AI’s Biggest Winners May Not Be the Companies Building AI

Salutations, Olio aficionados! 👋

Welcome to the 302nd edition of Weekly Olio. We’re thrilled to introduce a fresh new twist to your Sundays: Publisher Parmesan, our hand-picked, thoughtfully crafted edition designed to spark inspiration and insights for the week ahead.

It’s the perfect way to unwind, recharge, and prepare for the week with something truly worth savoring.

If you’re new here and you’re looking for more long-form, crispy writing, click the link to subscribe under this GIF 👇

Choose individual real estate investments backed by a $12B owner/operator that invests 20%+ alongside individual LPs  in every offering.

Accredited Investors only. $100k minimum investment.

Learn More

^{This communication is for informational purposes only and does not constitute an offer to sell or a solicitation of an offer to buy any securities. Any such offer or solicitation will be made exclusively through the definitive offering documents. All investments involve risk of loss, including the potential loss of principal. Past performance is not a guarantee of future results. Any targeted returns or projections are forward-looking statements, are based on current assumptions, and are not guarantees of future performance. Actual results may differ materially.}

Every technological revolution creates an obvious set of winners. The more interesting question is who captures the profits after the initial excitement fades.

In AI, the obvious winners are the companies building models, chips and data centers. But a different group may ultimately produce better returns: businesses that already own customers, proprietary data and deeply embedded workflows, and can use increasingly cheap intelligence to serve those customers better.

Amazon, Meta and Netflix offer three different versions of this opportunity. All are using AI to improve their existing businesses, but their investment strategies are diverging. Amazon and Meta are spending aggressively on infrastructure, while Netflix is using external models to improve content production without taking on comparable compute costs.

The distinction matters because AI is doing two things simultaneously. It is making products more valuable, and making the underlying technology easier to replicate. The enduring advantage may therefore belong not to whoever owns the best model, but to whoever can turn intelligence into better customer outcomes fastest.

The dot-com boom left investors with a familiar lesson: exciting technology does not automatically produce attractive investments. Companies can transform entire industries and still destroy shareholder value if they spend too much, grow without discipline or fail to build defensible economics.

That lesson remains relevant to AI. But it may not tell the whole story.

The internet made information cheaper to distribute. AI makes certain kinds of work cheaper to perform. It can help someone discover a product, recommend a video, write software, analyse a document or generate a scene for a film. In each case, the technology can reduce costs while improving the experience.

That combination is more powerful than cost reduction alone. If a company can serve customers better and do so at a lower marginal cost, it has the potential to expand both demand and profitability.

The evidence is beginning to appear in the operating metrics of established platforms. Amazon says customers using its AI shopping assistant spend more per order. Meta says improved recommendation systems are increasing time spent on its apps. Netflix is using generative AI to produce certain visual effects faster and more cheaply.

These are different businesses, but the underlying pattern is similar. AI is not simply replacing an existing cost. It is changing how effectively a company can serve a customer.

For investors, that raises a more important question than which model scores highest on a benchmark: who has the business model, customer relationships and data to turn better AI into lasting cash flow?

Consider Amazon. An AI shopping assistant is easy to describe as another interface layered on top of an online store. But that understates its potential.

If the assistant helps a customer find the right product, reduces friction and increases the likelihood of completing a purchase, Amazon can generate more economic value from a relationship it already owns. According to Amzon, customers using Alexa for Shopping spend more per order, while users who try Alexa+ sign up for Prime at higher rates.

The same logic applies to Meta, although its product is attention rather than transactions. Its recommendation systems determine which content people see, how long they stay and which advertisements are relevant to them. Better ranking can increase engagement and improve the likelihood that advertising produces a result.

Netflix is applying the technology further upstream. Its opportunity is not primarily to sell more compute or build a general-purpose assistant. It is to make entertainment more compelling and less expensive to produce. Case in point, the AI-enhanced footage for The American Experiment being produced in half the time and at half the cost of previous options.

