Research Firm Calls Anthropic's IPO 'Unprecedented Test of Investor Gullibity'
Skepticism is mounting over Anthropic’s rumored $2 trillion IPO valuation, with critics arguing that its soaring revenues mask massive training and infrastructure costs and an uncertain path to durable profits. Commenters question whether any closed LLM provider can sustain such a valuation as open-weight and cheaper competitor models rapidly improve and switching costs remain low. Others counter that frontier AI could still capture a meaningful share of future productivity and GDP growth, but even they see the current pricing as a high‑risk bet rather than a clearly justified value.
Valuation and Bubble Concerns
- Many see a ~$2T IPO as unjustifiable given fierce competition, low switching costs, and unclear moats; likened to WeWork-style overvaluation and a broader AI bubble.
- Others expect a classic “overpriced IPO → price drop → long-run recovery” pattern similar to Uber or SpaceX, with early volatility but strong eventual returns.
Revenue, Profitability, and Cash Burn
- Reported numbers: rapid revenue growth (2024→2025→2026) and huge “annualized run rate,” but posters stress this is not the same as realized revenue.
- Skepticism around “run-rate revenue” and EBITDA-style metrics; concern that training costs, data centers, hardware commitments, and reinvestment needs are being downplayed.
- Some say Anthropic has had “profitable” quarters only by excluding major expenses like model training, so true economic profitability is unclear.
Competition, Moats, and Open Source
- One side: Anthropic’s edge is being “ahead” on frontier quality, convenience, and enterprise readiness; inference is profitable per token, and demand plus falling unit costs could eventually outweigh training spend.
- Other side: being ahead is a weak moat when catch-up is easier than frontier R&D; Chinese labs and open-weight models (DeepSeek, Qwen, etc.) are rapidly “good enough” and much cheaper.
- Debate over open source impact: some claim little real-world adoption outside tech circles; others see many smaller firms already moving production to cheap/open models.
Demand for Intelligence and Use Cases
- Bulls argue demand for “intelligence” is effectively unbounded; better models unlock entirely new high-value tasks, not just current “AI use cases.”
- Skeptics counter that for many tasks there’s a ceiling where extra model quality doesn’t add value, and real-world bottlenecks (regulation, materials, trust) limit monetization.
Comparisons to Other Bubbles and Firms
- Frequent analogies to dot-com, tulip mania, Enron-style accounting, WeWork’s failed “tech multiple,” and to chip fabs/large manufacturing (high capex, modest multiples).
- Nvidia and memory makers are framed as current clear financial winners from the AI boom; frontier labs are seen as in a “game of chicken” over who runs out of cheap capital first.
On the Research Firm and Investor Stance
- The critical research note is viewed by some as biased and often wrong on past IPOs; others still find its skepticism useful.
- Thread contains both declared buyers (“transformative tech, solid lead”) and would-be short sellers who see the IPO as a retail wealth transfer.