The driver asked me to cancel the ride and pay him directly in cash. So I would pay less, and he would take home more. He told me Uber had been taking a much larger cut from drivers, sometimes, according to him, as much as 60%.

I would not take that number as the official average take rate. It was one driver, one ride, one conversation. But the feeling behind it did not seem random. There has been growing discussion about Uber’s take rate becoming more opaque. Some reports have suggested that, on certain rides, Uber’s cut can be above 50%.

The exact number is almost beside the point. What matters is that the relationship between Uber and its drivers feels more fragile than it used to.

My basic view is this: Uber is not a company I feel comfortable owning for the long term.

Fundamentally, I do not think Uber has fully figured out its long-term value proposition.

1. Uber is Not a Real Internet Company

Uber’s scaling power is limited.

A truly powerful internet company can usually build one product and distribute it across many markets at relatively low cost. Software is the easiest example. Once the code is written and the servers are running, users in many countries can start using it. Adding one more user does not add much cost, so as the company grows, margins can expand.

Uber is different.

On the surface, Uber looks like an internet platform. But it is not connecting a purely online world. It is connecting cars, drivers, riders, time, roads, airports, city rules, and local regulation.

That makes expansion much difficult.

Every new market requires Uber to solve a local version of the same problem. It needs to recruit drivers, attract riders, deal with regulators, handle insurance and safety responsibilities, adapt to airport and city rules, and compete with taxis, local ride-hailing apps, and public transportation. It also needs to understand how price-sensitive local consumers are, and what level of pay local drivers are willing to accept.

None of this can be solved by writing another line of code.

To be fair, Uber did benefit from the internet era. A big idea of that period was using software to connect fragmented supply and demand. Uber connected drivers and riders, and that created enormous growth.

But the things Uber connects are too physical.

That is why I never think of Uber as the ideal internet business model. It has platform characteristics. It has network effects. But it is not a truly asset-light internet company. Uber does gain advantages as it gets bigger, but it will always have to deal with the complexity of the physical world.

2. Uber is Entering the Harvesting Stage

Uber does not appear to be short on cash.

In the first quarter of 2026, Uber’s gross bookings were $53.7 billion. Its free cash flow was $2.3 billion. It also had $6.1 billion in cash and short-term investments.

So when drivers feel that Uber is under pressure to squeeze more out of each ride, that does not mean Uber actually has a cash flow problem.

My read is that after becoming profitable, Uber started shifting more of the economics toward the platform itself. Drivers take home less. Riders pay more. Uber keeps more. The marketplace feels worse on both sides, while Uber’s margins and cash flow look better.

A lot of this comes down to public market pressure.

Uber now has to prove that it is not just a growth story, but a mature business capable of producing consistent profit and free cash flow.

Investors need to see that mobility can be sustainably profitable, that delivery is not a permanent cash-burning machine, and that before autonomous vehicles arrive at real scale, Uber’s existing business can generate enough cash on its own.

So Uber will naturally, and maybe inevitably, push harder on the economics of each transaction. It can raise its take rate, reduce subsidies, limit how much value goes to drivers, and keep more of the economics on the platform.

In the short term, this makes financial sense. It can improve margins, increase free cash flow, and give shareholders a clearer path to returns.

But this feels more like financial optimization than long-term value creation. At its core, Uber is extracting more value from an existing network, not creating a new source of value.

That reflects the deeper tension inside Uber’s business model.

Uber wants to be a high-margin platform. But its supply side is not lifeless software. Its supply side is people. Drivers can feel their income falling. Riders can feel prices rising. When both sides of the experience get worse at the same time, trust between the platform and its users slowly erodes.

This is not a completely irrational choice. But it does make Uber feel less like a long-term compounder.

What I care most about is Uber’s awkward position in the transportation value chain.

Uber has always stood in the middle. It connects drivers and riders. It connects supply and demand. But Uber does not necessarily control the most valuable layer.

If the future still depends on human drivers, Uber is basically a marketplace. It has network effects, but those network effects are not especially strong or stable.

Drivers can work for multiple platforms at the same time. Riders can use multiple ride-hailing apps at the same time. Everyone is price-sensitive. City regulation is complicated. Supply quality can change by city, neighborhood, and time of day.

So Uber is not like Visa, Google Search, or Adobe. It does not have the same level of irreplaceability.

In the past, Uber connected drivers and riders, but the drivers were not Uber’s assets.

Now, Uber wants to connect autonomous vehicles and riders, but autonomous driving technology is not Uber’s asset either.

That is the core question for Uber: it has always occupied the middle layer, but does that middle layer have a durable moat?

3. The Conflict Between Autonomous Driving and Uber

My friend originally bought Uber because he thought autonomous vehicles could give the company another growth story.

That logic makes sense. Uber has users, demand, a dispatch system, and local operating experience. If robotaxis scale, Uber could become the demand layer that helps autonomous vehicle companies find riders.

But I think autonomous driving conflicts with Uber’s original model.

Uber’s core supply side is human drivers. The endpoint of autonomous driving is to reduce, or even replace, human drivers. If that happens, Uber has to redefine itself.

The bigger issue is that autonomous driving has deep technical barriers. It is not something Uber can easily control through acquisition or partnership. Companies like Waymo were built around this technology from the beginning, and they are already starting to scale commercially. By March 2026, Waymo had reached 500,000 paid robotaxi rides per week across 10 U.S. cities.

That puts Uber in an awkward position. It needs to protect its driver network today, while preparing for a future that may weaken that same network.

My view is that autonomous driving will not be completely fragmented, but it also will not be controlled by ride-hailing platforms like Uber. More likely, a few autonomous-native companies will control the core technology and vehicle operations, while Uber becomes a partner in some markets, mainly helping connect rider demand.

The Uber-Waymo partnership is a good example. In cities like Austin and Atlanta, Waymo’s autonomous ride-hailing service is offered through Uber. Uber handles fleet management and dispatch. Waymo provides the autonomous driving technology.

That shows Uber still has value. For Waymo, working with Uber can speed up adoption.

But Uber does not own the scarce assets: autonomous driving technology, safety record, mapping capability, regulatory approval, vehicle supply, and real-world driving data. Those are mostly in Waymo’s hands.

I think the robotaxi market will probably have three layers.

The first layer is the technology owner, such as Waymo, Tesla, or Zoox. This is where the deepest moat is.

The second layer is fleet operations: charging, cleaning, maintenance, inspections, software updates, accident handling, depots, and local operations teams. This layer is not yet a mature standalone industry, but it will likely become more specialized as robotaxi fleets grow.

The third layer is consumer distribution: Uber, Waymo’s own app, or Tesla’s own app.

Uber is strongest in the third layer. It may participate in parts of the second layer. But the strongest moat is still in the first layer.

That is why Uber may have a role in the autonomous driving era, but not necessarily the most valuable one.

4. Waymo Will Probably Do Both

Maybe the better question is whether Waymo will still need Uber once it has built enough consumer trust of its own.

The most likely answer is both.

In some cities, Waymo will build its own consumer habit. If it has a strong safety record, trusted brand, broad coverage, and short wait times, users can simply open the Waymo app directly. Over time, Waymo will not want to hand the customer relationship to Uber forever.

But in other cities, Waymo will still work with Uber. Entering a new market is difficult, and Uber can bring existing demand and local operating experience. For Waymo, Uber can speed up adoption.

So I do not think autonomous driving will directly kill Uber. But it may change Uber’s role.

Uber could move from being the dominant marketplace to being one of several distribution channels in the robotaxi market.

That difference matters.

If Uber is the dominant marketplace, it can keep more of the economics. If it is only one distribution channel, it has to share value with technology owners like Waymo, Tesla, and Zoox. The stronger those companies become, the weaker Uber’s bargaining power becomes.

Uber’s financial performance today is strong. It has traffic, local operating capability, dispatch experience, Uber One, and cross-selling between mobility and delivery. It has also proven that it can survive different cycles.

But I do not see it as a value investment.

Uber knows how to survive, pivot, and operate. But every time it enters a bigger future story, the most important technical asset does not fully belong to Uber.

Its past advantage came from the internet marketplace era. Its current profit comes from higher take rates and operating discipline. Its future story comes from autonomous driving.

But the most important capability in autonomous driving is not in Uber’s hands.

If the future has many autonomous driving companies, and none of them can control enough consumer demand on their own, Uber will still be valuable.

But if companies like Waymo control the technology and gradually build their own consumer entrance, Uber will be pushed into a weaker position.

I lean toward the second scenario.

Uber will probably survive. It may remain a strong operating company. But in my view, its long-term upside already has a visible ceiling.

The best answer is cost structure.

The most attractive technology companies have a premium business model. Once a software product is built, it can often serve many more users at a relatively low marginal cost. Revenue can scale faster than costs, which creates high gross margins, strong operating leverage, and capital-light growth.

This was one of the defining business ideas of the internet era: build the product once, distribute it widely, aggregate demand, and let scale improve the economics.

AI changes that equation.

AI looks like software from the user’s side. You open ChatGPT, type a question, and get an answer. The experience feels as simple as search, social media, or any other online service.

But every AI interaction requires computing power. Every answer, reasoning process, image generation, code generation, and agent task requires the model to run in the background. That creates inference cost.

As usage grows, the infrastructure burden grows with it. More users, more questions, and more complex agent workflows require more chips, data centers, electricity, cooling systems, networking equipment, and cloud capacity.

This makes AI structurally different from traditional software. A software product can often be copied and distributed at very low incremental cost. AI is different because each use draws on a large infrastructure stack behind it: chips, servers, data centers, electricity, cooling, networking, and cloud capacity. In that sense, AI is less like a finished software product and more like an intelligence service that has to be continuously operated. The interface is simple. The operating model is not.

The core question in the AI era is this—Can each unit of AI usage create enough value to justify the computing power behind it?

If AI remains a costly feature layered onto existing products, it could pressure margins. If AI enters enterprise workflows, completes high-value tasks, improves productivity, and gives customers a reason to pay more, then the infrastructure investment may make sense.

Now, the hyperscaler model is being tested. Microsoft, Google, Amazon, and Meta are spending enormous amounts of capital building data centers, buying GPUs, designing custom chips, and expanding cloud capacity. But whether that spending can turn into high-margin revenue and strong returns on invested capital is uncertain.

AI is bringing an old question back into technology:

How expensive is growth?

From Conversation to Execution

For AI, product capability is only the starting point. A strong model, reliable answers, and a usable interface are necessary, but the real battle is happening below the product layer, in data centers, chip supply, power contracts, cloud pricing, custom silicon, and the cost curve of inference.

Now, the AI race among large technology companies increasingly looks like an infrastructure war. Microsoft, Google, Amazon, and Meta are competing through models and product features, but the more durable advantage may come from data center capacity, GPU access, custom chips, cloud infrastructure, energy supply, and the ability to lower the cost of inference.

