The AI Boom May Have a Business-Model Problem Too

For most of this cycle, the AI debate sounded like a scale debate.

Who has the most chips? Who can build the most data centers? Who can spend the fastest? Who can lock up the best models, customers, and talent? That was the easy version of the boom. The harder version is showing up now: what if the real issue is not only how much the industry spends, but whether the business model underneath that spending is stable enough to justify it? Reuters reported this week that Big Tech is on pace to spend about $630 billion on AI infrastructure in 2026 and more than $800 billion in 2027, even as rising yields make that spending harder to finance and investors increasingly question whether returns will arrive fast enough.

That is the backdrop for a more uncomfortable question. The AI boom may have a business-model problem too. Not because demand is fake. Demand is real. The problem is that the revenue engine still looks less mature than the capex engine. Some companies are showing real monetization. Others are still leaning on run-rate metrics, new ad experiments, enterprise partnerships, and very large assumptions about future usage. Meanwhile, investors are becoming much less willing to fund open-ended AI spending unless they can see a cleaner path from compute to cash flow.

The spending model got ahead of the revenue model

The easiest part of the AI boom to understand is the spending. It is massive, visible, and getting bigger.

Reuters reported that the top hyperscalers are rapidly burning through cash to fund AI data centers, chips, and cloud infrastructure. Apollo estimates that about 60% of their operating cash flow funded capex at the end of last year; Reuters says that figure is now close to 70%. Morgan Stanley estimates cumulative capex this year and next could consume nearly 90% of expected operating cash flow. Bank of America expects hyperscaler debt issuance to hit $175 billion in 2026, up from $121 billion in 2025. In other words, this is no longer a story about casual internal reinvestment. It is becoming a story about debt, hurdle rates, and whether the payoff justifies the financing burden.

That by itself does not mean the boom is broken. It means the industry has moved into a phase where the quality of the business model matters more. When capital was cheap and investor enthusiasm was high, companies could get away with saying the opportunity was too large to miss. Now that yields are higher and balance-sheet pressure is rising, the market wants more than ambition. It wants evidence. Reuters described this as a shift toward skepticism that these investments will generate adequate returns, especially if borrowing costs keep rising.

AI leaders are still experimenting with how to make money

This is where the business-model problem gets more obvious.

OpenAI’s growth is real. Reuters reported that OpenAI topped $25 billion in annualized revenue at the end of February, up 17% from the end of 2025. Out of that, about $10 billion came from enterprise AI, according to a Reuters report on its private-equity talks. That sounds impressive, and it is. But it also tells you something important: even at this scale, OpenAI is still actively looking for new ways to distribute and monetize its products. Reuters reported that it is in talks with private-equity firms including TPG, Bain, Brookfield, and Advent to form a joint venture that would push its enterprise tools across portfolio companies and beyond. That is not the behavior of a company that feels its current distribution model is already sufficient.

OpenAI is also moving into advertising. Reuters reported on March 26 that its U.S. ads pilot inside ChatGPT crossed a $100 million annualized revenue mark within six weeks, with more than 600 advertisers already participating. The company says the ads do not affect outputs and do not share conversation data with marketers. Still, the strategic point is hard to miss: a company at the center of the AI boom is now testing an ad model to help fund the high cost of developing the technology. That does not mean subscriptions and enterprise contracts are failing. It does mean they are not enough to stop the search for additional revenue streams.

That is exactly what a business-model transition looks like. A company starts with paid subscriptions and enterprise seats, then adds consumption billing, partner revenue, and advertising as it tries to match soaring infrastructure costs with more diversified monetization. Investors should not read that as weakness by itself. They should read it as proof that the final revenue model is still being built in real time.

The revenue numbers are less clean than the hype makes them sound

Another problem is that AI revenue is often reported in ways that flatter momentum more than stability.

Reuters Breakingviews wrote this month that Anthropic’s reported revenue metrics illustrate the issue well. In a court filing, Anthropic’s CFO said revenue had exceeded $5 billion to date. At the same time, the company had touted a much larger annualized run-rate figure. Reuters explained that this gap partly reflects the way run-rate numbers are calculated and the fact that about 80% of Anthropic’s revenue comes from enterprises billed on a consumption basis. That makes the headline number sensitive to optimization, pricing changes, promotional credits, and temporary spikes in usage.

