(Bloomberg Opinion) — Call it the nerd mindset. Silicon Valley has spent the last few years believing that whoever built the smartest artificial-intelligence system would rule the industry. Today, that calculation is starting to look misguided.
The cost of building AI models is going up, and the price of usingthemis going down. AI companies’ revenuegrowthhas been impressive, but it’s also slowing. That doesn’t necessarily meanabursting bubble, but it does look like the realpower increasingly lies with the companies that distribute and power the technology: cloud giants like Microsoft Corp., Alphabet Inc. and Amazon.com Inc.
This trio of the Big Tech establishment, whose shares have risen an average of 10% since the start of 2026,controls the infrastructure, customer relationships and distribution channels through which AI is sold. They can profit whethera business uses Claude, or an older version of ChatGPT, or a cheaper open-weight model from China. That’s not the case for OpenAI or Anthropic.
The latteris on the runway to an initial public offering in November; it’s aiming for a $2 trillion valuation in what will be the biggest testof continued investor appetite for the AI boom. The problem: Anthropicrecently told investors that its third-quarter revenuewas$16.8 billion, a quarter-on-quarter growth rate of around 50%, compared with 140% earlier this year.
Go on X and you’ll find hysteriaover this news. A popular financial commentercalls Anthropic’s numbers a “crazy sharp” deceleration. “It’s over,” they add, ominously.
Actually, it’s hardly a disaster. At 50% quarterly growth, sales would increase roughly five times in a year and 130 times in three years. Thehard part issustaining it, as each additional quarter requires billions more in new revenue.
Compounding the problem is the fact that the most powerfulfrontier models cost a fortune to build. A chart published by Citadel Securities in August highlighted the painful economic trapfaced by OpenAI and Anthropic: The computing power needed to run AIis becoming more expensive, whileso-called token prices have fallen, making it much cheaper to use.
The authors of the Citadel chart were optimistic, saying that cheaperAIwas leading companies to use the technologymore, enough to sustain demand for all that expensive, underlying compute — AI boosters often point to this phenomenon, known as Jevons paradox.
But it gets harder to be optimistic about Anthropic and OpenAI sustaining thatpace of growth when more businessesare opting for AI that is just good enough,rather thanthe best. Cheaper, open-weight models now account for morethan a third of enterprise AI token usage, up from 23% a year ago, according to Enterprise Technology Research, which surveyed 200 companies in September.Respondents expected that share to reach 41% inthe next 12 months.
And over the past year, two majorAnthropic customers—Microsoft and Meta Platforms Inc. —have cut back on their internal use of Claude in favor of their own proprietary solutions in an attempt tobring down astronomical token costs.
That all makes the future look more hazy for the leading AI companies, particularly in the face of a yawning gap between the money they’re bringing in, and the collective $1.3 trillion that OpenAI and Anthropic are expected to spend on data centers, chips and computing power through the end of 2030.
Does that spell a catastrophic sequence of falling dominoes for the rest ofSilicon Valley? Some sort of correction seems inevitable when investors have committed such enormous sums to AI infrastructure and the debt financing it — which is already having second-order effects on the bond market — before anyone can reliably calculate the returns.
And if further growth struggles at Anthropic and OpenAI leadto a collapse in confidence, the cloud companies won’t be entirely immune either. They’ve invested billions of dollars in frontier AI firms and data centers to run their services. The use of cheapermodels will also squeeze their margins.
But when it comes to weathering a coming storm, the establishedcloudgiants are also far better placed than the newcomers. They have profitable businesses outside AI, longstanding relationships with corporate customers and the ability to make money whichever model comes out on top.
Recall that in the dot-com era, many investors lost their shirts because companies couldn’t quickly monetize expensive fiber-optic cables that had beeninstalled to supportthe boom. That infrastructure just took longer than many expected to beutilized.
If the AIbull run follows a similar path, the biggest winners may be the companies with the most resilience —notthe Sam Altmans or Dario Amodeis trying to build a machine god, but the ones who can afford to wait for it.More from Bloomberg Opinion:
–With assistance from Mark Gilbert .
This column reflects the personal views of the author and does not necessarily reflect the opinion of the editorial board or Bloomberg LP and its owners.
Parmy Olson is a Bloomberg Opinion columnist covering technology. A former reporter for the Wall Street Journal and Forbes, she is author of “Supremacy: AI, ChatGPT and the Race That Will Change the World.”
More stories like this are available on bloomberg.com/opinion
