Open AI – Achilles Heel

OpenAI is hostage to its fundraising.

• SoftBank has blown another hole in OpenAI sentiment when it admitted that nothing had been decided with regard to further investments in OpenAI, clearly demonstrating that if the money dries up, the game is over.
• SoftBank reported a return to profitability in FQ3 26 thanks to an unrealised gain of $20bn in its investment in OpenAI, but at the same time stated that it had not decided about further investments.
• This is very different to Jensen Huang’s statements a couple of weeks ago, where he was simply stating the nature of the deal that Nvidia has with OpenAI, much to the surprise of the press which had got it wrong from the start.
• This has been jumped on by the doomsayers as evidence that OpenAI is about to go bankrupt, but in reality, nothing has changed.
• When the deal with Nvidia was announced, it was stated that Nvidia would invest in OpenAI and at the same time, that OpenAI would purchase 10GW (around $100bn) of AI systems from Nvidia.
• The press took this as Nvidia would pump in $100bn immediately, but Nvidia and OpenAI were always fairly clear that this is not how the deal works.
• This is effectively a frame order from OpenAI split into a number of segments (probably 1GW each), and as OpenAI deploys each segment, Nvidia is issued shares in OpenAI as payment for the silicon.
• There may also be cash going backwards and forwards, but once the two sides of the equation are netted out, this is what one is left with.
• Frame orders are common in the capital equipment industry, which is where the price and amount of equipment are agreed upfront but the deal is not guaranteed.
• Instead, the customer can fulfil parts of the order as it rolls out the equipment, and there are no penalties for delays or even cancellations.
• This is the nature of the deal that Nvidia has signed with OpenAI, and in this context, Jensen’s comments make complete sense.
• If, for some reason, OpenAI doesn’t roll out the Nvidia equipment, Nvidia won’t buy the shares, and this has never been a secret to anyone paying attention.
• The situation with SoftBank is similar, and so while this represents no change to the situation, it has had a deleterious impact on market sentiment towards OpenAI.
• The nature of the frame orders has never been explicitly spelt out, and the deal has been very slow to be signed, which has led the market to think that OpenAI’s biggest backers are becoming more cautious on the outlook for the company.
• OpenAI is not a going concern because if the fundraising grinds to a halt, the company will probably go bankrupt within 3 months.
• The souring sentiment increases the risk of this happening regardless of what is going on behind closed doors.
• For example, if the market thought that OpenAI was going to go bankrupt and priced that into SoftBank’s shares, one can be pretty certain that SoftBank would not make a further investment.
• This is why sentiment matters so much, and so while sentiment rose meaningfully on the idea that Nvidia and SoftBank were going to pump billions into OpenAI as a result of sloppy reporting, the reality of the situation is beginning to emerge as it always does.
• The net result here is that these incidents expose the fatal flaw in OpenAI (and Anthropic) and highlight the major advantage that Google, Meta and Microsoft have over their start-up competitors.
• These companies have mature and established businesses that generate so much cash flow that they can afford to spend like crazy on AI data centres without destroying their balance sheets.
• Hence, should liquidity suddenly dry up, these companies will survive while the start-up crowd led by OpenAI and Anthropic will quickly run out of cash and be acquired for a much lower valuation.
• This, combined with the fact that Google in particular is challenging OpenAI and Anthropic for leadership in model performance, does make me increasingly concerned about the very large IPOs that are expected this year.
• Their F1 statements are likely to reveal that they will still need to raise vast amounts of money after going public before they become self-sustaining, which, in my opinion, is a major mistake.  
• Companies that go public should be going concerns, meaning that if all goes wrong, they will at least survive.
• One only has to look at the wreckage of the EV companies that went public and then struggled to raise more money for evidence of the nightmare that awaits OpenAI and Anthropic if the market gets tired of pumping cash onto the bonfire.
• It is clear to me that I would not want to touch either of these two companies and instead prefer to go where the money is and the valuations are at least semi-rational.
• Here, I find the memory companies that are already sold out for this year and trade on less than 10x 12-month PER, where Samsung is one that I hold.
• Elsewhere, long-term adjacencies such as inference at the edge and nuclear power are still positions that I hold.