When Anthropic filed a confidential S-1 with the Securities and Exchange Commission on June 1, the company's founders were understandably focused on the headline numbers: a $65 billion funding round, a $965 billion post-money valuation, and an annualized revenue run rate of $47 billion, roughly five times what the company reported twelve months earlier. For the people who built Claude, those figures told a story of exceptional growth. For Michael Burry, they told a different one.
Burry, whose fund Scion Asset Management made a fortune shorting mortgage-backed securities before the 2008 financial crisis, has been issuing warnings for weeks about the current wave of AI and space technology IPOs. His critique is not aimed at the underlying technology. It is aimed at the valuation logic, and specifically at what he sees as the market's failure to account for what happens when computing power stops being a competitive advantage and becomes a utility.
"There is no guarantee," Burry wrote to his subscribers in May, "and not even a strong likelihood, that Anthropic is long-term worth anywhere near $1 trillion." He expanded on that view following the IPO filing, calling frontier AI development "far too expensive, too much brute force," and predicting that compute "will be commoditized, like internet use."
The Specific Objection
Burry's argument has two components. The first is about the structural economics of building large language models. Training and running a frontier model requires continuous capital investment in GPU clusters, data center capacity, and the engineering talent to operate them. Anthropic's deal with SpaceX to lease the entire Colossus 1 data center in Memphis, at a reported $1.25 billion per month, illustrates the scale of that commitment. The company is spending at a pace that requires the revenue trajectory to continue upward without interruption.
What Burry questions is whether that spending produces a durable competitive moat. Internet bandwidth, he notes, was once a scarce resource that commanded premium margins. Today it is priced near zero and treated as a cost of doing business. The history of technology infrastructure, from long-distance telephone to cloud storage, tends toward the same outcome: strategic differentiator to commodity. This is not a new observation, but it carries more weight when applied to a company spending billions on GPU clusters every quarter. The competitive moat in AI, Burry suggests, is narrower than the current valuation implies.
Key Numbers at Time of IPO Filing
- Anthropic valuation$965 billion (post-Series H)
- Revenue run rate$47 billion annualized
- Series H raise$65 billion
- Business customers300,000+
- AI IPO capital (2026 wave, inflation-adj.)Comparable to ~300 dot-com IPOs in 2000
- Burry AI short positionExpanded in May 2026
The Dot-Com Parallel
The second component of Burry's case is historical. He calculated that the combined capital likely to be raised through the upcoming IPOs of SpaceX, Anthropic, and OpenAI, adjusted for inflation, is comparable to the aggregate proceeds from roughly 300 internet, media, and telecom company listings in the year 2000. That cohort was not composed of frauds. Amazon was in it. eBay had already gone public. Companies that, in the long run, proved foundational to the modern economy. And yet anyone who bought the cohort at 2000 valuations waited more than a decade to see their capital return at a profit.
Burry's parallel is intentionally imprecise. He is not predicting that Anthropic will fail, or even that its stock will collapse on listing. He is making a probabilistic point about what happens when an asset class attracts the volume of capital that AI has attracted: the most obvious winners get priced for perfection, and perfection is an unreliable assumption.
"Great companies can still become dangerous stocks if valuations assume perfection." Michael Burry, Scion Asset Management, May 2026
The Bull Case and What the Market Will Decide
Anthropic's defenders have a ready response. The $47 billion annualized run rate is not a projection; it is a reported figure. The company has over 300,000 business customers, and the enterprise adoption curve continued steepening even as the team grew past 4,000 employees. Anthropic has said it expects to report its first operating profit in the June 2026 quarter, which would put it ahead of both OpenAI and SpaceX on that measure at the time of their respective listings. The argument on the other side of Burry's trade is that Anthropic's model quality, its safety reputation, and its network of enterprise relationships form exactly the kind of durable advantage that compute commoditization alone cannot erode.
What the S-1, once made public, will reveal is the cost structure behind the revenue: the gross margin on Claude subscriptions, the trajectory of infrastructure spend relative to revenue growth, and the company's own guidance on what a path to sustained profitability looks like. Those numbers will either validate or complicate the bear case. Anthropic's $65 billion fundraising round already priced in substantial optimism. The IPO filing now asks the public markets to do the same.
Burry expanded his AI short position in May, which means he has money on his thesis rather than just words. The companies he is shorting have not been named publicly. Whether Anthropic is among them is not known. What is clear is that when the S-1 goes effective and trading opens, buyers and sellers will be arguing about precisely the question Burry has spent the last month trying to bring to the surface: whether $47 billion in annualized revenue growing at fivefold annual rates justifies a valuation that assumes the compute moat holds.
