Analysis
I keep coming back to a simple pattern this week: the AI companies actually reaching public markets in 2026 are not the ones with the biggest names. CoreWeave is trading. Cerebras priced in May. Crusoe is meeting Goldman, Morgan Stanley, JPMorgan and Bank of America off a fresh $30 billion private mark. Oura, a wearable-hardware company, just filed an actual S-1. Meanwhile OpenAI and Anthropic -- the two companies whose valuations dwarf every name I just listed combined -- remain confidentially filed with no public document and no committed date.
My view is that this gap is structural, not sequencing. A data center company or a wearable-hardware company sells something a public-market analyst already knows how to model: contracted capacity, unit economics, a subscription attach rate. Crusoe's $13 billion Jane Street contract is a number an infrastructure analyst can build a discounted cash flow around. Oura's 5 million paying members is a cohort-retention question with historical comparables in Peloton and other subscription-hardware businesses, however imperfect. OpenAI and Anthropic sell something genuinely harder to underwrite: a competitive position that can shift with the next model release from a rival lab, and a safety and regulatory risk profile that's actively evolving in public in real time -- this same week, OpenAI's own chief scientist admitted Astra is becoming harder to monitor as a side effect of capability gains nobody planned for.
That's not a knock on either company's business quality. It's an observation about what kind of story a prospectus can actually tell convincingly. "We have signed contracts and a data center under construction" is a story public investors have priced a thousand times before, across telecom, real estate and industrial infrastructure. "We built the most capable model in the world and we're not entirely sure how it reasons anymore" is not a story any S-1 template was built to contain, and it shows in how much longer both labs' registration processes are taking relative to their infrastructure-layer peers.
“Crusoe's $13 billion Jane Street contract is a number an infrastructure analyst can build a discounted cash flow around.”
Room for disagreement: the strongest counter to my read is that OpenAI and Anthropic aren't actually behind on process -- they're simply choosing to optimize for valuation over speed, and preparing a mega-cap listing takes more work the bigger the number gets:
- Crusoe's private mark -- $30 billion, a scale of infrastructure investors already know how to underwrite
- Oura's IPO target -- $16 billion-plus, a subscription-hardware business with historical comparables
- OpenAI and Anthropic's eventual listings -- reportedly $1 trillion-plus, requiring far more investor education and risk-factor drafting than either of the above
If that's right, the gap closes entirely once both labs are ready, and my structural explanation is just a slower clock, not a different one. I'd also note Anthropic's own revenue run rate -- reportedly $65 billion annualized -- is real and growing fast enough that a sufficiently patient set of underwriters could plausibly price it the way they'd price any other high-growth software company, safety caveats and all.
The number I'm actually watching: whether Anthropic files a public S-1 before Crusoe does. If the infrastructure company gets there first, that's confirmation the gap is about underwriting difficulty, not competitive priority. If Anthropic beats Crusoe to a public document, my whole thesis needs revisiting.
