AI investments are out of line with future profits, creating bubble that’s likely to pop
Editor:
It is important to put the proposed data center at the STAMP site in Genesee County into the broader context of artificial intelligence (AI).
AI technology in the US is experiencing a huge economic bubble, with investments far out of line with likely future profits. That financial bubble is likely to pop, thereby devastating the current AI landscape and dramatically reducing the need for new data centers. This vulnerability of AI has been overlooked in the irrational rush to invest in the technology.
AI depends on data centers. They hold the expensive, power-guzzling hardware on which the AI software runs. They are the physical brains of AI.
Very legitimate arguments against data centers like the one proposed at STAMP cite inefficient land use, excessive power use, large tax breaks for the ultra-rich, few jobs created, noise, infrasound, heat, air and water pollution, risk to wildlife, and quality of life effects on local residents.
Decent arguments all. However, to these we must add another: Data centers and the companies that build them are part of a looming AI crisis that could leave locals with a huge mess to clean up. We need to consider the economics of AI.
Consider the two “rock star” AI companies, OpenAI and Anthropic. Both are developing cutting edge LLM’s (large language models), both are privately held and lavishly financed, both want to go public via IPO’s, and both predict that an overwhelming demand for AI computations will ultimately flow through their systems.
Both put out incredible hype about the capabilities of their systems, how they will fundamentally change the world. Both downplay the obvious and well-documented risks of their systems (hallucinations, urging users to commit suicide, the recent major security breach, and many others). They also downplay the fact that for the foreseeable future, they are unlikely to generate any profits at all.
Right now they are losing large amounts of money by selling computations (or “compute”) for far less than their actual cost (Estimates: $13 of compute sold for $1 to $3). Customers have chafed at attempts to raise prices. Moreover, AI, unlike so many modern technologies, does not “scale.” Making it bigger does not mean that the cost per customer will go down.
When OpenAI and Anthropic go public the stockholders will demand profitable operations. There is no evidence that either company can deliver the promised value to their customers and still make a profit for their shareholders. This could pop the AI bubble, send the country into a recession, and leave the skeletons of half-built data centers across the USA.
So when you go over the list of reasons why the proposed STAMP data center doesn’t make sense, don’t forget to add the uncertainty of the future of AI, and how it may affect data center construction. The AI industry appears to be far less robust than its leaders would have us believe. Don’t be taken in by their hype.
Dave Giacherio
Kent



