These things are less visible than product demos, but they may matter more for long-term winners.

An AI coding agent does not just answer one question. It may need to read code, write code, run tests, fix bugs, and revise again. An enterprise AI agent needs to connect with CRM systems, ERP systems, email, databases, and internal documents, then complete an entire workflow. A customer service agent has to understand the problem, call tools, check orders, evaluate permissions, and produce a solution.

These use cases create more complex, more continuous, and more expensive compute demand.

So one of the most important questions is whether AI can move from conversation to execution.

If AI mainly stays in chat, content generation, and information organization, usage may still be large. But the commercial value may not be enough to support today’s infrastructure buildout. The willingness to pay for casual conversation is limited.

But if AI moves into execution and starts completing real work, the economics change. It becomes an execution system that can directly improve productivity.

In that world, computing demand will be higher, but it may also be supported by higher commercial value.

This is why I think the keyword of the internet era was attention, while the keyword of the AI era may become execution.

Internet companies were built around occupying user time. Search, social media, short video, content platforms, and advertising systems all depend on attention. The more time users spend, the more data, ad inventory, and monetization opportunities the platform creates.

AI companies have a different job. AI is trying to complete tasks for them. It writes code, handles customer service, analyzes data, generates reports, operates software, and moves workflows forward. Its value comes from how much valuable work it can complete.

The attention economy asks: who can occupy more of the user’s time?

The execution economy asks: who can complete more tasks, at higher quality, and at lower cost?

This is the deeper challenge AI creates for the internet business model. The internet era was built around attention: where users searched, watched, shopped, scrolled, and spent time. AI shifts the focus toward execution. Now, we should look at which tasks they are willing to trust AI to complete.

CapEx Is Becoming the New Product Roadmap

In the past, when we looked at technology companies, we usually started with the product roadmap.

What features will the company launch this year? What markets will it enter next year? How fast is user growth? How strong is enterprise adoption? Is there room to raise pricing?

Those questions are no longer enough in the AI era.

Now, the product roadmap is increasingly tied to the capital expenditure.

If a company says it wants to serve more users, train more complex models, support more powerful agents, and enter more enterprise workflows, it also has to answer a more basic question:

Where does the computing power come from?

An AI company cannot rely on product vision alone. Its roadmap now has to be supported by a credible infrastructure plan: lower computing costs, enough data center capacity, stable cloud partnerships, possible custom chips, and improving utilization over time.

In other words, an AI company has to prove that it can carry intelligence profitably.

The AI race has turned infrastructure into strategy. Nvidia is not just a chip supplier; it sits at the foundation of accelerated computing. Amazon’s Trainium and Google’s TPU reflect the same pressure from a different angle—hyperscalers want more control over their cost curves. Microsoft and OpenAI show how model capability, distribution, and cloud capacity are becoming tightly connected. Oracle’s renewed relevance comes from the same shift—AI demand has made data center access and cloud capacity much more valuable.

Things that used to sound like the background of technology, such as power, cooling, data centers, and networking, are now moving to the center.

AI is turning the background into the foreground.

In the AI era, the market is paying much more attention to the infrastructure behind the app. Intelligence is not produced out of nowhere. It has to be trained, deployed, run, powered, cooled, and maintained.

The focus now is whose infrastructure can support the product over the long term, and whose capital structure can support the business model.

This is not an abstract point. It is already showing up directly in Big Tech’s capital expenditure.

Amazon, Microsoft, Meta, and Google are planning to spend up to $725 billion in capital expenditure in 2026, about $100 billion more than their previous 2026 plans. Meta is also a good example. In its first-quarter 2026 earnings, the company raised its full-year capital expenditure guidance to $125 billion to $145 billion, up from its previous range of $115 billion to $135 billion.

The AI race is becoming a balance sheet competition.

Adoption Is Not Enough. Utilization Matters.

For AI infrastructure, one of the most important questions is utilization.

Are these expensive chips being fully used? Is data center capacity consistently generating revenue? Are inference workloads stable? Are agent workflows frequent enough? Can the capital expenditure made today turn into future cash flow?

This is very different from traditional software.

For a SaaS company, the market usually looks at annual recurring revenue, retention, sales efficiency, and gross margin. For a consumer internet company, the market looks at daily active users, monthly active users, engagement, ad load, and revenue per user.

But when a company depends heavily on AI infrastructure, adoption is only part of the story. The market also has to look at utilization: whether the computing power is being used often enough, productively enough, and profitably enough. Idle chips create waste. Poorly located data centers become long-term burdens. Excessive capital expenditure drags down returns, while high inference costs can limit product usage and compress margins.

One of the biggest debates in the AI market today is whether Big Tech is building the infrastructure of the future, or overbuilding for a level of commercial demand that has not yet been proven.

When Meta raises its 2026 capital expenditure guidance to $125 billion to $145 billion, it is essentially making an early bet on future AI demand. If agents and AI workflows become frequent enough, these data centers and servers become the foundation of future growth. If usage does not materialize, they become depreciation pressure and return pressure.

But, after the infrastructure is built, is the computing power being used? Does that usage generate revenue? Does the revenue cover the cost? Does it eventually become cash flow?

From my point of view, if AI agents truly enter enterprise workflows and become part of the working system for employees, developers, analysts, and customer service teams, then today’s capital expenditure may be early infrastructure for the future.

But if AI usage mainly stays in chat, content generation, information organization, or occasional assistance, then today’s infrastructure investment may be too heavy.

Instead of asking whether you hold an optimistic or pessimistic view about AI, the better question is whether AI can create value that is large enough, frequent enough, and monetizable enough to support the infrastructure cost behind it.

AI Is Demystifying Technology

What I see is that AI is bringing technology back to the real world.

The internet era made technology feel unusually detached from physical constraints. Once a product worked, it could spread quickly through software and distribution rather than through factories, inventory, or local infrastructure. A small team could build something in one market and reach users around the world. In the best cases, the scale arrived much faster than the physical weight normally required to support it.

Compared with manufacturing, energy, telecom, and real estate, internet companies looked lighter, faster, and less constrained by physical assets.

But AI challenges this narrative.

The stronger the model becomes, and the more frequently it is used, the more visible the resource base becomes. AI depends on chips, electricity, data centers, cooling systems, and capital, all of which are limited, expensive, and difficult to scale quickly.

AI may look like language, intelligence, and abstract capability on the surface, but its business reality is much more concrete. It depends on chips, electricity, land, data centers, depreciation, supply chains, and capital expenditure.

That is the demystification of technology.

AI forces us to see the physical infrastructure behind software again.

This may be one of the most important ways to understand the next phase of technology. The language of the internet era lost its magic. Now we talk about cost structure, infrastructure bottlenecks, control over computing power, inference cost, utilization, recurring revenue, and returns on capital expenditure.

The future competition is not only about who can generate the best answer.

It is about who can generate those answers at the lowest cost, with the highest efficiency, in the most stable way, and embed them into real workflows.

AI is making technology more expensive again.

It is also making technology more concrete.

This is the new weight of the AI era.

(If you want to read a full analysis of Dynatrace, here you are!)

The company takes scattered operational data and turns it into one chain of evidence, so engineers and business teams can make decisions faster and with more confidence. It maps system structure and dependencies, aligns different kinds of data on the same timeline, and reduces alert noise so teams can get closer to the root cause instead of wasting time going in circles. In that sense, what Dynatrace really does is make a complex system more explainable, more actionable, and more governable.

But the more important question is still this:

Before going into abstract metrics, I think it helps to first look at one customer case. Otherwise, words like observability, workflow embedding, or platform value can sound too abstract. A real case makes the value easier to feel.

A Real Example: Next

Next is a retail and e-commerce company with 9.3 million online customers and operations across more than 70 countries. It also has a business called Total Platform, where third-party brands such as Victoria’s Secret, Gap, and Reiss run on Next’s infrastructure. That means it is not only responsible for its own digital experience, but also for the experience of its partners. If the online journey breaks, the impact can spread very widely.

Its pressure came from two places.

The first was growth. Online website and mobile sales were growing quickly and had already become more than half of the company’s revenue. So online experience was central to revenue.

The second was speed of change. Next had around 2,000 developers making up to 7,000 changes to digital services every week. The more changes a company makes, the more chances there are for something to break. During seasonal peaks, even a small drop in online experience can turn into a meaningful revenue risk.

In that context, what Next needed was better visibility into the customer journey. It needed to see what was actually happening in the system from the moment a customer opened a page to the moment they checked out.

After evaluating the market, the company chose Dynatrace. One key reason was its real user monitoring and session replay capabilities. Unlike some tools that rely on random sampling, Dynatrace can capture every user session. That matters because many customer issues are occasional. If you only sample, you may miss the one session that actually explains the problem. And if you cannot catch it, you cannot explain it. If you cannot explain it, you cannot solve it.

After adopting Dynatrace, Next used it in four main ways.

First, it reduced the risk that comes with rapid innovation. Because the company changed things so often, it needed to know quickly whether a release had made the experience worse or whether it had actually delivered the expected result. If a journey became slower or abnormal, teams could identify and adjust it faster. They could also use the data to show internal teams and third-party partners whether an update was actually good or bad.

Second, it helped solve customer issues faster. Session replay allowed service teams to see what users actually saw and what path they actually took. That meant teams did not need to guess, and customers did not need to repeat long explanations over and over again. Teams could identify the root cause faster and give customers a clearer explanation of what was going on.

Third, it made cross-team collaboration faster. The platform was not only used by technical teams, but also by customer service, marketing, finance, and even third-party partners. Everyone was looking at the same data and the same chain of evidence. So when something went wrong, different teams could align much faster on what happened, how large the impact was, who should own it, and what needed to happen next.

Fourth, it helped reduce fraud risk. Because a meaningful part of Next’s online business was tied to credit, the company used Dynatrace to monitor customer journeys in real time and flag suspicious behavior. One example in the case was that if the same user tried to log into multiple accounts in the same session, the system could flag it so the fraud team could step in quickly.

I think this case makes Dynatrace’s value much more concrete.

The value is that it helps a company see where revenue risk, customer friction, organizational confusion, and even fraud risk are actually happening in real time. It turns technical complexity into something people across the company can understand and act on.

Once that becomes clear, the more abstract numbers become easier to understand.

How That Value Shows Up in a More Quantifiable Way

What I think Dynatrace did smartly is that the thing it sells can look optional at first glance. Because of that, it spent time and money quantifying the result, so customers could more clearly see the business value it created. It worked with IDC on business value white papers, which try to make this value visible in a more measurable way.

The clearest value can first be understood through two categories:

First, less downtime and less loss.
Second, less human time wasted.

