The numbers are staggering, almost obnoxious in their velocity. Anthropic’s annualized revenue run rate—from $14 billion in February to $47 billion in May. A private valuation leap from $380 billion to $965 billion in three months. The market now whispers about a $2 trillion IPO valuation. But for those of us who spent 2022 dissecting the Terra collapse, these numbers trigger a different kind of reflex. Not greed. Skepticism. The question isn’t whether Anthropic can grow. It’s whether the narrative around its growth is structurally sound, or just another liquidity sponge waiting to be squeezed.
Context: The AI Compute Machine
Anthropic is not a typical tech IPO. It’s a compute-hungry monster. The company secretly filed its S-1 on June 1, and since then, the details have leaked like a broken pipeline. The company has committed to spending over $100 billion on Amazon Web Services over the next decade. It has secured agreements for up to 5GW of new compute power from Amazon, another 5GW of next-generation TPU capacity from Google and Broadcom, and is tapping into SpaceX’s GPU capacity. In May, it raised $65 billion, partly to fund this compute expansion. The annualized revenue run rate is impressive, but so is the capital intensity. This is not a software company with high margins. This is a heavy-industry player dressed in AI clothing.
For context, consider the crypto mining sector. Bitmain’s dominance in ASICs, or the hash rate concentration after the 2024 halving. The similarity is uncomfortable. Anthropic is building a compute empire, but the underlying economics are eerily reminiscent of a proof-of-work network: high fixed costs, continuous reinvestment needed to stay competitive, and a race to the bottom on margins unless you control the supply chain. The difference? Anthropic doesn’t control the chips. It rents them from Amazon, Google, and Broadcom. That’s a structural vulnerability.
Core: The Narrative Mechanics of Pre-IPO Hype
Let’s deconstruct the valuation narrative. The market is pricing Anthropic as if it will capture the majority of AI profits. But history—and basic economics—suggests otherwise. In the 2020 DeFi summer, I watched liquidity providers pile into Uniswap pools, only to realize that the real value accrued to the protocol itself, not the LPs. Similarly, in the current AI stack, the value accrual is likely to concentrate at the infrastructure layer—compute providers, not model developers. The chipmakers (Nvidia, Broadcom) and cloud providers (Amazon, Google) are the equivalent of Ethereum’s base layer: they charge rent on every transaction. Anthropic is just a dApp running on top.
Based on my audit experience modeling liquidity congestion during high-volume swaps, I can see a parallel. The AI compute market is becoming congested. Anthropic needs to scale its compute capacity exponentially to maintain its model’s edge. But the marginal cost of compute is not falling as fast as the revenue growth. The data from the leaked S-1 suggests that Anthropic’s gross margin is already under pressure. In Q1 2025, the company spent 68% of revenue on compute and infrastructure. That’s worse than most crypto miners during the 2021 bull run. The narrative says “growth justifies the premium.” But growth without margin expansion is just a Ponzi schedule with better PR.
The market is currently discussing a $2 trillion IPO valuation. That would imply a price-to-sales multiple of roughly 42x based on the $47 billion run rate. Compare that to Nvidia’s 25x or Microsoft’s 12x. The premium is built on the assumption that Anthropic will maintain its technological lead and pricing power. But the competitive landscape is brutal. Open source models like Llama 3 are catching up. Google’s Gemini is no slouch. And the compute costs are not proprietary—anyone with enough capital can buy the same chips. The moat is not code; it’s capital allocation. And capital allocation is a game of diminishing returns.
Contrarian Angle: The Hidden Tax of Compute Leases
Here’s the blind spot that most analysts are missing. Anthropic’s $100 billion commitment to AWS is not just a cost—it’s a liability that locks the company into a single supplier. This is the same mistake that Terra made with UST’s peg to Luna. The correlation between cost and revenue becomes toxic when the supplier has pricing power. Amazon can raise prices on Anthropic at any time, and Anthropic has no alternative. The Google and Broadcom deals are small by comparison. The SpaceX GPU capacity is a hedge, but not a strong one. The result is that Anthropic’s gross margin is not a function of its own efficiency, but of Amazon’s goodwill.
I recall my 2023 work on EigenLayer’s restaking thesis. The core insight was that security is not a static asset; it’s a dynamic market where the buyer (the protocol) is at the mercy of the seller (the staker). The same applies here. Compute is the new security. Anthropic is leasing its security from Amazon, and the terms are not in its favor. The narrative that “AI is a winner-take-most market” ignores the fact that the winners are the infrastructure providers, not the application layer. The market is pricing Anthropic as if it’s the next Microsoft, but structurally, it’s more like a high-end SaaS company with a massive capex burden.
Furthermore, the IPO itself is a narrative event. The market is pricing in the “IPO premium” even before the deal is done. This is reminiscent of the 2021 Coinbase direct listing, where the stock popped 30% on day one, only to crash 80% a year later. The narrative of “public market validation” is a trap. The real value will be determined by cash flow, not hype. Anthropic’s cash flow statement, based on the leaked data, shows negative free cash flow of $12 billion in the first half of 2025. That’s a red flag in any market cycle.
Takeaway: The Next Narrative Shift
The Anthropic IPO is a litmus test for the entire AI narrative. If the market prices it at $2 trillion and the stock holds, it will validate the thesis that AI is the next internet. But if it stumbles, it will trigger a re-evaluation of the entire sector. The key metric to watch is not revenue growth, but compute efficiency. Specifically, the ratio of revenue to compute cost. If that ratio improves, the narrative is intact. If it deteriorates, the narrative is broken.
For crypto investors, the lesson is clear. The AI compute economy is a new frontier, but the same laws of narrative physics apply. Restaking isn’t a narrative shift in security—it’s a liquidity game. Similarly, Anthropic’s IPO is not a shift in AI value—it’s a liquidity event for early investors. The real alpha will be found in the infrastructure plays: the chipmakers, the data center operators, and the protocols that facilitate compute trading. Follow the narrative, but verify it with the math.
The question is not whether AI will grow. It’s whether the profits will flow to the model builders or the infrastructure landlords. Based on the data, I’m betting on the landlords. The IPO will be a liquidity event, not a value creation event. And the market will eventually realize that the narrative of AI’s infinite scalability is just another story waiting to be deconstructed.