Anthropic will be 'most ridiculous IPO' of year, analyst says

New Constructs values Anthropic at $150 billion and calls its potential IPO the “most ridiculous IPO of 2026,” warning that a valuation approaching $2 trillion would require profits exceeding Nvidia’s trailing $190 billion net income. The firm cites reported 2025 revenue of $4.6 billion and a $42 billion net loss, while Anthropic said its annualized revenue run rate reached $65 billion at the end of July, up sevenfold year over year, and was on pace for $100 billion by year-end 2026. The bearish assessment comes from a firm with both successful and unsuccessful IPO calls; Anthropic has not made its prospectus public.

Analysis

The key underwriting risk is not whether Anthropic can grow, but whether revenue converts into durable owner earnings after inference costs, model development, and customer acquisition. A reported $100B year-end annualized run rate would put a hypothetical $2T valuation at roughly 20x that run rate—but run rate is not audited revenue, and says little about gross margin, retention, or capital intensity. With no public S-1, the short thesis remains a valuation and disclosure-risk warning, not a verified earnings model.

The competitive spillover is two-sided. Open models and falling inference prices could cap closed-model pricing power; alternatively, cheaper models may broaden usage and increase aggregate demand for compute. That makes a reflexive short of NVIDIA (NVDA) premature: model-provider economics can weaken while compute demand remains strong, though efficiency gains are a medium-term risk to compute intensity. DoorDash (DASH) is a useful caution against treating a bearish IPO call as a timing signal, not a relevant operating comparable.

Near term, prospectus disclosure, IPO pricing, and demand will dominate. Over 1–3 months, track realized revenue versus run-rate claims, gross margin, customer concentration/retention, and cash burn. Over 6–18 months, open-model quality and inference-cost declines test whether premium closed models retain pricing power. The contrarian case is that rapid adoption could create operating leverage; it is falsified if growth fails to translate into improving unit economics. New Constructs has a track record of both notable misses and hits, so its conclusion should not be treated as an independent valuation endpoint.

Key Decisions for Investors

• Do not trade the bearish report as a standalone short signal. Keep Anthropic on an IPO watchlist and defer any position until the public S-1 permits verification of recognized revenue, gross margins, cash burn, customer concentration, and share-sale/lockup details.
• If the offering prices near the reported $2T scenario, require evidence that growth is converting into improving unit economics before participating; reassess after initial trading rather than chasing a first-day move. The thesis weakens if audited filings show durable margin expansion and strong customer retention.
• Avoid using this article alone to short NVDA. Monitor hyperscaler/model-provider capex and compute demand alongside inference efficiency; a sustained capex slowdown or falling accelerator demand would be a clearer negative catalyst than losses at one model developer.
• Treat open-source adoption and declining inference prices as a sector watch item: they could pressure premium closed-model economics, but may also expand total usage. Seek confirmation in disclosed pricing, retention, and gross-margin trends before taking a broader AI software short.