November 11, 2026, (Inside AI) — The artificial intelligence boom has a pricing problem. As Anthropic prepares to become the first frontier AI lab to go public, the industry faces a fundamental question: nobody really knows what AI costs or what it should cost. This uncertainty, amplified by the recent US midterms, threatens to reshape the economics of the entire sector.
The midterm elections delivered a blow to the AI data center buildout. Democrats won governorships in several states on anti-data center platforms, and they are now discussing moratoriums on new development. This could constrain the supply of compute, the essential fuel of the AI revolution, and drive up costs for companies like OpenAI and Anthropic that rely on massive computing power to train and run their models.
“It's going to be more expensive to acquire that compute,” said Mike Dolan, Reuters Editor-at-Large for Finance & Markets, on a recent podcast. “The frontier labs need compute to develop their models to an even higher rate so they're gonna be able to charge even more for them and raise revenue.”
The cost of compute is already staggering. Hyperscalers, the companies scaling up data centers, are spending close to $1 trillion this year alone. That figure is projected to exceed $1.2 trillion in 2027 and reach $5 trillion to $7 trillion cumulatively by 2030. By 2032, some estimates put total spending at $10 trillion. These are the bricks and mortar of the AI revolution: chips, servers, power plants, and network equipment.
But pricing this compute is opaque. Deals are multi-year and often confidential. The cost is typically measured in GPU per hour, and prices for the most advanced chips are rising sharply, a sign that demand outpaces supply. Yet there is no standardized market price. “It's very opaque, and it's not opaque because anyone's trying to throw a shroud over it,” Dolan said. “It's simply because it's only emerging as a thing.”
The Token Paradox: Falling Prices, Rising Usage
On the demand side, the picture is equally complex. The cost of tokens, the units of AI computation that customers pay for, is falling. This is partly because many processes have become commoditized and partly because AI companies are discounting heavily to attract users. “The token cost has come down partly because a lot of the processes have become slightly commoditized,” Dolan explained. “But how the AI companies would see it is that the amount of tokens being used is rising hugely.”
Anthropic's revenue has grown 14-fold in a year, according to Dolan, even as its expenses rise faster. The company's biggest cost is compute. The falling token price is not necessarily bad for revenue if usage grows fast enough. This is the Jevons paradox: as technology becomes more efficient, overall consumption increases. But the risk is that if token prices fall too far, the companies may not recoup their massive investments.
“If token prices are falling, that is not necessarily a bad thing for revenue,” Dolan said. “As you make the models more sophisticated, those models use more and more tokens.” Complex tasks, such as autonomous processes in banking, require models to check and recheck their work, consuming vast numbers of tokens. To make these models affordable, token prices must come down. But if the cost of compute does not allow for that, a crunch could ensue.
The lack of transparency in token pricing is a concern. “When we say token price is falling, that presupposes we have an overall price for tokens full stop. And there really isn't,” Dolan said. The Federal Reserve is paying attention. Kevin Warsh, chair of the Fed, has highlighted the need to understand token pricing and its impact on productivity and inflation. He has called for standardization.
Anthropic's impending IPO is expected to bring more clarity. Once public, the company will face quarterly earnings pressure and investor scrutiny. “They will be under greater pressure to make it clear exactly what they're charging those customers for and making it clear that they can see a path from what they are spending to ultimately raking it back,” Dolan said. The payoff, however, remains uncertain. To justify a $10 trillion investment over five to six years, the industry would need to generate $3 trillion to $4 trillion in revenue collectively, according to some estimates. That requires massive adoption across the economy.
Circular Financing and the Concentration Risk
Complicating matters is the circular financing that has emerged. AI labs like Anthropic have secured billions in investment from the same companies that supply their compute, such as Amazon and Alphabet. Nvidia, the dominant chip maker, holds stakes in its customers. This vendor financing keeps the ecosystem afloat but raises concerns about sustainability. “Unless the demand for AI starts to move out into the broader economy, into the rest of the enterprises across the United States, then this world is still a very small part and it's very very intertwined part that is different and separate to the broader economy at large,” Dolan warned.
The broader economy has yet to feel the productivity gains. Torsten Slok, chief economist at Apollo, has noted that margin increases and productivity gains are concentrated in the Magnificent Seven tech giants, not the other 493 companies in the S&P 500. “AI will probably not be successful if you think about it in those terms in terms of revenue earner over time... until those other companies are reaping the benefits from it and using it extensively as well,” Dolan said.
The financial markets are beginning to address the pricing void. The Shanghai Exchange plans to launch a futures market based on AI tokens, and US exchanges are exploring futures for GPU processing. “Financial markets fulfill a function... to give some exchange to allow people to set an accurate price that's efficient,” Dolan said. These instruments could provide transparency and allow companies to hedge against volatile AI costs, much like airlines hedge fuel costs.
As Anthropic's IPO approaches, the tech world will be watching closely. The valuation of the company has already swung wildly, from $2 trillion to just above $1 trillion in a matter of weeks. This volatility underscores the uncertainty. “If you don't know from one end of the month to the next, whether it's going to be worth 2 trillion or 1 trillion, that is all I'm saying is there's an extraordinary degree of uncertainty within that world about what AI is worth ultimately,” Dolan said. The answer will shape not only the future of AI but also the trajectory of the global economy.
Shamil has a low tolerance for AI that only works in controlled demos. He covers agentic systems and generative AI with the skepticism of someone who has seen too many “revolutionary” tools quietly disappear six months after launch.
