OpenAI is Lehman Brothers
A crash is coming.
In the 2007 subprime mortgage crisis, investment bank Lehman Brothers found itself heavily over-leveraged in billions of dollars of abominable mortgages. But behind the story was a gruesome tale of the herd mentality of the markets. Despite their underlying problems, securities backed by risky mortgages were given high credit ratings because the assumption was that the remarkable housing boom would continue unabated — despite research suggesting the market was leveling out.
Some in the media refused to accept what was happening. In a 2005 Wall Street Journal article, hedge-fund manager turned journalist Neil Barksy wrote, “The reality is this: There is no housing bubble in this country.” The next year, David Leonhardt of The New York Times suggested that the bubble bursting would be a good thing as it would lower housing prices. In a different article, he wrote that “homes seem to be much less vulnerable to crashes than other assets, because people rarely sell them in a panic.”
Lehman, once the fourth-largest investment bank in the US, was a media darling. In June 2005, The Times said “the party wasn’t over” for the firm thanks to “a strong franchise securitizing mortgages, essentially bundling them and selling them off in parts so that each individual holds less risk.” NBC News quoted a fund manager in December 2006 saying that Lehman “did a great, great job in a difficult environment.” Nothing could kill the firm, right up until it did.
By 2006, a fifth of mortgages were subprime. From July 2006 to January 2009, the national median house price dropped by 29%.
Fast forward to today. Last month, global venture funding reached $28 billion, with more than half of it going into companies in the artificial-intelligence sector. A September CNBC story warned that the flood of AI deals was distorting the VC market. Sound familiar?
As I write, the entire tech industry is being held up by investment and proliferation of generative AI. And with the media in its thrall — especially to OpenAI — this technology is being hailed as the path to an artificial general intelligence that has consciousness and the intelligence of a human, despite that there is no proof this is possible.
Yet the biggest problems are far more obvious: generative AI has no killer apps, requires endless training data to improve its outputs, has serious problems with “hallucinations” where it authoritatively states incorrect information, and every single company running models is burning billions of dollars.
To make matters worse, those building generative-AI models are finding diminishing returns in training them, partly because of a lack of data and partly because of obvious limitations of a model that guesses based on training data rather than knowing things, which skeptics like Gary Marcus have been warning about for years. Open AI hasn’t innovated anything, but members of the media buy the bullshit of the markets and marvel at how revolutionary it is to fall in love with a bot.
The real subprime AI crisis sits beneath this story. OpenAI and other model providers are burning billions of dollars to “scale” by charging unprofitable rates for model access. What happens when these companies need to charge what it actually costs to run them?
The answer is an apocalypse of sorts. Once OpenAI and other AI firms like Anthropic start running out of money, they’re going to have to raise their prices — the latter has already done so — which will eventually make any generative-AI integration unsustainable. This will in turn kill many startups that rely on these models, and actively scare away venture-capital dollars from the industry.
Perhaps this is all a little unfair on Lehman Brothers, which, unlike OpenAI, was remarkably profitable, and while its business model required it to funnel liquidity into unstable, illiquid assets, it actually made money.
Eventually the hyperscalers, who’ve pumped over $200 billion into generative-AI infrastructure, will realize that there isn’t a future in the technology. Failure will lead to a brutal haircut for the entire tech industry — and a continued dissent from consumers burned by Big Tech’s empty promise of sexy autocomplete being the future.
Read the other arguments for OpenAI's future here.
Ed Zitron is the CEO of national Media Relations and Public Relations company EZPR and the author of the newsletter “Where’s Your Ed At.”