Each example illustrates a different way AI can improve the economics of an existing customer relationship. Amazon can increase purchasing efficiency. Meta can improve discovery and advertising performance. Netflix can lower the cost of creating the content that attracts viewers.

The strategic advantage is not that these companies have discovered AI. It is that they have enormous distribution systems in which even a modest improvement can affect billions of customer interactions.

For years, software itself was a powerful competitive advantage. Building a sophisticated product required engineering talent, time and capital. Competitors could copy features, but doing so was expensive enough to provide some protection.

AI is changing that equation. As models improve and become more accessible, the cost of reproducing many software features is likely to fall. A product that once required a large engineering team may become easier for a smaller competitor to build.

That does not mean every software moat disappears. It means the moat increasingly depends on what sits behind the interface.

Amazon has transaction histories, product searches, purchase patterns and information about how customers respond to recommendations. Meta observes interactions across a vast social network. Netflix learns from viewing behaviour, completion patterns and audience preferences.

This information is valuable because it is connected to real outcomes. A model can generate a plausible recommendation, but a platform with years of customer behaviour can test whether that recommendation leads to a purchase, a longer viewing session or a more satisfying experience.

People in the industry call this dynamic Ontology Velocity: the rate at which a company combines proprietary data and AI to improve its product, attract more users and generate still more useful data.

The idea is intuitive even if the terminology is new. A company that learns slightly faster can deliver slightly better results. Better results attract more usage. More usage produces more feedback. That feedback improves the next round of decisions.

Over time, a small initial advantage can become meaningful.

But there is an important qualification. Data volume alone is not a moat. Much of the information collected by large platforms is noisy, redundant or difficult to translate into better decisions. The advantage comes from the ability to turn data into measurable improvements, repeatedly and economically.

The winners will not simply be the companies with the most data. They will be the companies that learn the most from it.

The DTC beauty category is crowded. To break through, Jennifer Aniston’s brand LolaVie, worked with Roku Ads Manager to easily set up, test, and optimize CTV ad creatives. The campaign helped drive a big lift in sales and customer growth, helping LolaVie break through in the crowded beauty category.

Learn more

This is where the comparison between Amazon, Meta and Netflix becomes particularly interesting.

Amazon and Meta are not merely using AI tools to improve their existing products. They are also investing heavily in the infrastructure that runs AI. That creates a substantial near-term cash burden.

The combined trailing capital expenditure has risen to $262 billion, up from $75 billion. Also the free cash flow per share for a basket containing one share of each company has fallen from a peak of $24 to $14.90, with Amazon's own figure turning negative.

Those numbers matter because stock prices ultimately depend on the cash a business can generate for shareholders over time. Revenue growth and better engagement are encouraging, but they do not automatically justify any level of spending.

The bullish interpretation is that Amazon and Meta are building a second platform alongside the first.

Meta can use its infrastructure to improve recommendations, automate internal work and develop AI assistants for consumers and businesses. Amazon can use AI to improve its retail operation while selling cloud computing and custom chips through AWS. Its infrastructure investments may therefore support both its internal operations and an external business.

This is a significant strategic option. The same infrastructure that improves an existing platform could become a product sold to other companies.

But an option is not the same as a realised return. The investment case depends on how much demand emerges, how quickly the infrastructure becomes productive and whether the eventual cash flows compensate shareholders for the capital committed today.

The market must eventually distinguish between spending that creates a durable second engine and spending that merely keeps a company in the race.

Netflix offers a useful contrast because it can benefit from AI without necessarily owning the infrastructure on which AI runs.

Its strategy is to use whichever models are best and cheapest for a particular task. If external providers compete to make their models more capable and less expensive, Netflix can potentially capture those improvements without funding the entire underlying infrastructure buildout.

That creates an appealing economic structure. Netflix owns the audience, the distribution platform and the viewing data. It can use AI to improve recommendations and production workflows while allowing model providers to compete on cost and capability.

Netflix's annual capital expenditure remains below $1 billion and free cash flow per share has risen roughly 70% since the end of 2023. These figures are presented in contrast to the much heavier infrastructure spending at Amazon and Meta.

