October 19, 2025 - 1:00pm
This week, a narrative has been building that AI is in a bubble. Huge amounts of money have poured into the industry, but it is yet to deliver the long-awaited productivity boom everyone speaks of.
In February of this year, the company materialised from nothing into a $12 billion valuation, on the back of a $2 billion capital raise, led by arch tech investors Andreesen-Horowitz. By some measures, this is the biggest initial seed round in history. What does Thinking Machines do? AI. More specifically? Well, no one quite knows.
Another AI star, Ilya Sustkever, has founded Safe Superintelligence. It aims to “build safe superintelligence, and nothing else”. On the basis of this terse prospectus, his company now has investments totalling $2 billion, against a $32 billion valuation, and zero products.
Yet within these black boxes lie the hopes of AI’s capital markets. Ten loss-making artificial intelligence start-ups have gained close to a trillion dollars in valuation over the past 12 months. Bank of America Global Research found that 54% of fund managers believe AI stocks are already in bubble territory, while 38% disagree. Jamie Dimon of JPMorgan Chase says he’s “far more worried” than others about a big correction.
The sums involved are staggering. AI infrastructure is so expensive that it requires companies to collectively fund a new Apollo Programme, not every 10 years, but every 10 months. Annual AI-related data-centre spending in 2025 is around $400 billion, but AI revenue in total still only sums to $60 billion.
A recent list in the Financial Times described the biggest AI firms of 2025 and their market capitalisations. It starts with well-known companies who sell real products with strong revenue models. By one metric, both Open AI and xAI are doubling their revenue year-on-year. Then there are the “selling-shovels-in-a-goldrush” firms, like Databricks, which does cloud stuff-plus-AI. Then there is corporate tech, like Sierra, a customer service chatbot specialist and Mercor, a recruiting app. There are pleasant but deeply inessential personal assistants like Claude and Cursor. And then there are the esoteric giants like Thinking Machines and Safe Superintelligence.
Taken together, it certainly feels as though there is a disconnect — between things that could make money, and the already sizeable impact on our everyday life.
The problem is that there are plenty of applications for AI. It’s just a question of whether they save more than they cost. For instance, Walmart says that its use of AI inventory bots has cut excess stock by 35%, and boosted accuracy by 15%. That’s a more pleasant experience for customers, and a vague tingle of savings for Walmart. But it is not the same as opening a new oil pipeline, or selling a car.
In the end, small constant wins that every company can access, in a competitive market, tend to leave everyone at the same competitive level that they were: the surplus value becomes social, it accrues to consumers, not producers.
We are in the strange position of being in a bubble and everyone knows it. If everyone is predicting downfall, surely that outcome must be priced in? If Thinking Machines ultimately makes nothing, its investors will only be on the hook for whatever portion of its capital is spent — and it could have spent very little.
Somewhere in all of that, fortunes will still be made; but it will be much slower than the stock price junkies can presently imagine. The banal truth is that in almost any bubble of the past half century, the boring investment would likely have been the right one. Buying more Apple, more Oracle, more Microsoft. Markets change, but the real winners often settle in early.
Join the discussion
Join like minded readers that support our journalism by becoming a paid subscriber
To join the discussion in the comments, become a paid subscriber.
Join like minded readers that support our journalism, read unlimited articles and enjoy other subscriber-only benefits.
Subscribe