The filing was secret. The numbers are not.
On June 1, Anthropic quietly submitted its S-1 to the SEC. By the time the market caught wind, the annualized revenue run rate had already jumped from $14 billion in February to $47 billion in May. That’s a 236% compound acceleration in three months. The private valuation followed: $380 billion in February, $965 billion by May. Now the chatter is about a $2 trillion IPO valuation.
Greeks don’t price this kind of beta. The options market has no framework for a company that burns cash faster than a DeFi protocol on a gas war and simultaneously demands a multiple that would make even the most aggressive growth-stock analyst blush. This is not a company — it’s a hypothesis dressed in compute power.
I’ve seen this pattern before. In 2017, I audited a token called CryptoGem that raised $2.4 million on a promise of “AI-enhanced gem trading.” The code had an integer overflow in the transfer function. I shorted it via Bitfinex’s uncollateralized lending, published the audit, and watched the token collapse. The founders collected their fees; the retail investors collected losses. The structure was the same: a narrative that outran the underlying mechanics.
Anthropic is more sophisticated, but the skeleton is familiar. The company has committed to spending over $100 billion on Amazon Web Services over the next decade. It has signed agreements for 5 gigawatts of new compute with Amazon, another 5 gigawatts of next-generation TPU capacity with Google and Broadcom, and is renting GPU capacity from SpaceX. The capital expenditure is not an investment — it’s a lease on a seat at a table that might not exist.
Context: The Infrastructure Arms Race
Let’s lay out the facts. Anthropic’s flagship model is Claude, a direct competitor to OpenAI’s GPT-4 and Google’s Gemini. The company has raised $65 billion in May alone, a portion of which is earmarked for expanding compute capacity. The commitments to Amazon, Google, and Broadcom are not optional — they are the price of admission to the frontier model race.
The revenue run rate acceleration is real. $47 billion annualized from a standing start in early 2024 is extraordinary by any metric. But the question is not whether revenue is growing — it’s whether the growth is sustainable, and more importantly, whether it translates into cash.
In my experience auditing DeFi protocols during the 2020 yield farming bubble, I learned that high revenue growth often masks structural flaws. Compound’s total value locked soared, but the protocol’s native token, COMP, was being inflated to incentivize liquidity. The yield was not real — it was a transfer from later holders to earlier ones. When the inflation model collapsed, the APRs dropped, and the capital fled.
Anthropic’s revenue is real in the sense that customers are paying for API access. But the gross margin is opaque. The company has not disclosed its cost of revenue, but given the compute commitments, it’s likely negative. Every dollar of revenue may require $1.50 of compute spend. That’s not a business — it’s a subsidy from venture capital to end users.
Core: The Order Flow Analysis
Let’s break down the numbers with the precision of a delta-neutral strategy.
First, the valuation. At $2 trillion, the implied price-to-sales multiple on the $47 billion run rate is 42.5x. That’s not insane for a hypergrowth company — but it assumes the run rate continues to compound at the same rate. If the growth slows to a still-impressive 100% year-over-year, the multiple contracts. If the market realizes that the revenue is not profitable, the multiple collapses.
Second, the capital structure. Anthropic has raised debt and equity to fund compute. The $100 billion commitment to AWS alone is a long-term liability that will consume future cash flows. Even if the company achieves $200 billion in revenue by 2030, the capital expenditure will eat a significant portion. This is similar to the dynamics we saw in the crypto mining space: miners with cheap power contracts outperformed those with high-cost debt. The difference is that Anthropic is the miner and the data center operator simultaneously.
Third, the competitive moat. Anthropic’s advantage is not its model architecture — it’s the secrecy around its training data and alignment techniques. But code is law, and bugs are justice. If a competitor like Mistral or xAI achieves similar performance with lower compute requirements, the pricing power evaporates. The AI model space is commoditizing faster than Ethereum layer-2s. In 2022, I wrote about the OP Stack vs. ZK Stack debate, arguing that the technical differences were less important than the ability to convince developers to deploy. The same applies here: the winner is not the best model, but the one that can lock in the most compute and customer relationships before the commoditization hits.
Fourth, the distribution of profits. Osman’s analysis correctly points out that investors need to focus on how future profits are distributed among model developers, chip manufacturers, cloud providers, and software companies. I would add a layer: the real value capture may be in the infrastructure providers, not the model developers. Nvidia’s GPU sales are already a $100 billion business. Amazon’s AWS is a $250 billion cloud business. Anthropic is a tenant in a building owned by Amazon and Google. The landlord always wins in a real estate boom.
I’ve seen this pattern in crypto. During the 2021 NFT mania, I tracked wash-trading in Bored Ape Yacht Club, connecting it to Aave liquidations. The infrastructure (OpenSea, Ethereum) captured more value than the individual NFTs. The same logic applies here: the model is the NFT, but the compute is the blockchain.
Contrarian: The Retail vs. Smart Money Disconnect
Here’s the counter-intuitive angle. The market is pricing Anthropic as if it will be the next Microsoft. But the historical analog is not Microsoft — it’s WeWork. Both companies had charismatic founders, massive revenue growth, and a narrative that the world was changing. WeWork’s IPO was a disaster because the market realized that the business model was a lease arbitrage, not a technology platform.
Anthropic is a lease arbitrage on compute. The company is borrowing money from VCs and cloud providers to rent GPUs, then reselling the output at a markup. The markup is currently positive because the demand for AI inference exceeds supply. But when the supply catches up — and it will, because every hyperscaler is building data centers — the margin will compress.
The smart money knows this. The $65 billion raise in May was likely structured with liquidation preferences that protect early investors. The IPO will be a liquidity event for insiders, not a wealth creation opportunity for retail. The public markets will be the exit liquidity for the venture capital funds that invested at a $380 billion valuation.
I’ve seen this exact dynamic in the ICO era. The CryptoGem token I shorted had a similar structure: a hot narrative, a secretive team, and a token that was priced for perfection. The founders sold before the reveal. The market learned the hard way that code is impartial.
Takeaway: The Only Trade That Matters
If you are a retail investor, the question is not whether to buy the IPO. The question is how to profit from the volatility that will follow.
Anthropic’s IPO will be the most anticipated listing of the decade. The hype will push the stock to absurd levels on day one. But the structural flaws — negative margins, massive capex, commoditization risk — will eventually surface. The smart trade is to short the post-IPO euphoria using protective puts or put spreads. The exact timing is uncertain, but the direction is clear.
In my 2024 ETF approval trade, I profited from the mispricing of implied volatility by selling premium on CME Bitcoin futures. The same principle applies here: the IPO will create a volatility spike, and the market will overprice the call options. Sell the premium, wait for the decay, and collect the theta.
Code is law, but bugs are justice. The bug in Anthropic’s valuation is the assumption that revenue growth is synonymous with value creation. It is not. The only question is how long the market can ignore the flaw before the correction.