Trust is a bug. The market is preparing to assign a $2 trillion valuation to an AI company whose most critical numbers—cost structure, revenue concentration, net retention—are hidden behind a veil of 'pipeline optimism.' I've spent 28 years auditing protocols, from the DAO's recursive call to Optimism's gas estimation flaw. I know a vulnerability when I see one. And the Anthropic IPO narrative is full of unverified assumptions.
This is not a hit piece. It is a forensic audit of a market narrative. The facts are public: Polymarket gives a 70% probability of an October 2026 IPO, WSJ reports meetings with bankers, Reuters cites a 2028 revenue target of $1900–2000 billion and current annualized revenue of $470 billion. The goal: a $2 trillion market cap. That is larger than the entire crypto market cap at its peak. The question is not whether Anthropic is a good company—it is. The question is whether the market is pricing a story or a sustainable business.
Context: The Protocol of Public Markets
Anthropic is the developer of Claude, a leading AI model. It is backed by Amazon and Google, has raised substantial capital, and is now preparing to go public. The IPO is expected to be the largest tech listing in history, surpassing even the Alibaba debut. The valuation narrative is built on three pillars: (1) explosive revenue growth—$470B annualized now to $2T by 2028; (2) a market that believes AI is the next SaaS; and (3) the scarcity of high-quality AI assets in public markets. But like any protocol, the value is only as strong as the weakest assumption.
Core: Stress-Testing the Valuation Circuit
Let's start with the revenue projection. A CAGR of ~60% from $470B to $2000B over three years is aggressive but not impossible. However, the base is deceptive. Annualized revenue is not the same as booked revenue. It is a snapshot of the last month multiplied by 12. In AI, where customer acquisition is lumpy and enterprise deals take quarters, this metric can be misleading. I have seen DeFi protocols report 'TVL annualized' that collapsed within weeks. The same principle applies.
Now, the valuation multiples. At $2T market cap and $470B current revenue, the trailing P/S ratio is ~42.6x. That is higher than any major SaaS company at their peak. Salesforce's highest was ~25x. Zoom's was ~40x during COVID. But Zoom was profitable. Anthropic is not. The cost structure is the black hole.
AI inference costs are non-trivial. Each query to Claude consumes compute, and as usage scales, so does the AWS or Google Cloud bill. If gross margins are 50% (optimistic for a model provider), the net income margin at 25% requires massive operating leverage. That means the company must grow revenue faster than costs. In a competitive market where OpenAI and Google are also spending, that is not guaranteed.
I stress-tested a similar model during the 2022 DeFi crash. I traced the collapse of lending protocols to oracle latency and impermanent loss. Here, the 'oracle' is the market's expectation of AI adoption. If enterprise AI budgets slow—even by 10%—the revenue projection fails. The 2028 target becomes a mirage.
Contrarian: The Blind Spots the Article Ignored
The original source material from CryptoPotato is itself a case study in selection bias. It elevates Polymarket data—a prediction market with limited liquidity and a crypto-native user base—to the same level as Reuters and WSJ. That is a vulnerability. Polymarket is a fun tool, but it is not a reliable price discovery mechanism for a $2T IPO. The participants are not institutional investors with billions to deploy. They are speculators. The probability of 70% reflects sentiment, not reality.
More dangerous is the omission of infrastructure costs. The article never mentions the capital expenditure required to support the 2028 revenue target. If Anthropic must spend $100B to build data centers (a plausible number given current AI capex trends), the net present value of the company drops significantly. This is like a DeFi protocol that pays high gas fees for every transaction—the economics work only if volume is high and fees are low. But fees are rising.
And then there is the competitive landscape. The article treats Anthropic as a standalone winner, but the market is a duopoly with OpenAI. OpenAI has a more established ecosystem, broader product suite, and deeper Microsoft integration. Anthropic's 'safety-first' positioning is a differentiator, but it is also a governor. Safety-first means slower deployment, more red tape, and higher compliance costs. That is a feature for enterprise clients, but it is a cost center.
Takeaway: The Crash is Already Written into the Contract
If the market prices this IPO at $2T, it is buying a narrative. The narrative is not fraudulent—it is simply unverifiable. The company has not disclosed its cost structure, customer concentration, or net revenue retention. The revenue projection is based on a single year's target that is five years out. In software, that is a lifetime. In AI, it is a geological epoch.
I have seen this pattern before. The DAO was a brilliant idea with a fatal code flaw. The flaw was hidden in the assumption that external calls would not re-enter. Today, the flaw is hidden in the assumption that revenue growth will outpace cost growth forever. It won't. At some point, the market will demand proof. And if the proof is not there, the valuation will re-enter the downside.
Proofs over promises. The Anthropic IPO is a test of the market's ability to distinguish between a protocol and a story. I am watching. I am auditing. And I am not buying at these prices.
If it's not verifiable, it's invisible. The $2 trillion is a promise, not a proof. Until the S-1 reveals the code—the revenue breakdown, the gross margins, the capex plans—the wise investor treats this as a speculative overhang. The crash is already written into the contract. The only question is when the execution hits.