OpenAI – The Pick and Shovels Trade

OpenAI promotes suppliers’ shares over its own.

• OpenAI’s revenue is on track to hit $10bn in 2025 and $125bn in 2029 but the fact it won’t make money until the company is a gigantic global corporation is an indication that its business model is fundamentally flawed with the real winners from this profligacy being its suppliers such as Nvidia, CoreWeave. Jony Ive and so on.
• Some details have leaked regarding OpenAI’s expectations for the coming few years that are being used to support its $40bn fundraising at an eye-watering valuation of $300bn post-money.
• This includes revenues doubling in 2025 to $10bn and continuing to grow at a compound annual growth rate of 88% until 2029, when revenues will hit $125bn.
• Given the current excitement around AI and the rate at which everyone is spending on AI infrastructure and services, this number is vaguely credible, but if there is a correction (as I expect), then OpenAI will miss this by a country mile.
• This correction is likely to be caused by the realisation that statistical-based systems like LLMs have no understanding of causality and are incapable of reasoning, meaning that the world is not going to change overnight, but instead simply become much more productive.
• Hence, what is going on right now is a substantial overbuilding of capacity, which will result in prices falling and the business model of providing AI services not being as profitable as everyone seems to think.
• This would make everyone’s medium-term forecasts much too high, which in turn will trigger a valuation correction and a shake-up in the global pecking order in AI.
• It is at this moment that I expect Apple to come in and acquire one of the big AI suppliers that are currently burning money with gay abandon and turn into the in-house solution that it badly needs.
• However, even if I am wrong, there is something badly wrong with OpenAI’s forecasts.
• Most start-ups that survive will start breaking even around the time that they become a small to medium-sized enterprise, but OpenAI says that it will not make money until it is already a gigantic global corporation.
• I think that this is a crazy state of affairs, as a business that is unable to make money before it is gigantic is unlikely to be a going concern and is therefore at great risk of going bankrupt the minute there is a disturbance in the market, such as a correction or global crisis.
• However, it is clear that OpenAI intends to continue to spend money like there is no tomorrow, and while everyone seems to think that this makes OpenAI worth more money, I think it makes OpenAI worth less and its suppliers more.
• Between now and 2029, OpenAI’s cumulative revenues should be around $256bn, and if its forecasts are correct, its expenses will be about the same.
• This means that large suppliers of infrastructure to OpenAI, either directly or indirectly, are going to make out like drunken sailors.
• Consequently, the real place to invest when it comes to the AI trade remains the picks and shovels.
• These are the companies that make money now, giving them the reserves to withstand any market fluctuations.
• This continues to lead me directly to Nvidia, where, although competition is certainly heating up, Nvidia has seen it coming and has moved to mitigate the threats.
• The adjacencies, such as all of the other components that go into the data centre, inference at the edge suppliers, nuclear power and so on, are also going to fare well.
• It is the contenders themselves that I worry about, and I suspect that by 2029, OpenAI, Mistral, Anthropic, and so on will not exist as independent companies as they will be acquired when they get into trouble.
• Hence, the only direct AI investment I would consider is Nvidia, but I continue to think that there is more value to be had in the adjacencies of inference at the edge and nuclear power, where I hold positions in my portfolio.