According to the study, the number of severe incidents, meaning major system failures and serious service disruptions, fell from 52.4 per year to 33 per year, a 37% decline. At the same time, the average repair time dropped from 2.9 hours to 1.3 hours, a 56% decline. As a result, annual revenue loss linked to these incidents fell from 4 million dollars to 1.1 million dollars, implying about 2.9 million dollars in annual loss avoidance.

Source: lDC Business Value In-Depth Interviews, March 2024

The second category is efficiency. The study says platform administrators improved efficiency by 53%. IT, DevOps, and application performance teams improved by 28%. Developer productivity rose by 18%. At the company level, the combined value of saved money and avoided losses averaged about 17.3 million dollars per year, and the annualized benefit per 100 supported applications was around 2.6 million dollars.

Source: lDC Business Value In-Depth Interviews, March 2024

From here, Dynatrace’s business value can be summarized in three lines.

First, fewer major incidents and shorter incident duration, so direct losses go down.

Second, less human time is wasted on finding causes and aligning information, so team efficiency goes up.

Third, when it becomes a shared language across teams, management can connect technical issues to business outcomes faster, and make trade-offs faster.

A more summarized way to see this is ROI. In the white paper, the reported ROI is 451%, with a payback period of about six months.

Source: lDC Business Value In-Depth Interviews, March 2024

A very simple way to feel that number is this: If a company spent 1 million dollars on the platform over three years, including subscription and implementation costs, then a 451% ROI would imply total benefits of around 5.5 million dollars. Roughly 1 million dollars would cover the cost, and around 4.5 million dollars would be net benefit.

*By the way, this is only a simplified example to make the ratio more intuitive. It is not real company data.

Then We Come Back to the Business Itself

If we want to understand Dynatrace as a company, then we have to go back to operating metrics.

For this kind of observability platform, growth does not come from selling a product once and being done. Usually, a customer starts small, then expands inside the same company, then adds more modules, and eventually turns the product into a shared platform used across many teams. If that is the path, then the company’s most important questions also appear in sequence. And the metrics we watch should follow that same sequence.

I think the framework can be understood in three steps:

Step one, land.
Can the product actually get adopted in a real environment and solve an immediate problem.

Step two, expand.
Can the company deepen usage inside existing customers in a way that is economically attractive.

Step three, become a platform.
Can the product move from being a tool to becoming shared infrastructure inside the organization.

Subscribe now

Step One: Land

At this stage, the main risk is whether it can sell it and get the customer to actually use it.

Customers usually begin with one system, one team, or one business line. So the key question is whether the product can solve a real problem quickly enough that the customer wants to renew and expand spending.

That is why the first thing to watch is growth and retention, especially Annual Recurring Revenue (ARR) and Dollar-Based Net Retention Rate (DBNRR). These tell us whether customers are moving from pilot use into continued expansion.

Source: Dynatrace Q3 26’ Investor Presentation

Dynatrace disclosed about 1.97 billion dollars of ARR in fiscal 2026 Q3. That matters for two reasons. First, it shows the company is already operating at meaningful enterprise scale. Second, the fact that ARR is still growing suggests it is not only defending its installed base, but also still adding new business and expansion.

Its NRR was 111%. That suggests existing customers are still deepening usage and paying more overall, so expansion is still covering contraction and churn. At the same time, 111% does not look extreme. It looks more like a healthy number for a mature enterprise software business, showing that the expansion engine is still working, but it is not running at an unusually aggressive speed.

Source: Dynatrace Q3 26’ Investor Presentation

This number only becomes meaningful through comparison. Compared with its own history, the broader trend has been downward, which suggests Dynatrace has already moved past its highest-growth phase and is now in a more stable stage. Compared with peers, 111% looks healthy, but not exceptional. So the number is good, but the trend still matters.

Step Two: Expand

At this stage, the main question is how deeply it can expand within existing ones.

Once enterprise customers believe in the value, expansion is often cheaper and more stable than acquiring new customers. But this stage also exposes whether scale is really improving the business or whether cost is starting to rise with usage. Observability goes deeper as customers generate more data, and that can create structural pressure on storage and processing costs.

So here, the main things to watch are unit economics and cost structure. Is subscription gross margin stable? Is free cash flow improving? Is the company having to spend more and more on sales just to get the same kind of growth? If those metrics weaken, expansion may be happening through discounting, more people, or higher delivery costs, which would be harder to sustain.

Dynatrace reported subscription gross margin of 85% and services gross margin of 3% in fiscal 2026 Q3. An 85% subscription gross margin is strong. It suggests that customers are willing to pay, and that Dynatrace is not being dragged down too heavily by cloud or data processing costs. The 3% services gross margin looks low, but in enterprise software that is not necessarily a bad sign, because services are often used to help customers adopt the product rather than to generate profit themselves. The real concern would be if services became a larger share of revenue while still remaining low-margin, because that could suggest the business depends too much on human delivery.

Source: Dynatrace Q3 26’ Investor Presentation

The company also reported quarterly free cash flow of 27.234 million dollars and free cash flow margin of 5%, while guiding to a full-year free cash flow margin of 26%. A single quarter can move around, so I would not read too much into that one number alone. But if full-year free cash flow margin can really stay around 26%, that suggests the company is still able to turn revenue into meaningful cash generation.

Source: Dynatrace Q3 26’ News Release

More broadly, the business still looks like a subscription-led enterprise software model. Subscription revenue remains the main revenue source, while services are a relatively small piece. That matters because it means the company still looks like a software business with scale benefits, not a services-heavy business where growth is tied too closely to labor.

Step Three: Become a Platform

This is the stage where the real question becomes whether Dynatrace is becoming shared infrastructure.

At this point, the most important thing is not customer count by itself, but how deeply and broadly the product is used inside each customer. Is it spreading across teams. Is it spreading across modules. Is it becoming part of real daily workflows.

I think there are four signals to watch.

1. Multi-module penetration

Are customers buying more than one product? Is management increasingly talking about broader platform adoption rather than single-product adoption? Dynatrace’s own filings emphasize expanding relationships inside existing customers, broadening platform capabilities, and using its DPS licensing model to drive wider platform consumption?

2. Large customers going deeper

Platformization often shows up first in large enterprises, because they have more complexity, more teams, and more to gain from using one shared system. So what matters is whether more of them are expanding from narrow use into broader organizational adoption.

3. Usage depth

Ideally, we would have direct numbers like monitored services, containers, active users, queries, or automation actions. If the company does not disclose those, then NRR and large-customer expansion stories can work as rough substitutes, because deeper usage usually shows up through existing customers spending more.

4. Workflow embedding

This is where a product starts feeling like infrastructure. Does it sit inside ticketing systems? Does it connect to on-call systems? Are customers using it to guide service-level management, alert reduction, and automatic root cause analysis? Companies do not always disclose one clean number here, so often this has to be judged through recurring patterns in customer stories and management language.

Share

What Would Real Competitive Pressure Look Like

At the platform stage, competition does not always show up first as obvious churn. More often, it shows up as slower decisions, larger discounts, shorter contracts, and weaker visibility.

So I think there are four practical things to watch.

1. Pricing and discount pressure

If the company has to cut price or promise more costly delivery in order to win deals, subscription gross margin should start to weaken. So far, Dynatrace’s margin still looks relatively stable. That suggests it is not being badly squeezed yet, even if it is clearly no longer in a high-growth phase.

2. Contract structure

Dynatrace disclosed 3.2032 billion dollars of RPO as of December 31, 2025, and said 53% of that would be recognized over the next twelve months. That gives a rough near-term revenue visibility figure of about 1.6977 billion dollars. On its own, that does not tell us whether the trend is improving or weakening. But it does give a clearer framework for thinking about short-term visibility and contract depth.

Source: Dynatrace Q3 26’ 10Q

3. Sales cycle length

If enterprise decision-making gets slower, approvals become stricter, and customers become more cautious, that usually shows up first in weaker order visibility and slower new ARR formation.

4. Replacement and bundling pressure

At this stage, Dynatrace is not only competing with other observability vendors. It is also competing with cloud-native tools and larger vendors that may bundle logging, security, and cost management together. So one of the most important things to watch is whether Dynatrace can keep convincing customers that a unified data model, unified governance, and unified workflow really matter enough to justify a standalone platform.

Final Thought

At this point, I think the more useful takeaway is not just how to understand Dynatrace, but how to analyze tech companies more generally. The key is to start with how a company actually creates value for its customers in the real world, then ask what metrics can help quantify that value. From there, you can test the business both horizontally, by comparing it with peers, and vertically, by looking at whether that value creation holds up over time.

That is the framework I am trying to build for myself when I study companies. First identify the real source of customer value. Then connect that value to the operating and financial data that can support it. Only after that does the model start to feel meaningful, because the numbers are no longer floating on their own. They are tied back to the actual business reality.

Most infrastructure categories sound boring until the world gets complicated enough that they suddenly become essential.

I think observability is one of those categories.

For a long time, software systems were much easier to understand. A company might have had one main application, or just a small number of services. Deployments were relatively stable. Changes did not happen all the time. If a page became slow, engineers could usually trace the issue back to a clear source. If errors appeared, they could often identify the failing module without needing to reconstruct the entire environment.

That was the world in which application performance monitoring, or APM, made sense.

APM was built for a simpler technical reality. It helped engineers see how an application was performing, where a request spent its time, and where errors appeared. And in systems with clearer boundaries and fewer moving parts, that was often enough. A team did not need a complete model of the entire environment. It just needed enough visibility into one application to diagnose the problem.

But that is no longer the world most companies operate in.

Modern software systems are much harder to understand than they used to be, and the deeper reason is that companies now need software to change faster, handle more traffic, and stay online even while different parts of the system are constantly being updated.

That pressure changed the way software gets built.

Instead of putting everything inside one large application, companies now often break a system into many smaller services (micro-services) because smaller pieces are easier for different teams to build, update, and scale on their own. If one part of the product changes often, or needs much more computing power than another part, it makes sense to separate them rather than force everything to move together.

To make those smaller services easier to deploy and run, companies often package them into containers. A container is basically a standardized bundle that includes the application and the environment it needs in order to run. The point is to make software easier to move across different machines and environments without constantly worrying that it will behave differently in each one.

But once a company has many containers, it needs a system to coordinate them. That is where Kubernetes comes in. Kubernetes is software that helps companies manage large numbers of containers. It decides where they run, restarts them if they fail, and adds or removes capacity as demand changes. This means systems now scale up and down continuously. If traffic rises, the company may need more running copies of a service so the system does not slow down or crash. If traffic falls, it may reduce those copies to save cost. In other words, the shape of the system is no longer fixed. It keeps changing while the business is running.

Many companies also no longer operate in one clean environment. They use multiple cloud providers, or a mix of cloud and on-premise infrastructure, because different workloads, regulations, legacy systems, and cost considerations push them in different directions. On top of that, their systems depend on a growing number of internal services and third-party tools, from payments to identity verification to message queues to analytics.

