The big questions remaining after Anthropic’s blockbuster IPO prospectus

Anthropic reported $16.29 billion in revenue and $2.38 billion in cost of revenue in the first half of 2026, implying an 85% gross margin; revenue was up from $1.16 billion a year earlier. Sequential growth slowed from about 144% in Q2 to roughly 47% in Q3, while compute costs, capacity, future margins and investor acceptance of founder voting control remain open questions. Analysts estimate a potential IPO valuation of $1 trillion to $2 trillion, but Anthropic has not disclosed a target.

Analysis

The key underwriting question for Anthropic (ANT) is not headline gross margin but marginal economics: if serving demand and training new models require compute commitments to rise alongside revenue, current gross margin may overstate the cash economics available to equity. The undisclosed training-versus-inference split is therefore a central diligence item, as is whether capacity constraints are limiting monetization. Slower sequential growth raises the hurdle for a valuation anchored to rapid scaling; founder voting control adds a potential governance discount that cash on hand does not remove.

Near term, an IPO could act as a valuation and liquidity test for private AI assets, while also drawing investor capital away from already crowded AI exposures. Over 1–3 months, monitor IPO pricing, share-class terms, compute commitments, and evidence that growth is converting into operating leverage. Over 6–18 months, the structural winners are suppliers able to monetize sustained infrastructure demand, including SK hynix and GPU vendors; the risk is that debt-funded or capex-heavy buildouts outrun durable customer returns. Microsoft’s cost discipline and SpaceX’s financing highlight that compute demand does not automatically imply attractive returns for every buyer.

Contrarian view: abundant cash and strong demand may give Anthropic room to invest through near-term margin volatility, and constrained capacity could represent foregone revenue rather than weak demand. But neither establishes attractive marginal returns. No valuation is supplied, so an IPO trade is premature; a rich offer price could make even strong growth insufficient.

Key Decisions for Investors

• Do not initiate an ANT position before the final offer valuation, share-class terms, and lock-up details are available. Reassess if pricing embeds continued rapid growth without evidence on marginal compute cost or operating leverage; founder voting control warrants an explicit governance discount in the underwriting.
• Set an IPO watch item for the compute mix: seek disclosure separating inference from training expense, capacity commitments, and customer demand that cannot currently be served. A rising compute burden relative to revenue or a further sequential growth slowdown would falsify the margin-scalability thesis.
• Avoid chasing NVDA, AMD, or SK hynix solely on aggregate AI demand headlines. Revisit supplier exposure over the next 1–3 months against customer capex plans and funding capacity; a pullback in announced infrastructure spending or evidence of underutilized capacity would weaken the demand thesis.
• Treat the IPO as a possible near-term liquidity and valuation catalyst for AI equities, not as a standalone sector buy signal. Watch the offer price and subsequent trading: a weak reception despite strong demand claims would argue for reducing crowded AI exposure; resilient pricing with improving compute economics would support revisiting the stance.