AI companies would have to achieve unprecedented profits to justify the financial investments they have received: the more likely scenario is that this bubble will burst sooner or later.
Joseph Grosso is a librarian and writer in New York City. He is the author of Emerald City: How Capital Transformed New York (Zer0 Books).
Cross-posted from Counterpunch
Browsing the Business pages in recent times one finds an interesting dichotomy. On one hand, investment in AI continues to rage, sucking in around 80 percent of venture capital in the U.S. in the first quarter of 2026. AI companies make up about 40 percent of the value of U.S. equity markets. Almost all the S&P 500’s gains in 2026 have come from AI companies. As Ruchir Sharma put it a column for the Financial Times last year: ‘America is now one big bet on AI.’
On the other hand, the Wall Street Journal reported in July that ‘‘Corporate America Has Suddenly Decided to Stop Blowing Money on AI.” From Uber, which blew through its whole 2026 AI budget in a half year, to Salesforce companies are spending less (and appear to be shifting more towards open-source models from China). According to Wired, even Meta is formally ending what amounted to a ‘tokenmaxxing’ incentive program for employees. The company told workers that their performance evaluations would no longer be dependent on how much they used AI tools (of course, it wasn’t long ago Meta flushed $83 billion down the drain on its Metaverse thing).
The job apocalypse so touted by the AI companies, as the main reason to buy their products, hasn’t yet come to pass. A Census Bureau survey about AI (the Bureau has been running it since 2023) reported that most companies had no change in overall employment. In July the Wall Street Journal reported an EY-Parthenon survey that found the percentage of CEOs expecting significant AI-driven headcount reductions fell from 46% in January 2025 to 20% by May 2026. Software engineers, who recently were on the way to extinction with the introduction of Claude and Codex (back in February, a selloff wiped out $1 trillion from software and services stocks), are rising as a percentage of total employment. There is apparently more to software than writing code. Even translator jobs are holding steady. That expected productivity boom hasn’t shown up in the data.
A new report published by Glassdoor (as described on the Blood in the Machine Substack), a webpage where workers anonymously talk and complain about their jobs, reveals growing anti-AI sentiment. The report shows jobs as diverse as insurance claim adjusters (98 percent negative), writers/journalists (81 percent), accountants (80 percent), customer service (78 percent), designers (70 percent), and tech support/IT (68 percent) giving AI overwhelming bad reviews- while perhaps predictably the more positive views come management positions. Afterall, research published by Harvard Business Review early this year show that AI intensified work, rather than reduced it, through a combination of blurring boundaries and making multitasking easier. The research showed speed expectations shifting naturally even without pressure from management. What is emerging is not AI eliminating jobs but making jobs more tedious and miserable. Karl Marx described something to like back in his first volume of Capital.
Meanwhile, the big AI companies are racing toward IPOs. Anthropic, which passed OpenAI in revenue and valuation earlier this year, mainly with Claude as OpenAI had struggled for years to find a way for its chatbots to actually earn money, is targeting sometime this fall. The Wall Street Journal reports that the IPO could raise up to $100 billion with about a $2 trillion valuation.
This would break the record for IPOs that was set by SpaceX just this past June. SpaceX has surely been a successful launch and satellite-internet company, lowering the launch price per kilogram substantially, yet it insists on being an AI company (Elon Musk merged SpaceX with xAI in February). For its IPO, SpaceX got the honor of an immediate investment-grade rating. Previous tech heavyweights from Meta to Netflix waited a decade or more after their listings to achieve a top-tier credit rating. Needless to say, OpenAI and Anthropic are seeking the same rating. SpaceX also benefited from changes to index rules that meant billions in passive investment tracking the S&P 500 and Nasdaq flowed into its stock. The company is burning cash on AI, just posting a $541 million quarterly loss (that’s a pace of well over $2 billion for the year) and to hit its market cap (the company’s IPO pitch included some harebrained plans like putting data centers in space) would have to become perhaps the most successful company ever. According to Fortune’s estimates, SpaceX would have to increase sales by a whopping 50 percent every year for a decade, something no company has even come close to accomplishing.
The outlook is similar at Anthropic and OpenAI. Ed Zitron, who has become the Cassandra of the AI craze, explains in a post titled ‘What Happens if OpenAI Dies?’ there are only two eventualities given the company’s projections and commitments: OpenAI either becomes the largest, most successful company of all time or OpenAI runs out of money at some point. OpenAI’s projections have it almost tripling its 2025 revenues, doubling its 2026 revenues, nearly doubling its 2027 revenues, growing its 2028 revenues by 68 percent, and then growing its 2029 revenues by 64 percent. Quite an awful lot would have to go right.
Anthropic has worked to position itself as the ‘good guy’ AI startup (it was formed by seven former OpenAI employees who left OpenAI over safety concerns). During the most recent Super Bowl, the company ran spots slapping OpenAI for introducing ads on ChatGPT. There was something of an ethical moment in the aftermath of the abduction of Nicholas Maduro in Venezuela this past January. When it was reported that Claude was used in the operation, allegedly through Anthropic’s partnership with Palantir, the company accused the Trump administration of breaking usage guidelines which prevents Anthropic models from developing autonomous weapons or conducting surveillance. This soon led to Trump declaring Anthropic a supply-chain risk (Anthropic sued and won a ruling in August, though it is unclear if anything has changed). Of course, this also meant that Anthropic was the first AI company approved for use on classified U.S. government networks and that it works with the likes of Palantir. Then there was Anthropic announcing in April that its new Clause Mythos Preview model was too dangerous to release to the public, though it was made available to companies like Apple, Amazon, and CrowdStrike (Really, is there a better instance of marketing?).
The high point had to be co-founder Chris Olah landing a seat at the table in the Vatican when Pope Leo XIV released his encyclical “Magnifica humanitas”, calling to preserve human dignity in the AI age. For his part, Olah’s remarks included:
The first is our duty to the global poor. There is a real possibility that AI will displace human labor at very large scale. If that happens, supporting those displaced will be a moral imperative of historic proportions…How can we ensure the gains of AI are shared globally? We do not have a mechanism for this. It is an unsolved problem, and it is the kind of problem the Church has historically refused to let the world ignore.
It was three days later that Anthropic announced a $65 billion funding round and a few months later announced that its annualized revenue also hit $65 billion. Perhaps sharing some of this through more progressive taxation and open sourcing would be a start? But actually, annualized revenue is a lousy metric, where a snapshot of a month’s income is multiplied by twelve. Anthropic still isn’t nearly profitable on a consistent basis.
When the bubble does burst it will take at least a third of the stock market with it. While it is a fact that the great majority of stocks are owned by the wealthiest people (the wealthiest 1 percent own more than the bottom 90 percent combined), plenty of other people will see their retirement accounts negatively affected. Past bubble manias, such as the 19th-century British railroad bubble at least left behind useful infrastructure. It is difficult to see what other uses there can be for all these data centers, the opposition to which has united Americans like nothing else in recent times.
Some would point to the dot-com bubble and say while that bubble popped the internet itself survived and remained a major force. One difference with AI foundation models is that there is no sign of them becoming cheaper to build. It gets more expensive every time it improves and every time it adds a user. In his latest book, The Reverse Centaur’s Guide to Life After AI, Cory Doctorow argues what can be salvaged from the residue, along with worker skills that may be put to good use, are stand-alone, open-source, ‘toy’ models that run on personal computers and devices (As for data centers, Doctorow argues that perhaps they can be good laser tag arenas).
The bottom line is it will be up to us to go forward with what is useful and ensure that the nebulous plans of billionaires never come to pass.
Be the first to comment