Release cycles are faster. Teams may push several code releases or configuration changes in a single day because the business expects software to improve continuously, not through large slow updates.

Therefore, software has become more modular, more distributed, and more changeable because companies want speed, flexibility, resilience, and efficiency. Once a system is built this way, it also becomes much harder for any human team to understand what is happening across the whole environment at a given moment. This is the point where traditional APM starts to feel insufficient.

You may be able to see that one service is slow, but not whether that service is actually the root cause. You may see error rates climbing, but not know whether the source was a recent deployment, a database lock conflict, or an external API becoming unstable. You may have traces, but they may be incomplete. You may have metrics, logs, and alerts, but they may live across different tools and different teams. What you end up is fragments, but we all want a coherent understanding of the system to answer the questions that actually matter during an incident.

What changed? What depends on what? Which issue is the real root cause? Which failures are only downstream symptoms? How large is the blast radius? Which users or transactions are affected? What should the team do next?

That is why observability has become so important.

A good observability platform has to do more than show that something is wrong. It has to help teams understand what the system actually is, what is changing inside it, and what matters most when something breaks.

The first layer is structure.

Before a team can diagnose a system well, it needs a credible picture of the system itself. That means knowing what services exist, what databases, queues, and APIs they depend on, and how those relationships change over time. A strong observability platform tries to build that picture automatically and keep it updated as the environment evolves. This may sound basic, but it solves one of the most wasteful parts of incident response. In many companies, teams spend far too much time simply figuring out what the current system looks like before they can even start solving the actual problem.

Once that structure is visible, the next layer is evidence.

During an incident, engineers need one usable chain of evidence. If latency rises, they want to know which request path is slowing down and which dependency is involved. If they begin with an error log, they want to know what was happening to load, response time, and adjacent services at that same moment. That is why a real observability platform has to connect metrics, traces, and logs into one coherent context. The point is not to show more signals. The point is to let teams move from symptom to explanation as quickly as possible.

The third layer is prioritization.

In a complex system, many things can look broken at once, but not all of them matter equally. Some anomalies are root causes, while others are only downstream reactions. So we want to know is what should be fixed first. A strong platform needs to reduce noise, rank likely root causes, estimate blast radius, and connect anomalies to real changes in the environment. Otherwise, it may give teams more visibility while still leaving them overwhelmed.

The final layer is workflow.

In large organizations, incidents involve routing alerts, escalating ownership, validating fixes, updating runbooks, connecting to ticketing systems, and learning from past incident reviews. The value of an observability platform is whether it fits into how a company actually operates. Once a platform becomes part of that operating rhythm, it becomes much harder to replace.

In short, observability focuses on making an entire distributed system legible by combining different forms of telemetry, keeps track of changing dependencies, and helps teams move from anomaly to evidence with less guessing, less tool-switching, and less cross-team friction.

Why Dynatrace

Plenty of companies are trying to build that kind of platform. I think Dynatrace is especially interesting for the specific way it approaches the problem.

At its core, Dynatrace is trying to build a continuously updated map of reality.

That matters because Dynatrace realizes that the central bottleneck is no longer what autonomous systems could do in theory. It is whether enterprises can trust them to act in production. The biggest blockers are all forms of lost control, from security and compliance concerns to limited real-time visibility and unclear approval boundaries. Companies need autonomy that is controllable. That is why Dynatrace is trying to build a safe and reliable autonomous systems, and the focus should firstly be on “safe and reliable”. Without that, autonomy may be technically possible, but still too risky for enterprises to adopt at scale.

This is the first reason I think Dynatrace has a real moat.

The second part of that moat is switching cost. Once a company calculates service-level objectives on one platform, encodes alert rules and noise-reduction logic there, ties runbooks and ticketing systems into it, and trains teams to work inside that environment, the platform becomes part of the company’s operating system. Migrating away from it means re-encoding a large body of operating knowledge and revalidating it so the new setup does not create blind spots or alert fatigue.

The third part of the moat is the semantic and causal model.

Many vendors can collect signals. Fewer can reliably turn messy signals into answers inside messy real-world environments. Enterprise systems have inconsistent naming, changing topology, rolling deployments, incomplete traces, imperfect timestamps, and many layers of dependencies. The hard problem is building a model strong enough to explain what happened in a way people trust.

This seems to easy to be replicated by other competitors but it actually requires long-term refinement across data collection, correlation, diagnosis, presentation, and workflow integration, which requires true understanding of the product architecture underneath it.

The fourth part of the moat is enterprise delivery.

Very large enterprises are difficult customers. Security teams care about data residency, access control, and auditability. Platform teams care about performance overhead and maintainability. Application teams care about speed and usability. Management cares about reliability, risk reduction, and return on investment. Even if a product works well in a proof of concept, expanding it across multiple teams and business units is still hard.

Execution matters. Specifically, the vendor has to know how to onboard customers in the right order, define useful service-level objectives, control alert noise, integrate into existing processes, and introduce automation carefully. The more successful deployments a company has completed, the easier it becomes to win the next one. Over time, that creates an execution advantage that is hard for newer entrants to replicate.

Different Ways to Build Observability

Of course, Dynatrace is not alone.

Datadog and Splunk are both serious competitors, but they are strongest in different ways.

Datadog is more oriented around integration breadth, speed of onboarding, and developer flexibility. It is very good at helping teams bring many services and tools into one place quickly, then query and slice the data through tags and dimensions. That makes it especially appealing to teams that want fast adoption and a flexible workflow. In lighter or medium-complexity environments, or in organizations that want more freedom in how they build their monitoring setup, that approach can be very powerful.

Splunk comes from a more log-centric and security-adjacent tradition. Its historical strength is in search, investigation, and evidence-based workflows. It is especially strong where logs, auditability, and security operations matter most.

It is naive to say that Dynatrace can do everything and competitors cannot. In fact, these companies optimize for different centers of gravity.

Dynatrace seems to be optimizing for the customer who lives in the most complicated enterprise environment, where too many systems, too many teams, and too much change make manual interpretation unscalable. Datadog is relatively more optimized for breadth, speed, and flexibility. Splunk is relatively more optimized for logs, investigation, and security-oriented workflows.

I do not think this difference is accidental.

There are real trade-offs in trying to win all of these dimensions at once. A company that wants to dominate automated system modeling and root-cause diagnosis has to invest deeply in structured models and default automation. A company that wants to dominate developer workflow and integration breadth has to invest heavily in connectors, templates, APIs, and usability. A company that wants to dominate log search has to invest heavily in indexing, storage, retrieval, compliance, and security operations. These are all expensive, and they do not always fit naturally into the same product philosophy.

That is why I think the observability market does not necessarily converge into one perfect product that is best at everything. It produces a few companies that each become exceptionally strong along one path, then extend outward from there.

After having the big picture in mind, I still think Dynatrace is easy to underestimate.

If you only look at the category at a surface level, observability can start to look like a feature race. More dashboards, more AI assistants, more traces, more alerts, more integrations. But at the deepest level, the real product is faster and more reliable organizational understanding inside a system that has become too complex to reason about manually.

And if that is the product, then Dynatrace’s emphasis on automatic discovery, causal explanation, and enterprise complexity starts to look much more important.

Where the Thesis Could Break

That said, the risks are real too.

The first risk is standardization, especially through OpenTelemetry. As telemetry collection becomes more standardized, customers will find it easier to move data across platforms and compare vendors. That weakens lock-in at the collection layer.

But that does not necessarily destroy value. Collection becoming commoditized is not the same thing as diagnosis, workflow integration, and governance becoming commoditized. If anything, it may make the higher layers of the product even more important. When everyone can collect more or less the same raw data, the real question becomes who can turn it into the most trustworthy and useful operational answer.

The second risk is stronger native tooling from the cloud vendors themselves. In simpler single-cloud environments, native monitoring products may become good enough for many customers, especially on price. But large enterprises usually operate in hybrid environments with multiple clouds, on-premise infrastructure, legacy systems, and many external dependencies. That still creates room for a neutral platform that can sit above the whole environment.

The third risk is AI feature convergence.

Every vendor will talk about AI copilots, AI diagnosis, and automated root-cause analysis. Over time, many products may start to sound similar. That can create pricing pressure if the market begins to treat AI as a checkbox feature.

My own view is that AI is not the real product. The real product is whether the platform actually helps teams solve problems faster in a trustworthy and controllable way. If the workflow is weak, the governance is weak, and the evidence chain is weak, then AI is just a nicer interface layered on top of an unreliable process.

There is also a broader industry signal worth watching, which is consolidation. The fact that recognizable companies in this space have been acquired by larger firms suggests that observability is being treated less like a niche tool and more like an important part of enterprise infrastructure. That cuts both ways for Dynatrace. On one hand, it reinforces the idea that strong platform assets in this category are valuable. On the other hand, it means competition may increasingly come not only from pure-play observability companies, but from larger ecosystems bundling observability into a broader enterprise stack.

The Bet on Complexity

In short, I think the Dynatrace case comes down to a fairly simple judgment.

The company matters because modern systems have become too dynamic and too interconnected for manual debugging to scale. Observability matters because enterprises need more than visibility. They need a way to convert complexity into operational understanding. And Dynatrace matters because it is trying to solve that problem in a way that is especially well suited to large, messy, high-stakes enterprise environments.

That does not make it risk-free. But if the market continues moving toward larger, more complex, more hybrid, and more AI-heavy systems, then I think Dynatrace owns the highest growth potential.

We are betting that complexity itself is becoming one of the most important problems in enterprise software, and that the winners in observability will be the platforms that can make that complexity legible, actionable, and governable.

That is a serious thesis, and I think it is one worth paying attention to.

Tech companies created value by connecting people to information, people to people, people to products, and people to services. The internet reduced the cost of distance. It made search easier, communication faster, and transactions more convenient. In that world, the most powerful companies were usually the ones that could build the biggest networks, control the main gateways, and become the default place where users started.

Google wanted to be where you searched. Meta wanted to be where you spent attention. Apple wanted to be the operating environment you lived inside. Amazon wanted to be where you bought. Each company was trying to become a default entry point. Once it controlled that entry point, it could keep expanding outward.

But I think the AI era may change the game

If the internet era was mainly about connection, the AI era may be more about time compression. The question is no longer only who can connect users to the right information. It is who can shorten the whole path from intention to result, who can understand what the user wants, process the information, automate the workflow, and save time.

That shift matters because saving time is different from simply providing access. In the internet era, a company could win by helping you find something. In the AI era, the stronger company may be the one that can help you finish something. The internet helped people reach. AI may help people complete. That means every tech company now has to rethink its role. The question is where in the user’s workflow the company can remove steps, reduce waiting, and turn intention into action.

Seen from that angle, different tech companies may have very different paths.

