OpenAI – The Elusive Trillion

OpenAI could be the dark horse of the AI industry.

• OpenAI appears to be delaying its IPO, which will prevent an embarrassing rejection by the market but also reveals a company that I have long believed is not ready for the heat and scrutiny of the public market.
• OpenAI’s valuation remains stubbornly stuck at $730bn while its arch-rival Anthropic has blasted through the $1tn barrier, is likely profitable and should attract strong demand when it goes public.
• Delaying the IPO is a sign that this is not the right time for OpenAI, and I suspect that its internal systems are not yet ready to be able to report reliable quarterly earnings.
• Furthermore, the governance of this company also remains a mess, although cleaning up the relationship with Microsoft has clarified the structure somewhat.  
• These are all reasons to delay the IPO, but the biggest one appears to be OpenAI’s likely failure to receive a valuation in excess of $1tn, which appears to be unacceptable to Sam Altman.
• It is also a clear sign that OpenAI’s financial performance has been eclipsed by Anthropic which I think may have broken even in Q1 2026.
• This is a result of Anthropic commanding a premium price for its product, evidenced by the fact that when it comes to coding, no one talks about anything else and my estimate that Anthropic is earning around $20bn – $25bn per GW of compute capacity.
• Compute that costs around $41bn per GW to build, and $1.5bn to run annually offers a return of more than 20% when it generates more than $20bn / GW in annual revenues which is why I think Anthropic is currently enjoying healthy gross margins.
• However, Anthropic has just leased capacity from xAI at a price of $37.7bn per year per GW, meaning that it will need to double its prices to continue being profitable.
• At this price, it makes more sense to build the capacity in-house, but that means waiting at least 2 years for compute to come online, and most of the companies think that they do not have that kind of time.
• Everywhere one looks in the market, the token-maxing trend has given way to token rationing, as the sellers of tokens have shifted from all-you-can-eat to pay-as-you-go as scarce compute has driven prices up substantially.
• This is where OpenAI may end up being the dark horse.
• Anthropic was very late in securing compute, and so it is having to pay a large premium, whereas OpenAI closed its deals out to 2030 over a year ago.
• This means that there is a good chance that OpenAI is paying Oracle and its other providers a fraction of what Anthropic has signed for, meaning that its cost base going into 2027 and 2028 could be far lower than its peers.
• This could allow it to substantially undercut Anthropic and be profitable, leaving Anthropic unable to compete.
• This would be the time to go for IPO and so a decision to delay to next year does make more sense than one man’s vanity.
• This assumes that its suppliers are unable to wiggle out of their contracts and have no capacity to raise prices, and if I were Oracle or CoreWeave, I would be looking at exactly this.
• Hence, I am warming up to the idea of OpenAI as the market increasingly dislikes it and it may have an ace up its sleeve.
• In the meantime, the focus will now switch to Anthropic’s IPO where the numbers will need to be spectacular to justify a valuation north of $1tn.