OpenAI – Stalking Horse

OpenAI may come from behind in 2027.

• OpenAI looks to be raising money again at a valuation of $1.2tn, and given the evolving fundamentals in the AI industry, I think that OpenAI could very well be the company to watch in 2027.
• This news comes hot on the heels of the news that SoftBank is looking at raising more debt, meaning that there is more to this possibility than just a wild rumour.
• OpenAI has had a rough 2026, having been soundly beaten and overtaken by Anthropic, but I think that the stars are lining up for the company to fare far better than Anthropic in 2027.
  • First, cost of compute: In 2024 and 2025, everyone watched agape as OpenAI made massive capital commitments to AI data centre infrastructure, but because it was first, it appears to have been able to negotiate the best price.
  • Indications are that OpenAI has contracts with Oracle at somewhere between $10bn – $16bn per GW while Google and Anthropic have just agreed to pay $30bn – $50bn per GW to xAI.
  • This makes OpenAI far and away the cost leader.
  • Second, token consumption: where OpenAI is claiming that GPT-6 is on par (or slightly better) with Fable-5 at many tasks but at the same time uses substantially fewer tokens.
  • How OpenAI has managed to achieve this is not clear, but it has come at the cost of visibility, as GPT-6 does not output the same thinking trail that its predecessors do.
  • This means that customers will burn through their token budgets more slowly and get more bang for each million tokens that they buy.
• Both of these factors will have a substantial impact on OpenAI’s ability to compete against Anthropic (and everyone else) as its total cost of compute should be substantially lower.
• Furthermore, if it can complete the same tasks but using fewer tokens, customers will find that their bills are lower by another significant margin, further enhancing its attractiveness for customers.
• OpenAI has just passed $40bn in annualised run rate, while the expectation is that Anthropic will pass $100bn by the end of the year, explaining how Anthropic can look at going public at $2tn while OpenAI is raising at $1.2tn.
• For 2027, it looks like compute pricing will remain at supernormal levels ($30bn/GW – $50bn/GW), which is great news for anyone with capacity to sell, as this will generate 5-year returns of better than 60%.
• It not such good news for the AI service providers as they will have to increase their prices in order to make an acceptable return on the services that they provide.
• So far, demand has remained relatively insensitive to price and even now, there is still more demand than anyone can supply the capacity for.
• This is why everyone who has capacity is looking at selling that capacity into the spot market, as the returns are far greater than using the capacity for one’s own products and services.
• Against this backdrop, OpenAI should enjoy the industry’s highest gross margins by a significant margin in 2027, assuming that its suppliers are unable to renegotiate the contracts they signed in 2024 and 2025.
• Hence, I think that 2027 could be the year when OpenAI grows faster than Anthropic and potentially takes a lot of market share.
• This makes 2027 the right time to go public, as the company should be nicely profitable by then and have a credible growth trajectory ahead of it.
• Whether that trajectory is enough to justify a $2tn valuation at IPO remains to be seen, but OpenAI is the company to keep an eye on next year.