Some companies are naturally strong at interpretation. They can answer questions, organize information, summarize complexity, and guide decisions. Their advantage is that they sit close to human intent. They understand what the user is asking for.

Other companies are naturally strong at execution. They may not always be the best at interpretation, but they control the system that completes the task. They have logistics, payments, infrastructure, devices, software environments, or enterprise workflows that let them actually carry things through.

Share

This is where Amazon could have a unique position

Amazon was already powerful in the internet era because it built a dense system of connection. It connected consumers, merchants, products, logistics, fulfillment, cloud infrastructure, and later advertising and devices. Its strength was that these pieces kept reinforcing one another. That is why the old language around Amazon, the flywheel, customer obsession, and turning internal capability into external interface, still matters. It captured the logic of how Amazon became powerful in the first place.

But in the AI era, if competition shifts toward time compression, then Amazon needs to sit at the layer of recommendation or information, and also control a large part of the execution layer.

If you want to buy something, Amazon should help you search, store the item, process the payment, move the product, deliver it, and handle the return. If a business wants to build in the AI era, Amazon should offer access to models, cloud infrastructure, compute, tools, and enterprise systems through AWS. In both cases, Amazon helps you think and also helps you complete the whole action chain.

Many companies have the brain or the hand. Some can interpret intent but cannot execute. Some can execute but do not control the intelligence layer. But Amazon is one of the few that has a real chance to combine both.

This explains why products like Alexa matter. I think people sometimes focus too much on whether it is a good chatbot. We should focus on whether Amazon can use AI to shorten the distance between what the user wants and what actually gets done. Search, compare, decide, order, deliver. The more of that chain Amazon can compress, the stronger its position becomes.

And that, to me, is the broader lesson. Amazon is not just a company story. It is a way to see how tech power may be changing. In the internet era, power often came from building the biggest network and controlling the default gateway. In the AI era, power may increasingly come from building the shortest path from intention to completed action.

Connection still matters, because it remains the starting point for users. The next winners may still need networks, data, scale, and distribution. But those things will matter most when they help a company save users time in a deeper way, by helping them get it done.

The real strategic question for tech companies is becoming much sharper. What part of the customer’s time can you eliminate? What part of the workflow can you compress? What part of the gap between wanting and getting can you own?

In the internet era, the strongest companies reduced distance. In the AI era, the strongest companies may be the ones that reduce time.

In the concluding part of the last article, I posed a few questions:

• How does Generative AI allow us to surpass human limits again, and what clues does it provide us?

• What might be the future directions and major trends in the technology industry?

• What significance does ChatGPT’s newly introduced voice chat feature hold?

• And what deeper meaning lies behind Meta’s recent launches of glasses and graphic generation?”

Let’s tackle these questions one by one.

Why ChatGPT’s New Features Are So Important?

The functionalities of ChatGPT that we currently enjoy without spending a dime involve typing and engaging in text-based conversations. However, around the end of September, ChatGPT gained the ability to speak and perceive visual content. In other words, you can now have a lifelike voice conversation with AI, a capability I’ve personally tested and found remarkably realistic, akin to talking to a stranger on the phone.

Additionally, you can send a photo to ChatGPT for it to analyze. These two new features are game-changers! Ben Thompson also highlighted the significance of these features in his article “AI, Hardware, and Virtual Reality,” emphasizing how they make human-machine interaction more natural and significantly reduce the sense of distance when using ChatGPT. Yet, I believe there’s another way to analyze this.

Let’s consider the evolution of trends in social media software from ancient times to the present.

• Facebook emerged in 2004;

• Twitter followed in 2006;

• Instagram joined the scene in 2010;

• Snapchat in 2011, and…

• currently, the hottest platform is TikTok, introduced in 2014.

I cite examples from U.S. social media despite the possibility that these platforms might not be extensively used in China, simply because the tech industry there has developed more naturally from start to finish with fewer regulations, making trends more apparent. Can you spot the trend in the progression of these social media software over time?

Facebook became the social media giant of its era by ensuring real user identities and providing initial information exchange and community building. Twitter gained popularity because it required all user messages to be under 150 characters, significantly increasing information density and dissemination efficiency.

During that time, due to technological and network speed limitations, communication and information dissemination were text-dominated. With the advent of 3G networks, Instagram became possible. On Instagram, we can share photos, short videos, GIFs, beauty filters, and various effects. Snapchat revolutionized communication by using photos instead of text, and the messages sent would disappear after a certain period.

Platforms like TikTok shifted to entirely video-based content. J and Chris Paik believe that people always want more information, want it in the shortest time possible, and technology determines the ceiling of information density. In summary, they observe a trend towards information density. However, I hold a different perspective.

Trends in the Development of Social Media

Firstly, I don’t believe people always crave more information. While more information is good, an excess of it can overload our brains. Nor do I think people want to get information in the shortest time possible — at least not at an always-high information density. I believe the most crucial aspect of the overall trend is that social media is making our online communication more and more like real-life interaction.

What does our communication look like in real life? We constantly use our eyes to see (all the photos around us); the scenery before us is dynamic (videos); we speak instead of typing (voice); and our spoken words aren’t recorded or screenshotted — they dissipate in the air like the wind (a significant feature of Snapchat, and Instagram’s story function has a similar aspect).

Our diverse interpretations of this trend reflect our different perspectives on a particular software — BeReal. This recently trending app sends users a random notification every day, prompting them to share a photo of themselves and their surroundings within two minutes, with no filters or beauty enhancements — just raw authenticity! Isn’t authenticity the most prominent characteristic of social interaction between people? BeReal utilizes time constraints and random reminders to make authenticity, which is often challenging to achieve in the online world, a reality. It’s another software that makes the online world more like reality.

Of course, as a standalone product, BeReal might need more hooks to retain users, but as an additional feature on existing social platforms (such as Instagram acquiring and integrating BeReal), it could become a more valuable use case.

I spent so much time discussing trends in social media software to emphasize that the ability to communicate with ChatGPT through voice and photos is a significant breakthrough. It makes interacting with computers more natural and seamless.

However, ChatGPT and the trend we’re discussing have a fundamental difference: ChatGPT is not a social media platform. Social media helps convey information, but ChatGPT creates information. While social media makes the online world more like reality, what does that sound like? It sounds like Virtual Reality. And ChatGPT has a grander mission — to merge the real world with the electronic world. What’s that? That’s Augmented Reality. Of course, this mission doesn’t necessarily have to be fulfilled by ChatGPT; any AI model could be a candidate for this mission.

Now, let’s turn our attention to Meta’s latest release — the Meta Smart Glasses.

Smart Glasses: Virtual Reality and Augmented Reality

Let’s first consider, from a sensory perspective, what virtual reality and augmented reality entail.

Virtual reality, in a sense, is creating a parallel universe (let’s call it World B) while living in our current World A. Why do we create this alternate reality? Surely, it’s because we are dissatisfied with certain aspects of A and want to explore more in a new world. However, completely abandoning A is not feasible because some aspects of A cannot be entirely replaced by B. B can come very close to A, but it can never fully substitute for it. Therefore, in the future of virtual reality, humans are likely to switch between these two worlds.

Now, what is augmented reality? Augmented reality is a technology that overlays digital information, such as images, videos, 3D models, or data, onto the real world. It enhances users’ perception of the physical environment through devices like smartphones, tablets, smart glasses, or specialized AR headsets. Unlike creating a parallel universe, augmented reality transforms and upgrades our current world. This means we don’t have to give up the beauty of any existing world, and there’s no need to constantly switch between different worlds. In a way, virtual reality creates two planes, while augmented reality adds dimensions to the existing plane. From this, two dimensions become three-dimensional and even multi-dimensional.

When we think about the verb “overlay,” how can we overlay another world onto our existing world? The current approach focuses on the human senses. Since humans perceive the world through various senses — ears, eyes, nose, skin — changing our sensory channels directly can transport us to another world. So, what kind of sensory device should we choose? Unanimously, everyone has chosen glasses. For me, this is a very natural choice because, compared to other organisms, humans are highly visual creatures. Most of our daily sensory experiences are dominated by vision, which also captures the majority of our attention. Therefore, glasses and head-mounted devices have become the most natural choice.

Zuckerberg, when introducing our glasses, said:

“I think that smart glasses are going to be an important platform for the future, not only because they’re the natural way to put holograms in the world, so we can put digital objects in our physical space, but also — if you think about it, smart glasses are the ideal form factor for you to let an AI assistant see what you’re seeing and hear what you’re hearing.”

Ben Thompson agrees with this statement.

I have no objection.

But what I find most exciting in this statement is the word — Platform. Yes, as everyone starts focusing on augmented reality, glasses, and head-mounted devices, the attention in our tech world has begun to shift from software to hardware. When new hardware emerges, just like the birth of the iPhone once did, a new platform comes into existence. What is a platform? From the most basic definition, a platform is a cross-section where various activities and things happen.

When smartphones appeared, all software development, innovation, and revolutions began to unfold. Everyone was thinking about creating their products on this platform. And now, a new platform is born, and everyone will start thinking, about how to design products on this new platform. How to design a product that is more naturally usable for the glasses platform? How to design a product that fully leverages the advantages of generative AI? How to bring value to users? How to ride the next wave?

This new wave is incredibly exciting, but at the same time, we must be aware, going back to one of the characteristics of TMT at the very beginning, the enormous upfront investment costs. This implies that the rise of this wave will bring rapid capital shifts and investments, but can such investments really bring the expected returns? Hardware development clearly involves more upfront investment costs compared to software, meaning higher risks, and slightly lower revenue, requiring more rational choices, and more efficient capital operations.

In fact, thinking about the current market in the United States, the overall pessimistic market has provided investors with a sober and calm environment.

In this environment, I am full of enthusiasm and anticipation.

Now, tackling such topics usually involves a predefined analytical framework, and my go-to approach is the dichotomy method. It’s a structured way to delve into the industry, starting with the External Market before diving into the nitty-gritty of the Internal Business. This method provides a clear and comprehensive framework that covers all aspects, offering a holistic view from past to present. For instance; it’s akin to categorizing people as biological men and women — a straightforward approach that ensures inclusivity.

Let’s kick off with the coffee industry, a quintessential player in the Consumer & Retail sector. The sheer scale of this industry is staggering; after all, who doesn’t enjoy a good cup of coffee? In the United States, coffee has transcended mere beverage status — it’s a life necessity. The Total Addressable Market for coffee, consequently, is vast.

But, and it’s significant, this expansive market comes with intense competition. Coffee shops are a dime a dozen worldwide, and the industry landscape is in a constant state of flux, with new players entering the scene regularly. The challenge lies in attempting to fully saturate such a dynamic market. While the Total Addressable Market is colossal, the room for growth is somewhat limited. Almost everyone is already sipping on their daily brew, and there’s a cap on how much coffee an individual can consume in a day. With a finite increase in daily coffee consumption, the industry finds itself under a metaphorical ceiling, leaving little space for substantial improvement.

