Steve Eisman, the American investor who became a household name by shorting the subprime mortgage market during the 2008-2009 financial crisis, has a pointed view on artificial intelligence. In his reading, the AI boom carries one glaring vulnerability that markets are not fully pricing in.
OpenAI and Anthropic, the two dominant frontier labs, are each burning billions on compute. Both companies spent much of last summer in a breathless race toward public markets to secure fresh capital. Their spending patterns might suggest no cheaper alternative can emerge, but that assumption deserves scrutiny. What happens if lower-cost competing models gain traction? Both giants would face pressure to cut prices, threatening the revenue growth they need to keep scaling.
Eisman Makes His Case on CNBC
In an interview with CNBC, Eisman argued that chip manufacturers and cloud providers are dangerously overexposed to the financial fortunes of the AI giants. His view, stated plainly:
“The entire advanced electronics supply chain, at this point, depends on the health of Anthropic and OpenAI.”
If the two frontier labs were to run into serious funding constraints, the downstream consequences for chipmakers, data center operators, and infrastructure providers could be severe and wide-ranging.
Not everyone shares Eisman's alarm. Morningstar, for one, reads the situation very differently, and its analysis offers a substantive counter-argument worth considering.
Morningstar: Scale as a Competitive Moat
Morningstar analysts contend that the sheer scale of the two frontier labs could protect them over the medium and long term, and by extension, shelter the electronics supply chain that depends on them. By purchasing compute at volumes no smaller competitor can match, OpenAI and Anthropic have locked in pricing advantages that compound over time. According to Morningstar's analysis, this scale translates into an estimated annual cost advantage of $57 billion on technology spending.
A recent sector finance analysis led by Morningstar experts estimated that Anthropic and OpenAI hold multi-year preferential agreements with Nvidia. Under those terms, both labs can purchase DGX B200 servers (the hardware foundation for frontier AI model development) at $2.25 per GPU-hour. The standard list price for the same ecosystem sits at $16.11 per GPU-hour, per the same Morningstar analysis.
Put those numbers together and one gigawatt of compute capacity costs the two labs an estimated $9 billion annually, according to Morningstar's figures. A smaller rival attempting to compete on the same terrain, without the leverage of multi-year volume commitments, could pay as much as $66 billion for equivalent compute with no discounts applied. The gap is not incremental. It's structural.
“The large labs are, today, the only buyers capable of committing to the duration and scale needed to justify such price reductions.”
Morningstar also flagged a second compounding advantage: the larger customer base that OpenAI and Anthropic command generates a higher volume of feedback, which accelerates product improvement in ways that leaner rivals simply can't replicate at the same speed.
Cheaper Alternatives Are Gaining Ground
Market data tells a more nuanced story. According to a production index published by the Vercel platform, enterprise customers are spreading AI spending across a growing roster of models that compete with Claude Duet and OpenAI's frontier offerings (including models capable of operating without an internet connection). The overall AI market is expanding fast. Per Vercel's data, spend on the most advanced models grew tenfold in roughly ten months since last December, and lower-cost alternatives have been gaining share consistently.
No credible challenger has yet threatened the leadership of the two largest labs in absolute terms. But a growing ecosystem of smaller AI companies is steadily accumulating users, and if that trajectory holds, Anthropic and OpenAI will face meaningful market-share erosion over time.
Once competitive conditions level out more than they are today, the question Eisman is already asking becomes unavoidable: for how long can the AI giants freely set their own pricing and protect their margins? The answer matters not just for the two labs but for every chipmaker, cloud provider, and infrastructure investor that has tied its growth forecast to their continued dominance. Nvidia's DGX pricing agreements, Morningstar's $57 billion cost-advantage estimate, and Vercel's market-share data are the three signals worth tracking closely in the months ahead.
