A prediction market assigns a 91% probability that Anthropic will be worth $1.25 trillion by year-end. A rumored $10 billion compute lease with Meta is the catalyst. The math is absurd. The logic is worse. Yet the market moves.
The deal, if confirmed, would see Meta renting tens of thousands of GPUs to Anthropic over three years—an infrastructure commitment larger than most nation-state budgets. Polymarket, a crypto-native prediction platform, has priced in a valuation that would make Anthropic more valuable than NVIDIA was in early 2023. Let’s dissect the numbers.
The Compute Arithmetic $10 billion for compute. At current rental rates for H100-class hardware, that translates to roughly 30,000–50,000 GPUs over a three-year lease, depending on power, cooling, and markups. That’s a cluster capable of training models at the scale of GPT-4 but not beyond. The math holds: $10 billion is real money, and the hardware is real. But the humans who set the lease terms did not verify the revenue trajectory. Anthropic’s current annualized revenue is in the low single-digit billions. Even the most aggressive growth curve cannot support a $10 billion annualized compute bill without sustained external funding. This is not a business model; it is a burn rate disguised as investment.
The Valuation Mirage 1.25 trillion. Let that number settle. Compare to NVIDIA’s ~$2 trillion market cap after a decade of dominance and >50% net margins. Anthropic has no net margins—it loses money on every API call. A 91% implied probability on Polymarket means the market believes this is almost certain. Polymarket is a low-liquidity prediction market where a few whales can move odds. The 91% is not consensus; it is a signal that someone with capital wants to create the impression of inevitability. Correlation is the comfort of the unprepared. The correlation between the compute lease rumor and the Polymarket spike is not evidence of truth—it is evidence of coordinated narrative engineering.
The Strategic Logic (What Bulls Got Right) Meta’s motivation is clear: lease idle compute, lock in a strategic ally against the Microsoft-OpenAI axis, and gain preferential access to Anthropic’s models. For Anthropic, securing dedicated compute outside the hyperscaler oligopoly grants independence. The alliance is real. The compute need is real. The valuation target is not. The bulls correctly identify that AI is now a capital and compute arms race, and that Anthropic must secure capacity or die. But they conflate necessity with value. A burning platform is not worth $1.25 trillion; it is worth the cost of the fire truck.
The Footgun in the Fine Print I have audited similar infrastructure-heavy deals during my risk management days. The hidden liability is the “take-or-pay” structure. If Anthropic’s revenue does not grow fast enough, the lease payments become a fixed cost that consumes operating cash flow. The downside is asymmetric: the compute provider (Meta) is diversified; Anthropic is not. Assumptions are just risks wearing disguises. The assumption that revenue will scale to cover a $10 billion opex line is a risk disguised as a growth projection. In a bear market for AI tokens and private valuations, this risk is magnified by the lack of liquidity.
The Takeaway The Polymarket prediction will likely be wrong. The compute deal may close. But the gap between infrastructure reality and narrative valuation is so wide that only a serious re-rating or a massive revenue surprise can close it. Investors should treat the 91% as what it is: a synthetic probability generated by a low-liquidity market, not a truth oracle. Value is consensus; truth is optional. The only truth here is that $10 billion of compute hardware will be deployed. Whether it generates $1.25 trillion of enterprise value depends on whether the consensus holds when the bills come due.
Ask yourself: Who benefits from you believing the 91%? The answer reveals the provenance of the story.