Furthermore, delving into the annals of coffee, it becomes evident that the industry has long since stepped into a mature stage, where the scope for early growth is considerably limited. Yet, a noteworthy aspect of consumer-facing markets lies in the fact that their products aren’t mere commodities but rather thrive on differentiation.

Let’s shift our focus to the Internal Business aspect. Internally, operations can be neatly bifurcated into quantitative metrics and qualitative analysis. The former involves the somewhat mundane task of scrutinizing financial details like the income statement (encompassing revenue, costs, expenses, losses, and profits), the cash flow statement, and the balance sheet (considering factors like working capital efficiency, capital intensity, capital utilization allocation, capital structure, and the driving forces behind these metrics).

Now, let’s spice things up with the more captivating qualitative analysis, which deserves a prime spot in our discussion. To grasp the prevailing industry consensus, I delved into The Business Research Company’s comprehensive industry study, highlighting various business strategies employed by coffee markets.

First and foremost, coffee shops set themselves apart by offering superior-quality coffee, elevating the experience beyond the ordinary. This might entail sourcing premium coffee beans, adopting organic and ethically-driven standards, or featuring unique and niche varieties such as pour-over coffee.

Secondly, there’s a delightful array of coffee shops putting in extra effort to tantalize taste buds. Think walnut coffee, rosemary and Bergamot latte, honey cinnamon latte, lavender latte, chocolate coffee, and spice coffee, just to name a few.

And let’s not forget the coffee establishments that go the extra mile with their service. With technology becoming ubiquitous, “coffee subscription” services are gaining popularity. Subscribers receive a regular delivery of high-quality coffee beans, ensuring freshness and convenience, often accompanied by samples of new products. This not only enhances the customer experience but also allows coffee shops to control costs effectively and forecast profits for the upcoming quarter, thereby providing a more streamlined and efficient service.

Diversifying the target audience is yet another strategic avenue. Take, for instance, coffee shops that tailor their offerings to teenagers aged 13 to 18. This demographic is enticed through youth-centric promotions, flavor adjustments to cater to richer and sweeter tastes, and an increased emphasis on online ordering services.

These approaches essentially align with shifts in consumer behavior. With the growing habit of coffee consumption (although the trend has plateaued), an increasing focus on health and quality, and coffee morphing into a lifestyle choice, coffee shops, and indeed all players in the Fast-Moving Consumer Goods (FMCG) industry, need to adapt their product and sales strategies accordingly.

While these strategies may sound like industry basics, they can feel somewhat cliché and fail to stir much excitement upon reading. Yes, this framework facilitates a structured presentation of my surface-level understanding of the industry and a typical company. But how do I inject more depth, provide insightful perspectives, and spark engaging discussions?

It was at this point that a thought struck me — seek out an unconventional company!

Enter the coffee industry, and what name pops into your head? None other than the magical Starbucks.

Starbucks stands out as an extraordinary coffee company. When you think of Starbucks, it’s not just about their products; there’s something more, something intangible…

Let’s distill these intangibles into three keywords: third space, unique language, and corporate culture.

Let’s delve into the enchanting concept of the “Third Place.” Once a relatively novel idea, the term has now seamlessly integrated into our daily lexicon, a testament to its deep-rooted resonance with consumers. The Third Place, as conceptualized by American urban sociologist Ray Oldenburg in his book “The Great Good Place,” represents an informal public space essential for a thriving city life. It’s a space where people can gather, shed their concerns momentarily, engage in diverse conversations, and spend time with those they cherish. Oldenburg contends that the existence of such places is integral to the establishment of a free, equal, and open culture in modern society.

In this Third Place, our visits aren’t driven by specific needs like quenching thirst or satisfying hunger. While we may consume food and drink, or rest, these actions aren’t the primary purpose — we go simply because we want to. Starbucks seamlessly embodies the essence of the Third Place in the ongoing transformation of our social fabric. Howard Schultz, the visionary behind Starbucks, once articulated, “In the third space of Starbucks, people’s relationships are free and equal, there is no sense of hierarchy in the workplace, and there is no bondage of various roles in the family, and people can release their true selves.” Starbucks, in essence, aspires to inspire and nurture the human spirit, fostering a sense of community with every customer and every cup of coffee.

This sentiment resonates as a poetic and beautiful mission, evoking memories of opening the door to a Starbucks on a chilly winter day. The gentle warmth, soft chairs, and the ebb and flow of conversation create an atmosphere that beckons you to relax your heart and soul. Behind this seemingly effortless ambiance lies clever design and business acumen. Starbucks stores aren’t randomly laid out; every design element is orchestrated to enhance connections and communication. The low bar allows you to witness the artistry of a Starbucks barista crafting your drink. Smiling staff, akin to a night French bar, not only serve your coffee but also engage in conversations about coffee culture and the tales behind each cup.

The layout of Starbucks, from tables to chairs, strikes a delicate balance — neither too distant nor uncomfortably close. Whether you choose solitude or companionship, the environment accommodates both seamlessly. The horizontal orientation of the line encourages natural interactions. Observing others’ orders or striking up conversations becomes second nature, fostering an invisible connection that weaves through the Starbucks space, making it both comfortable and integrated. The design brilliance of Starbucks lies in its ability to create an environment that goes beyond serving coffee — it crafts a sanctuary where individuals can connect, converse, and be themselves.

Now, let’s explore another intriguing facet — Starbucks’ unique language. A viral video from the past humorously mocked Starbucks for deviating from the conventional “small, medium, large” ordering model, opting instead for “short, tall, grande.” At first glance, this linguistic departure might seem unnecessary. Yet, this distinctive language serves a purpose — it fosters a sense of community and erects a subtle threshold. Once you grasp this unique lexicon, you’re not merely placing an order; you’re proclaiming your membership in the Starbucks community. It’s akin to joining an exclusive group on an American campus — the cool, beautiful girls who operate with their own set of cultural norms and inside jokes. Although they project an air of exclusivity, secretly, they hope for new members to join, creating an air of sophistication. Starbucks employs a similar psychological strategy, striking a delicate balance. Their “threshold” isn’t too high; they aim to welcome as many customers as possible, creating an inclusive yet distinct environment.

Lastly, let’s dive into corporate culture — a vital compass that guides a company’s decisions. Corporations are faced with countless choices, big and small, and the key to maintaining a cohesive company image lies in unified, continuous, and consistent decision-making. Starbucks, in this regard, stands out. A company must navigate its relationships with three key stakeholders: employees, customers, and shareholders. The crucial question is determining the priority among these stakeholders. For Starbucks, the answer is clear — their employees take precedence.

At Starbucks, all employees are considered equal and are encouraged to respect one another, evident in their nomenclature as “Partners.” Regardless of rank, everyone is addressed as a partner. Work is rotated among various tasks — cash register, drink preparation, cleaning, and even toilet maintenance — eliminating hierarchies in responsibilities. When employees make mistakes — whether it’s a broken glass, a hiccup at the register, or a simple error — there’s a humane approach. As long as the mistake doesn’t involve moral wrongdoing, there’s no punitive action. If an employee faces unfair treatment, they have a direct line to the Seattle headquarters to voice their concerns.

While Starbucks employees may not boast the highest salaries, they find themselves in the upper-middle class. This compensation comes with a slew of benefits, including top-tier employee training, comprehensive health insurance, and support programs. Notably, all employees are afforded the opportunity to become shareholders in the company. They can purchase company shares at a discount, fostering a sense of ownership and aligning their success with the company’s growth. Starbucks, through its prioritization of employees and the cultivation of a supportive corporate culture, showcases a commitment to both its workforce and the overall Starbucks experience.

Delving deeper into Starbucks’ internal dynamics, its promotion system adds an intriguing layer to the brand’s allure. As foreshadowed earlier, keen observers will notice the distinct colors of Starbucks aprons, each carrying its own significance. The green apron denotes a regular clerk. However, for those aspiring to don a black apron embroidered with their name, a higher level of coffee expertise is requisite. This includes knowledge about coffee production, familiarity with various coffee flavors, awareness of the stories and cultural aspects of coffee, and the successful completion of a written test and interview. For the crème de la crème, the coffee-colored apron signifies the highest-ranking employee — a Coffee Master. These elite individuals compete in the Coffee Master competition, showcasing proficiency in areas such as latte art, tasting, creative drinks, blind tasting, and overall expertise.

Starbucks’ treatment of its employees with respect and equality isn’t just a token gesture; it’s well-earned. Those who choose to make a long-term commitment to Starbucks typically share a genuine interest and passion for coffee culture. Transforming their pride in coffee and the Starbucks brand into a driving force, employees channel this enthusiasm into providing exceptional customer service. By fostering a workplace environment where employees genuinely love their work, their company, and the culture of coffee, Starbucks ensures the delivery of the best third space and service. In strategically selecting employees, Starbucks is also, in a way, choosing its customers.

When scrutinizing the magic of Starbucks, it’s evident that they go beyond providing high-quality coffee and excellent service targeted at different customer segments. However, attributing Starbucks’ success solely to these aspects would be oversimplifying. The real secret sauce lies in their Branding Power. With a formidable brand, a company becomes irreplaceable, attracting customers, fostering loyalty, and gaining a subtle but powerful business capability.

Imagine weaving the enchanting tale of Starbucks into an analysis of coffee companies, emphasizing the enchantment of brands! The term “brand” isn’t exclusive to Starbucks — it’s a concept any coffee shop can harness during interviews, as each has the potential to create its own unique brand.

Yet, for Starbucks, there’s an added layer: culture.

Reflecting on Starbucks’ growth history, it becomes evident that Starbucks offers more than just coffee. It provides an extensive array of services, experiences, and a culture that underpins the brand. The relationship between the brand and people is a multifaceted story. Coffee itself is a culture — a symbol often portrayed in workplace dramas by a stylish, empowered woman, strolling down the street with a cup of coffee, signaling her identity, lifestyle, and social class. Similarly, pushing open the glass doors of Starbucks on a chilly winter day isn’t merely about the brand; it’s about personal identity, belonging to a social circle, and choosing a particular culture.

Yet, it’s worth noting that Starbucks has shifted over time, becoming less sophisticated and cultured. This evolution prompts us to ponder whether this change is an inherent part of the brand’s evolution or a strategic move in response to evolving consumer preferences and cultural shifts. The Starbucks story continues to unfold, shaped by the delicate interplay between its brand, culture, and the ever-changing coffee landscape.

