AI Money Squeeze: What You Need to Know

Earlier this month, millions of OpenClaw users woke up to a stark reality: the viral AI agent tool, which had taken the global tech industry by storm this year, was now severely restricted by Anthropic. The company, like other leading AI labs, faced immense pressure to ease the strain on its systems and start turning a profit. If users wanted Claude AI to power their popular agents, they would have to pay handsomely for the privilege.

“Our subscriptions weren’t built for the usage patterns of these third-party tools,” wrote Boris Cherny, head of Claude Code, on X. “We want to be intentional in managing our growth to continue to serve our customers sustainably long-term. This change is a step toward that.”

The announcement signaled a broader shift. Investors have poured hundreds of billions of dollars into companies like OpenAI and Anthropic to scale up and build out compute power. Now, they expect returns. After years of offering cheap or free access to advanced AI systems, the bill is coming due, and downstream users are feeling the pinch.

Over the past few years, top AI labs have introduced new subscription tiers to court power users. OpenAI and Anthropic revamped their pricing for enterprise clients. OpenAI rolled out in-platform advertisements. Anthropic restricted third-party tools. This pattern echoes the tech boom of the 2010s, when venture capitalists subsidized rapid growth in ride-hailing, e-commerce, and delivery services. Once companies cemented their dominance, they raised prices, added revenue streams, and delivered returns to investors. Or they didn’t, and they crashed.

But AI companies have burned through investor money faster than any other sector in recent history. They’ve broken ground on data centers worldwide, dedicating billions to promises of better models, lower costs, and AI for everyone. Even stemming losses is difficult, let alone generating the kind of profits investors expect. “When you sink trillions of dollars into data centers, you’re going to expect a return,” said Will Sommer, a senior director analyst at Gartner who specializes in economic forecasting and quantitative modeling.

“Is the era of basically free or close-to-free AI kind of coming to an end here?” asked Mark Riedl, a professor in the Georgia Tech School of Interactive Computing. “It’s too soon to say for certain, but there are some signs.”

Sommer studies long-term economic trends related to generative AI, including calculating the money at stake. Between 2024 and 2029, Gartner estimates that capital investment in AI data centers will reach about $6.3 trillion, a “massive amount of money.” To avoid write-downs, major AI model providers would ideally need a return on invested capital (ROIC) of about 25 percent, similar to what Amazon, Microsoft, and Google earn on their overall capital investments. If returns fall below 12 percent, institutional capital loses interest. Below 7 percent, you enter write-down territory, which Sommer calls “an unmitigated disaster for all of the investors in this technology.”

To reach that 7 percent minimum, Gartner forecasts that large AI companies would need to earn cumulatively close to $7 trillion in AI-driven revenue through 2029, roughly $2 trillion per year by the end of the period. For “historic returns,” they’d need nearly $8.2 trillion. OpenAI has already made $600 billion in spending commitments through 2030, a “massive step down” from the $1.4 trillion it had planned earlier. Based on OpenAI’s revenue forecasts and potential compound annual growth, Sommer predicts that even in the best-case scenario, the lab would only hit a fraction of the required spend.

How do model providers make this money? By selling access to tokens, units of data input that AI models process. One token equals about four characters in English. A paragraph is roughly 100 tokens, and a 1,500-word essay about 2,050 tokens, per OpenAI. To meet investor expectations, providers would need to process a “mind-bending” number of tokens, Sommer said.

Google announced it was processing 1.3 quadrillion tokens in October. Adding all providers’ estimates gives 100 to 200 quadrillion tokens per year. But to achieve the $2 trillion annual spend Gartner calculated, providers would need to generate a cumulative 10 sextillion tokens per year. Even with a generous 10 percent profit margin per token, token consumption would need to grow by 50,000 to 100,000 times by 2030.

Currently, companies aren’t capable of processing this many tokens, and they’re likely taking a loss on them. Sommer estimates that direct infrastructure and electricity costs yield reasonable margins, but those margins tighten or disappear with newer, token-hungry models. Indirect costs like building compute and training the next big model eat up any profit. “As soon as you then add all of the infrastructure that needs to be built for the next generation of model, and you look at how these models are going to scale, it becomes increasingly untenable,” Sommer said.

He predicts many companies “won’t be able to sustain their burn rate,” and market consolidation is inevitable. In his view, no more than two large language model providers in any regional market will survive. The era of generous free tiers probably won’t last. “For the [labs] that have a lot of users that were free, I think the question was never really if you’d monetize the free tier but it was when, and how badly do you do it,” said Jay Madheswaran, cofounder of legal AI startup Eve, a client of both OpenAI and Anthropic.

Even with a viable business model, building customer loyalty is complicated. Top labs constantly leapfrog each other on model debuts, feature releases, strategy shifts, and hiring. Engineers and developers often switch models daily, making it easy to jump ship. So labs emphasize locking users into their platforms. Anthropic, focused on enterprise clients, has gone all in on coding efforts. OpenAI has pledged to mirror that focus, with both companies reportedly racing to IPO by the end of 2026.

For now, that competition benefits end users. “It’s an arms race where you cannot let up at all because the switching cost is zero,” said Soham Mazumdar, cofounder and CEO of Wisdom AI. “As a common man, I’m going to be the winner longer-term.”

In AI’s early days, most compute costs went to training models, while inference was cheaper. But as models have advanced, inference has become far more resource-intensive. AI agents, which complete complex tasks autonomously, now use vastly more tokens than basic chatbots did a few years ago. Reasoning models, which power many agents, are notoriously expensive on the inference side, said Riedl. They “think through” multiple paths, launch sub-agents, and verify accuracy, consuming thousands of tokens behind the scenes.