The broader lesson is not that Netflix's approach is automatically superior. A streaming service and a hyperscale cloud provider have different requirements. Netflix may not need to build foundational compute infrastructure because its core business does not depend on selling that capacity to the world.

But its position highlights a strategy worth watching: own the customer relationship and the workflow, rent the intelligence, and keep the savings.

As AI becomes more capable and less expensive, companies following this model could see their margins improve without having to finance the entire technology stack.

The risk is that competitors can do the same. Netflix's advantage will depend on its ability to convert cheaper intelligence into better entertainment, not simply on access to external models.

There is a third model: companies that do not just use AI internally, but sell systems that help other businesses turn their own data and workflows into better decisions.

Palantir is the example here. Its proposition is not simply access to a language model. It connects data, operational processes and AI within a customer's environment, allowing the system to support decisions across areas such as defence, manufacturing and law.

The argument is that each deployment can deepen the customer's operational system. As more workflows become connected, replacing the platform becomes more difficult. Improvements in underlying models can also make the product more capable without requiring the vendor to develop every improvement itself.

That is an attractive business model if it works. A company can benefit from advances across the AI ecosystem while retaining the customer relationship and the workflow layer.

Palantir's reported net dollar retention is 157% with a rise in quarterly free cash flow per share from $0.06 in September 2023 to $0.47 in June 2026. Those figures suggest expanding customer spending and improving cash generation, though they do not by themselves establish how much of the performance is attributable to AI.

Palantir also illustrates why the distinction between owning data and owning the system that makes data useful matters. Customers may already possess the information they need. Their problem is that the information is fragmented, difficult to interpret and disconnected from daily operations.

A platform that makes it actionable can become deeply embedded.

Still, there is a valuation question. Even a business with strong retention and attractive unit economics can be a poor investment if its stock price already assumes extraordinary growth. The quality of the business and the price paid for it remain separate questions.

The AI investment debate is often framed as a contest between model builders. Which laboratory has the best system? Which chipmaker will capture the most spending? Which cloud provider can build the largest infrastructure base?

Those questions matter. But they focus on the supply of intelligence rather than the economics of using it.

As models become more capable and competition pushes down their cost, the ability to access intelligence may become less distinctive. The harder problem will be applying it to a specific customer, within a real workflow, in a way that produces a measurable improvement.

That favours companies with trusted distribution, proprietary feedback and problems worth solving. It also favours companies that can benefit from falling AI costs without having to finance every layer of the stack.

Amazon and Meta are betting that owning the infrastructure will create strategic options beyond their existing businesses. Netflix is betting that it can capture the benefits of better models while remaining relatively capital-light. Palantir is betting that the operational layer connecting data to decisions can become indispensable.

These are different paths, not a single formula.

But they point toward the same question for investors: when intelligence becomes cheaper, who gets to keep the savings?

The answer may determine which AI businesses merely grow, and which ones turn that growth into lasting free cash flow per share.

Missed reading the recent editions? Check out our previous coverage here:

What does it look like when a consumer brand starts to break out?

Popsmith generated $5.8M in revenue in 2025 and is projecting $9M in 2026, with more than 100,000 customers and $20M+ in lifetime sales.

The brand has already earned national recognition through Oprah’s Favorite Things and partnerships with Costco, Sur La Table, and Williams Sonoma, where Popsmith is rolling out nationwide this holiday season.

Now Popsmith is raising capital to accelerate its next stage of growth, and individual investors have the opportunity to become shareholders.

Explore the Popsmith investment opportunity.

That’s all for this week. If you enjoyed this edition, we’d really appreciate if you shared it with a friend, family member or colleague.

We’ll be back in your inbox 2 PM IST next Sunday. Till then, have a productive week!

Peeyush and Mohit

Disclaimer: The views, thoughts, and opinions expressed in the text belong solely to the author, and not necessarily to the author's employer, organization, committee or other group or individual.