As we harken back to our initial exploration of the coffee industry, one undeniable limitation looms large — the industry’s inherent ceiling. When attempting to calculate the overall market potential, a rudimentary formula surfaces: the total number of coffee drinkers multiplied by the average cups of coffee consumed per day, multiplied by the profit per cup, equals the potential daily profit. Yet, within this basic equation, an upper limit emerges — while consuming 0 to 1 cup a day is common, progressing from 1 to 2 cups is feasible, but the leap to 10 cups is highly improbable. Who, after all, consumes 10 cups of coffee daily? Thus, when Starbucks ascended as the unrivaled leader in the coffee market, the looming concern for the boss was unmistakable: How can we continue to grow when we’ve already conquered the market?

Starbucks brainstormed various strategies to address this challenge. One approach involved enhancing sales at individual stores — since coaxing people to drink more coffee had its limits, the foray into the food business was a logical next step. Starbucks evolved from being solely a coffee destination to a place where customers could enjoy breakfast and order baked goods. The geographical expansion also became a key tactic; having saturated the U.S. market, Starbucks set its sights on China, resulting in a proliferation of countless Starbucks stores across the country. Additionally, the integration of new internet and digital technologies aimed to boost Starbucks’ efficiency and customer convenience, allowing each store to serve a larger clientele.

However, there are critical nuances that deserve attention. The dynamics of the Chinese and American markets differ significantly. While Starbucks in the U.S. may now seem less sophisticated, Starbucks stores in China still exude an air of refinement. Yet, it’s essential to recognize that Starbucks in China and Starbucks in the United States aren’t necessarily identical. Starbucks originated in the United States, where coffee culture took root earlier, making coffee a more familiar and accessible choice for Americans. Consequently, Starbucks’ development in the U.S. was swift, but so too was its decline. In present-day New York, Starbucks stores lack the elegance and sophistication one might associate with the brand — they are more akin to ordinary coffee shops, with simple furnishings and no plush sofas, catering to a clientele that contrasts sharply with the urban white-collar workers of yore. In this evolving landscape, Starbucks appears to be transforming, inching closer to the model of a fast-food giant like McDonald’s, as noted by Zhang Xiaoyu in a classic business class observation. The once-sophisticated coffee experience is giving way to a more streamlined and standardized version, reflecting the shifting tastes and preferences of its clientele.

Certainly, the Starbucks narrative in the United States encompasses more than just the mainstream coffee shops — it extends to the exclusive Starbucks Reserve Roasteries. These establishments, one or two per city, boast luxurious decor, elaborate displays of coffee-making equipment, historical explanations, an extensive array of Starbucks souvenirs and merchandise, and a unique menu not found in regular stores. Rather than mere coffee shops, these locations transcend their utilitarian purpose to become tourist attractions.

In essence, Starbucks has effectively bifurcated its brand into two distinct entities: one aligns with the commercial profit-driven model akin to fast-food giants like McDonald’s, while the other aspires to be a “third space,” preserving the brand’s image and offering a space for service, experience, and culture. The success of this dual-pronged approach remains uncertain and warrants observation.

However, in the Chinese market, the cultural, lifestyle, and personal identity facets embodied by coffee haven’t diminished. Starbucks still retains its luster, especially in the realm of light luxury. This is why Starbucks aims to tap into the underlying business logic of continuing to generate profits in these emerging markets.

This observation prompts us to acknowledge that, beyond the alluring charm of the Starbucks brand, at its core, the company faces the imperative to return to the essence of business — seizing more interests, expanding stores, refining coffee taste, enhancing efficiency, reducing costs, and improving earnings. It underscores the fundamental truth that, despite the captivating allure of brand narratives and cultural aspirations, every business ultimately exists to make money.

Yes, a conventional business analysis model may appear mundane and cliché, but its solid framework ensures we don’t lose sight of a company’s essential mission. Starbucks may seek culture, brand, and the creation of free spaces for people, but, above all, it aims to turn a profit.

After all, who doesn’t want to?

In recent times, the tech industry has been abuzz with the advent of OpenAI’s ChatGPT. From the viewpoint of everyday individuals, there’s a growing concern about AI potentially replacing jobs. Bloggers are quick to create short videos sharing their insights, and youngsters are experimenting with integrating ChatGPT into their homework routines. However, it’s worth noting that many people are panicked with this “vicious” technology. Today, I want to demystify ChatGPT from a business perspective as a student, exploring the impact it may have on the industry. While it might seem challenging to encapsulate the vast landscape of the technology industry and generative AI in a single article, consider this piece a small appetizer, hoping it sparks your interest.

Now, let’s start by clarifying what ChatGPT exactly is.

1. What is ChatGPT?

The prowess of this machine chat box lies in the last three letters: GPT, which stands for Generative Pre-trained Transformer. “Generative” refers to its ability to generate responses, “Pre-trained” indicates its learning from vast amounts of data beforehand, and “Transformer” denotes its deep learning model structure. In essence, GPT signifies a pre-trained, generative model based on the Transformer architecture.

Let’s draw a parallel with how humans learn.

In school, we often experience supervised learning, where the teacher provides standard answers for subjects like chemistry, and defining right and wrong. Sometimes, teachers explain the principles behind formulas, while other times, they may opt for a more direct approach, saying, “Just memorize the formula!”

Besides supervised learning, there’s unsupervised learning, where individuals, driven by a strong self-learning inclination, study independently, relying on textbooks without internet checks, solving problems on their own, and mastering the material.

A third approach is reinforcement learning, analogous to an athlete receiving guidance from a coach during practice, with corrections made for any mistakes.

Supervised learning and reinforcement methods, while effective, necessitate continuous oversight. On the contrary, self-learning imposes a substantial burden on machines. How does ChatGPT tackle this challenge? Through direct data feeding. Envision artificial intelligence as a black box; after inputting diverse parameters, it boldly asserts, “I’ve completed the learning process!” Upon testing its capabilities, we discern, “Wow, it indeed learned.” Yet, the enigma persists concerning what precisely it has acquired.

We, at least myself, always have a vague awareness that different parts of the world connect together and share some universal features. There is just this sort of emergence of numerous, intricate connections in the world, which is hard to summarize by a universal formula. Humans invest substantial time in comprehending these inexplicable phenomena, but we are sort of being restrained by our physical limits such as memories or calculating abilities. In contrast, artificial intelligence thrives on limitless data, where quantitative shifts lead to qualitative transformations. Unfatigued and minimally energy-consuming, these machines eventually decipher the intricate laws of the world, often elusive to human understanding. The result? Robots are capable of capitalizing on this newfound knowledge.

Let us keep in mind that GPT is actually an extraordinary language model. The way they learn language is similar to humans. I can speak English and Chinese, but it is hard for me to write a Chinese textbook to teach an English speaker how to speak Chinese. I naturally grasp those languages without consciously calculating those grammar rules in my mind. Similarly, during GPT’s training, no rules of natural language were explicitly taught; it autonomously organized itself after consuming vast amounts of data. The GPT model resembles the human brain more than a conventional computer program.

Analogous to the intricate neural structures within our brains that elude our full comprehension, GPT’s learning process mirrors a self-unfolding journey. Various neural networks intertwine, and suddenly, clarity emerges. When GPT operates, it generates text, predicting the next character by assessing the likelihood within the context of the preceding and succeeding characters. It doesn’t truly comprehend the content; it merely recognizes that, through data consumption, it can formulate responses to your questions.

If you are a student struggling with Math 220, you might observe that ChatGPT falls short in performing calculations. It’s by design! When you pose a question, ChatGPT, fueled by the extensive data it digests, calculates probabilities to assist in formulating a formula and outlining the logic. However, the numerical output after the formula is presented is inherently flawed. Why? ChatGPT doesn’t compute the formula; it gauges the probability of a 1 or 2 occurring before or after the “=” sign. Hence, ChatGPT is not here to do rigid calculations. Instead, it is designed for brainstorming and creative works.

I used to write an article (not posted it yet, will do so in the future) delved into my experiences using ChatGPT to enhance daily life, and addressed concerns about job displacement (spoiler alert: it’s unlikely on a large scale). Curious minds can explore those insights. Yet, let’s steer this discussion toward the business realm, concluding our exploration of ChatGPT and AI in a broader context.

So, what makes ChatGPT significant from a business standpoint? How has it reshaped the tech industry, and what lies ahead?

2. Let’s get down to business

In fact, it’s not just ChatGPT but Generative AI as a whole that holds commercial significance. Across the corporate landscape, nearly every company is integrating Generative AI into its business model, pondering over the strategies to capitalize on this transformative technology. The key question echoing in boardrooms is, “How do I leverage Generative AI for profitability, and what measures can I take to navigate this paradigm shift?”

To grasp Generative AI’s impact on the industry, a foundational understanding of the tech landscape is essential. Have you ever wondered why the Technology, Media, and Telecoms (TMT) sectors are grouped together? Ben Thompson, in a Stratechery article titled “AI, Hardware, and Virtual Reality” (September 28), provides a compelling explanation — they share colossal upfront costs.

Consider a Chinese app, WeChat, a software entity. Initiating revenue generation requires a substantial investment in hiring engineers and programmers to develop the WeChat software. The early stages of crafting this program entail numerous unknowns, risking significant expenditure for a product that may not resonate with users. Media giants like Disney face hefty upfront costs for both theme parks and movie production, while telecom giants like China Mobile must invest extensively in building communication networks before reaping profits.

Despite these hefty upfront costs, a common thread ties these sectors together — near-perfect scalability once operational, equating to almost zero marginal costs. Once WeChat software is deployed, increased user adoption directly boosts revenue without a proportionate increase in costs. This scalability principle applies similarly to Disney’s parks and China Mobile’s communication networks.

However, these TMT sectors, while breaking the constraints of human physical existence (space and time), grapple with a shared limitation — human attention and time.

3. Transcending Boundaries

Human attention and time — finite resources tethered to our biological makeup — have long dictated the choices we make. With only twenty-four hours in a day, sixty minutes in an hour, and sixty seconds in a minute, the constraints of time are inescapable. However, the advent of Generative AI propels us into a realm where we inch closer to the divine, surpassing the limitations of the human experience on the supply side.

Limited attention compels us to make choices — read a novel, watch a movie, or engage in conversation — within the confines of a set timeframe. The dream of enjoying all three simultaneously remains elusive due to the finite nature of time. The scarcity of attention and time on the demand side presents a challenge, a coveted and contested resource in the competitive landscape.

Yet, on the supply side, Generative AI emerges as a transformative force. The once significant upfront costs associated with media production, telecommunications infrastructure, and human interactions are poised to dwindle rapidly. Picture a scenario where scripts are initially brainstormed and written by Generative AI, potentially evolving to complete autonomy in later stages. Article writing, news generation, and film production — all integral components of the creative process — will undergo functional differentiation with Generative AI at its core, enhancing productivity and cutting costs. This paradigm shift extends to the software realm, where programs traditionally crafted by numerous engineers and programmers are gradually infiltrated by Generative AI, an indispensable ally in production efficiency and cost reduction. Telecom, too, undergoes a similar transformation.

