Bloomberg reported this weekend that both SpaceX and Anthropic are eyeing IPOs. Simultaneously, TechCrunch noted that Anthropic has not disclosed its actual user numbers, with estimates ranging from 18 million to 30 million, a 12-million-person uncertainty that would be disqualifying for any traditional public company filing. This is the central tension in AI's IPO moment: companies that have built their brand on epistemic humility, on being the safe AI company, are about to enter a disclosure regime that demands numbers, not vibes.
The Transparency Paradox of Safe AI
Anthropic's entire market positioning rests on being the responsible alternative to OpenAI. But The Verge's reporting on TikTok's broken AI ad labeling system illustrates what happens when AI companies hit the real world: disclosure frameworks collapse under commercial pressure. Samsung is running AI-generated ads without labels. Anthropic doesn't know how many users it has within a 12-million margin. The AI industry's relationship with transparency is, charitably, aspirational. A 2025 paper in Nature Machine Intelligence by Bommasani et al. found that fewer than 15% of major AI model releases included sufficient documentation for independent third-party audits, a metric that becomes legally significant the moment these companies file an S-1.
What IPO Markets Actually Reward
The SpaceX pairing is instructive. Both companies have cultish consumer followings, both have founder-driven narratives that transcend their financials, and both are going public in a moment when live VC intelligence signals suggest the window for AI infrastructure plays is narrowing as rate sensitivity returns. The conventional wisdom about not spray-and-praying investors applies to IPOs too: the Anthropic roadshow will need a tight story about who their customer actually is, how many of them there are, and why that number isn't embarrassing. The gap between 18 and 30 million users is not a rounding error. It is a thesis.