OpenAI – What IPO?

IPO hopes are receding.

• OpenAI missing its forecasts is not a big deal from a financial perspective, but it hammers home that enterprise is adopting AI before the consumer and that the torch of industry leadership has passed from OpenAI to Anthropic.  
• This also raises doubts about OpenAI’s readiness for IPO in 2026 as well as SoftBank’s wisdom in betting everything on this company.
• The end of 2025 should have seen OpenAI pass 1bn weekly active users (WaU), but instead it looks like upstarts like Gemini have taken market share despite ChatGPT being a global household name.
• It is still the market leader, but the gap is closing and when one plans to spend $1.4tn on infrastructure and raises money at a valuation of $850bn to do so, there is very little margin for error.
• This slender margin is clearly on display today at SoftBank, which has committed something in the order of $65bn to OpenAI and where news of this issue has sent the shares down 10% in Tokyo.
• This, combined with what I think is a misunderstanding by the Wall Street Journal (WSJ), is leading the market to think that if OpenAI misses its numbers, then it is in financial difficulty.
• This stems from reports that OpenAI’s CFO has voiced concerns that, unless the revenue targets are achieved, the company will not be able to afford its data centre build-out commitments.
• This is being taken as a sign of financial distress, but I think this is very far from the case.
• This is because none of the commitments that OpenAI has made to buy compute from others or build it itself are not set in stone.
• Instead, they are frame agreements where the volume and price are agreed, but if it turns out that OpenAI decides to delay or cancel these commitments, then this can be done with no penalties.
• Hence, missing a revenue estimate at the end of 2025 or in the first months of 2026 does not spell the end of the company, but it does raise valid concerns as to whether it will be able to go public this year.
• In this environment, anyone involved in AI will want to go public as soon as possible, as right now the supply of paper (shares) is at its lowest, meaning that whoever goes first is likely to get the best price.
• By mid-2027, there are likely to be many more AI companies in the public domain, and if the economics of compute have not improved, we could be well into a financial correction.
• This is why if OpenAI is going to go public (I am far from convinced that it should), it needs to go sooner rather than later.
• However, reports of missing targets and the fact that its equity is now trading below both the last raise valuation and Anthropic in private markets are more troubling signs.
• Hence, I think the probability of OpenAI going bankrupt remains low as missed revenues can be dealt with by cutting capex with no penalty, but whether OpenAI is a going concern long-term is an open question.
• This is because its revenue per GW of compute capacity remains below what I would consider to be an economically viable threshold, meaning that it must increase this number or go under.
• Revenue per GW remains static at around $10bn/GW and RFM Research has calculated that it needs to rise to at least $15bn/GW or better $20bn+/GW in order to have any hope of growing into the $1tn+ price that is likely to be asked for at IPO (see here).
• Failure to get the IPO away successfully would be a disaster and could easily trigger a valuation reset across the industry.
• Hence, I think the probability of an IPO from OpenAI this year is fast receding, which will not please Mr Son and his backers at all.