Consider ChatGPT as a prime example. While I cherish conversations with friends on WeChat, the limitations of response times and availability often lead to moments of boredom. With ChatGPT, these constraints vanish. Interaction becomes boundless — I can converse at any time, on any topic. Despite the acknowledgment that ChatGPT isn’t human, the engagement provides an avenue for learning and joy. This AI chat box, designed not solely for conversation, exemplifies the evolving landscape. Platforms like Meta’s Meta AI and Character.ai, fueled by diverse datasets, introduce AI entities with distinct personalities, including those of celebrities and actors. Imagine conversing with Elon Musk’s AI on startup advice or discussing future songs and relationships with Taylor Swift’s AI. The richness of interaction rivals, if not surpasses conversations with real friends.

In the past, the object of chat was a limited resource, epitomizing scarcity. Now, Generative AI’s proliferation transforms this dynamic, making the number of chat companions infinite — an era of abundance. Celebrities, once scarce and revered, become highly objectified entities, accessible for interaction anytime, anywhere. This shift from scarcity to abundance forms a compelling framework for understanding contemporary business phenomena.

Yes, Generative AI is pushing us beyond human limits once again, what clues does this give us? What is likely to be the way forward and the big trends in the tech industry? What’s important about ChatGPT’s new voice chat feature? What are the implications of Meta’s new eye and graphics generation?

Let us discuss this more in the next article!

So let’s dive in!

Amazon is one of the greatest companies in the world because it not only influences how we live in the world but also because of its CEO Jeff Bezos’ philosophy and his great ambition. While many people might interpret Amazon as an e-commerce retail company, platform company, or maybe AI company, that is far more simple compared to the real intention of Amazon. What Amazon finally wants to do is “collect tax” from ALL online activities.

In order to understand this great ambition, we need to understand four big strategies that Bezos created in the long term: Flywheel Effect, Customer-Is-the-God, Dominant Strategy, and its “Super Interface”, which all reinforce each other to create an open and flowing ecosystem.

1. First, let us talk about the Flywheel Effect.

At the start of Amazon’s 2014 Letter to Shareholders, Bezos wrote:

“A dreamy business offering has at least four characteristics. Customers love it, it can grow to a very large size, it has strong returns on capital, and it’s durable in time — with the potential to endure for decades. When you find one of these, don’t just swipe right, get married.

Well, I’m pleased to report that Amazon hasn’t been monogamous in this regard. After two decades of risk-taking and teamwork, and with generous helpings of good fortune all along the way, we are now happily wed to what I believe are three such life partners: Marketplace, Prime, and AWS. ”

Prime services serve consumers. If you pay membership for Amazon, you can enjoy a series of amazing services, including free shipping; unlimited streaming of movie and TV episodes; discounts to all shopping categories like Amazon Essentials, Whole Foods, Amazon Fresh, etc; free and early access to e-books on Kindle. Currently, the subscription is over 200 million, and the annual revenue is $35B.

Marketplace serves third-party sellers. Except for Amazon’s own products, all the other sellers can also join the Amazon platform and directly sell their own products. There are over 2 million third-party sellers on Amazon, representing 60% of the units sold.

Amazon Web Services (AWS) serves big enterprises and small businesses by providing cloud services with databases, storage, software, machine learning, etc., where all enterprises can set up their IT system on this AWS platform. It is cheap, efficient, and very stable, much more convenient compared to establishing a system by itself. The customer base includes Verizon, Netflix, and GE, which is one of the best cloud service providers, generating $62B in revenue, which is the most profitable segment.

These three services created the flywheel effect, which happens “when small wins for your business build on each other over time and eventually gain so much momentum that growth almost seems to happen by itself”. Think about these three services, prime services increase customer loyalty. If you purchased the prime services, you tend to buy more — because you buy more you save more based on the fixed amount of money you paid every year — and customers are not going to leave. With Marketplace, customers’ choice of products increases, and they are more willing to buy Prime Services and also stick with Amazon in the long run. With more customers, third-party sellers are also more willing to join the Amazon Platform to sell their own products. When customers increase, Amazon’s market power to negotiate with the supplier will increase, so they will also be able to share profits with their customers, pushing third-party sellers to control their price or quality at the same time. Amazon also has Fulfillment by Amazon (FBA) services that help third-party sellers transport their products, making these third-party sellers sticky to the platform as well. While with the Marketplace, more third-party enterprises will also stick with Amazon because their whole IT system is established upon it!

These wheels rotate increasingly fast, generating amazing cash flows for Amazon in the long run.

2. But how does Bezos come up with this system? It is his belief that “Customer is the God.”

Every business has a slogan saying how much they care about their customers, but no one is willing to sacrifice their own profits to do it- except Amazon.

For example, Amazon developed a function for customers to put comments on their products– however, many customers will post negative comments that decrease purchases. In fact, according to their survey, customers do stop purchasing when they see negative comments. However, Bezos said: “We don’t make money when we sell things. We make money when we help customers make better purchase decisions.” In fact, this function seems to hurt the profits of Amazon but helps provide better services for customers, thus increasing revenue in the long term.

Moreover, think about Marketplace, which is also hurting Amazon’s own profits, because sometimes customers will choose products from third-party sellers instead of Amazon’s own products. However, from Bezos’ perspective, if there is a product, and its prices or quality is better compared to Amazon’s own product, it is better for customers to get it easily on Amazon instead of spending great efforts to find it somewhere else. Same logic– the customers’ profits are Amazon’s profits.

Another crazy example that can demonstrate this belief is this: Target used to be a ten-year customer of Amazon’s AWS services, but they are planning to leave. Amazingly, Amazon even helped Target to transfer all the data from AWS to their own IT system, although their own system broke down after a while.

Bezos’ strategy of putting customers’ profits in the first place is the core of their strategy, which not only explains why their three services can create such a Flywheel Effect but also makes Amazon the most profitable firm in the long term.

We know that Amazon starts by selling books online. But in fact, Bezos knows that selling books is not his only goal in the long term, but is the best starting point.

3. Why is that? It is because of the Dominant Strategy.

For the retail industry, the theoretical framework is quite mature: the most important thing about customers’ experience is rich product offerings, fast speed, good quality, and cheap prices.

Usually, when a company starts to do business, it will focus on one or two areas to focus on and use that to attract customers. For example, Walmart focused on rich product offerings and cheap prices, and Costco focused on high quality and cheap prices. In order to compete, they need to sacrifice some areas in order to take some part of the market share. However, if you find a strategy that does not change whatever your competitors do, you should do it, because this is the Dominant Strategy. In the retail industry, if you can find a way that ensure four qualities at the same time, you should ignore your competitors and do it directly.

Bezos found this Dominant Strategy, and the reason he is the one who discovered is that he realized the advanced technology of the internet can achieve all in the long term period.

At the initial stage, Bezos has already said: “Our vision is to be earth’s most customer-centric company; to build a place where people can come to find and discover anything they might want to buy online. To achieve this, we have initially focused on the book industry.”

And the reason he chose the book industry is obvious if we think about the advantage brought by the internet:

Without physical barriers, the book offered online will definitely be richer compared to the physical bookshop, so it satisfies “rich product offering”. There is no need to consider the quality of the book because the book is pretty standardized– the one you bought online will be the same in the bookshop, which directly skipped “high quality”. Because there are no rental costs, the online market has lower expenses, Amazon is able to provide cheap services. And finally, the delivery services can improve and be super fast depending on the future internet advancement.

Therefore, Amazon started by focusing on the book market because it satisfied all the requirements of customers. At the same time, Amazon is transporting different products to online shops, while polishing their own services to expand to different areas. With their long-term planning and belief in a dominant strategy, they are able to form a very wide moat that no one can challenge at a later stage.

Amazon’s other core concept of Amazon is to transform its internal functional businesses into external service-oriented ones.

4. I would like to call this Amazon’s “Super Interface”.

Think about Amazon’s FBA services. Establishing an internal delivery system is the essential task of a large e-commerce company, but Amazon prompts this system to a new level, which not only serves Amazon itself but is also standardized and modularized to serve external customers. That goes the same with their AWS system. When Amazon first thought about developing a more efficient internal IT system to support business operations, they realized that an independent cloud service IT system is not an easy task, and in fact, all companies will need this. Therefore, they decided to make AWS an external service as well.

There are several advantages to doing this. First, it definitely provides a new channel to generate revenues. But more importantly than that, it gives Amazon’s internal group to try new businesses with little cost like start-ups. They can even do what Startup does “Agile development with rapid iteration” to establish the services. This kind of service will have the first major customer– Amazon themselves, so they will know how to satisfy customers’ demand on the right track, even if this business is not so successful, the worst situation is that Amazon itself can still use it.

Moreover, once Amazon puts these services as an external one, the service is exposed to a real competitive environment and faces real survival issues, turning them into a real successful enterprise. Think about the Flywheel effect again, with more customers for the services, they can’t leave Amazon, and they really “grow” on the Amazon platform and this makes you lose the ability to imagine a world without Amazon. In fact, Amazon, in this sense, is no longer a company anymore, it is an interface that everything connects to and cannot leave with it. Just like an open ecosystem that supports all the creatures in the world.

While with all these four features of Amazon, we can see how Amazon is a really great company. But there is a final step in order to understand the essence of Amazon, which is …

5. Amazon’s final ambition — to “collect taxes” from all online activities.

If you want to use the internet, you will pay for Amazon.

Think about the business models of FBA and AWS, they all have a common feature: They all have high upfront costs and fixed costs, which are beneficial from Economic of Scale — more people using it, those costs become relatively cheap.

Now let us look at Amazon’s recent acquisition of Whole Foods. Whole Foods is not as standardized as books or other retail products, therefore Amazon is not so successful in this market. Therefore, in the past, the only thing you cannot buy from Amazon is fresh and organic food. However, fresh groceries also have the same feature described above– high upfront costs but benefits from Economies of Scale. Therefore, it fits into Amazon’s system very well, while also reconstructing a whole supply chain for fresh groceries, something Amazon does not have in the past, and making Amazon comprehensive at a new level. Whole Foods will help Amazon to provide services to individuals, small restaurants, and big catering enterprises.

From this acquisition, we can clearly see the ambition of Amazon.

While living in 2023, what do we usually do online? Buy necessities, shop, entertain, write, start your own business, manage big enterprises… And in fact, Amazon provides all these services.

If you are a consumer who wants to purchase anything– buy Prime.

If you own a small business — come to Marketplace and sell your products to more people.

If you own big enterprises– build your IT services on AWS.

If you are a creator– let us create good content together and Amazon will help you to publish it.

…

Therefore, with Amazon’s roots spreading over the earth, you live with it and rely on it. You pay for all the activities you do online for Amazon!

Thanks to Ben Thompson from Stratechery and Zhang Xiaoyu from DeDao.