That does not make the growth fake. It makes it noisy.

When much of your revenue depends on metered enterprise usage, customers can optimize queries, switch workloads, negotiate discounts, or slow deployments. Reuters also noted that open-source competition is scaling quickly and putting pressure on pricing across the industry. That matters because a business model built on heavy compute and usage-based billing needs customers to consume at very large scale without getting too efficient too quickly. If customers learn to get similar output with fewer tokens, cheaper models, or smaller deployments, the monetization curve gets harder to defend even while industry adoption keeps rising.

This is why the AI business-model problem is not the same as a demand problem. People clearly want the tools. The harder question is whether the industry can charge enough, consistently enough, to cover what it is building.

The economics are still brutal

The raw arithmetic is what makes this issue impossible to ignore.

Reuters Breakingviews reported that HSBC estimates OpenAI may need $207 billion in additional financing by 2030 and could burn nearly $280 billion in cash over that period. The same piece said Anthropic had spent more than $10 billion on training models and serving user queries while generating about $5 billion of cumulative revenue. Reuters concluded that the arithmetic looks unforgiving unless computing costs fall sharply or customers pay much more.

That sentence gets to the heart of the problem. There are only a few ways out. Compute gets dramatically cheaper. Pricing power gets dramatically stronger. Usage grows so fast that scale overcomes margin pressure. Or some combination of all three. Those outcomes are possible. But when the industry is simultaneously spending hundreds of billions on infrastructure and still leaning on experimental or evolving revenue streams, investors have a right to ask whether the business model is catching up to the capital structure.

Big Tech is now exposed to the labs’ business models too

This is not only a problem for OpenAI or Anthropic.

Big Tech’s AI capex increasingly depends on AI labs and customers monetizing the infrastructure they are consuming. Reuters reported in January that OpenAI accounted for 45% of Microsoft’s cloud backlog, a concentration risk that immediately mattered more once Azure growth underwhelmed and Microsoft’s shares fell 10%. Reuters also noted that Meta gained 10% the same day because investors could see AI already improving ad targeting and revenue. The contrast was sharp: Wall Street will tolerate huge AI spending when it can see the payoff, but it is quick to punish spending that arrives ahead of obvious returns.

Amazon faced the same pressure. Reuters reported that its plan to spend $200 billion in 2026 sent the stock down 11.5% after hours, even though AWS demand remained strong. Analysts quoted by Reuters said the market had a clear message: soaring AI spending can continue only if companies show commensurate operational or financial returns. One analyst said Amazon has to invest at those levels just to stay in the race, which is exactly the point. The race itself may be economically rational, but that does not mean every participant earns a good return on the capital required to run it.

This is where the business-model problem becomes a market problem. Hyperscalers are not only betting on their own engineering and scale. They are also betting that the companies consuming their AI infrastructure can turn usage into profitable, durable businesses. If the labs or AI software layers fall short, the entire stack feels it.

This does not kill the AI story. It changes the test.

The market is not saying AI is over.

It is saying the next phase needs proof. Reuters reported that investor unease has grown as Alphabet, Amazon, Microsoft, and others raise spending plans while software and data companies also face disruption from the same AI wave. The market is beginning to punish heavy spend even when the underlying business is still good, simply because the spending now looks more capital-intensive and less obviously self-funding than it did a year ago.

That means the AI boom now faces two tests at once. One is financial: can the returns outrun the rising cost of capital? The other is commercial: can the biggest AI companies lock in business models that are durable enough to support that spending? If those answers come back yes, today’s doubts will look temporary. If not, the problem will not be that AI was overhyped as a technology. It will be that the economics were less mature than the buildout assumed.

The bottom line

The AI boom may have a business-model problem because spending scaled faster than monetization certainty did.

Big Tech is still spending at historic levels, and leading AI companies are still growing extremely fast. But the industry is also experimenting with ads, private-equity distribution, metered enterprise billing, and other monetization paths while cash burn remains enormous and funding needs keep rising. That is not a sign the boom is fake. It is a sign the business model is still being negotiated under the pressure of real capital markets.

For investors, that is the key shift. The question is no longer just who can build the most AI. It is who can turn that buildout into a durable, high-return business before the bill gets even bigger.