“You put in your one-sentence prompt… and it’ll talk out loud to itself for thousands and thousands of tokens, thousands and thousands of words, maybe even tens of thousands when you get into coding,” Riedl said. “If you have thousands or millions of people using these things every single day, the inference costs of just the users generating tons and tons of tokens all the time really outweighs the training side of things.” If providers made a profit on those tokens, it wouldn’t be a problem, but as things stand, it’s a strain.

“The use cases have exploded, and we’re out of capacity,” said Aaron Levie, CEO of Box. Top AI labs have recently changed policies on API usage and third-party tools, like Anthropic banning OpenClaw unless subscribers pay extra. “You’ve got these tools that are basically just sitting as background processors on everyone’s laptops and desktops, just continuously waking themselves up, generating some tokens, doing some stuff, and putting themselves back to sleep,” Riedl said.

No matter what you’re doing with a reasoning-model-powered agent, there are likely wasted tokens. In an era where labs lose money on some tokens and are strapped for compute, the industry is trying to reduce waste and build more focused models. Although it’s good for customers and labs to use fewer tokens, it works against the goal of massively increasing token usage. As Sommer puts it, there’s a “narrow space on the treadmill” between short- and long-term goals.

Big AI companies are at a transition point: they’ve attracted huge user bases with free access, and now they need to keep those users while charging more. “On one hand, they want to see more tokens being generated but they have to either suck up the costs, which they can sort of do as long as venture capital is flowing, or pass the costs back on to [customers],” Riedl said. “Maybe the economics are a little upside down right now.”

These days, OpenAI and Anthropic weigh flat-rate subscription plans against metered fees. Their enterprise plans are now token-based, since usership is “uneven,” as Andrew Filev, founder of Zencoder, noted. Some users log in occasionally, while others run agents around the clock. In consumer chatbots, some model makers are turning to advertising. OpenAI recently introduced ads within ChatGPT, and it’s reportedly working on a tool to track their effectiveness. Anthropic decried the move in its 2026 Super Bowl ads.

For companies building tools on top of models like GPT-5 or Claude Opus, token prices are rising, and the extra cost trickles down to customers. Multiple tech companies told The Verge that they or their customers are changing strategies to offset new pricing. Some are moving to open-source models, while others spend resources evaluating how expensive models perform compared to cheaper alternatives.

David DeSanto, CEO of software company Anaconda, recently returned from a five-week trip speaking to customers. Many were moving to self-host AI models on Amazon Bedrock or Google’s Vertex AI for more control, or switching to open-source or open-weight models, which have improved significantly. Some companies worry about security when sending IP to commercial labs, so they only use ChatGPT or Claude for “mission-critical applications,” he said.

“Everyone I spoke to had some version of this problem , their token usage has gone up, so their usage-based billing cost has gone up, or the tier they were on no longer has the same cap, and now they’re having to go to a more expensive tier to try to keep the same amount of usage per month as part of their flat rate,” DeSanto said.

Eve constantly balances quality and token costs, Madheswaran said. The company’s token usage has gone up 100x year-over-year, so it switches between open-source models and ones from Anthropic and OpenAI. But even a 1 percent regression in output quality negatively impacts customers “quite significantly,” which is why Eve spends resources tracking model quality. It uses newer reasoning models about 25 to 30 percent of the time, splitting the rest between its own open-source variants and cheaper models from leading labs. Madheswaran said some cheap models are just as accurate as expensive ones, depending on the query.

“What open source is really doing is it’s putting pressure on these companies to make their cheaper models cheaper because their profit margins there are much, much better,” Madheswaran said.

Wisdom AI hasn’t had to pass on cost increases yet. The team tests how different models perform on various tasks and budgets accordingly. Mazumdar said it’s been testing Cerebras, popular for open-weight models, “in anticipation of how expensive things will get” from premier labs. “[Big AI companies] have been giving this away for free,” he said. “What they’re trying to do is, the moment they sense there’s an enterprise at play, or there’s propensity to pay, they absolutely jack up the prices drastically.” But he noted that for coding, open-source models don’t come close.

Box’s Levie believes the changes will play out over the next 24 months. The VC-subsidized era was likely necessary for growth, but now it’s time to build efficiency. “The size of the market is so large that I think it actually will sort of all work out,” Levie said. “At an individual company level, you have to decide: Can you keep up with this flywheel, or are you going to be priced out based on an inability to raise capital or an inability to make the model more efficient for your tasks?”

Eve’s Madheswaran thinks the industry will shift from focusing on the “best” model to what works for niche use cases. “That’s my guess, and obviously I’m betting our entire company on it.”

Sommer likens the scenario to the “stegosaurus paradox.” When scientists first discovered the stegosaurus fossil, they didn’t understand how a large body could be supported by a small head with a tiny mouth. The theory was that it needed to constantly eat a highly nutritious diet. “We see AI as kind of being the same deal,” Sommer said. For AI labs to survive, providers need to find more food for it (the entire global economy) and it has to be highly nutritious (earning a margin rather than subsidizing). If the paradox isn’t resolved, it will lead to write-downs, falling valuations, dried-up financing, and a broad resetting of expectations. A sustainable business model “would require that genAI be infused in everything from billboards to checkout kiosks,” with providers taking a cut of transactions.

“The free era was really a land grab , it’s a common strategy used by startups,” said Eve’s Madheswaran. “That’s just not a business model. You can’t do that for too long.”