The last tranche of those warrants can be exercised if AMD’s share price, which leapt more than 23 per cent on news of the deal to $US203.71, hits $US600.
If all those warrants could have been exercised immediately, in agreeing to purchase AMD’s GPUs OpenAI would have created $US6.3 billion of value for itself from the $US63.4 billion the announcement added to AMD’s market capitalisation.
Generative AI apps are tearing through workplaces and are now increasingly used for coding.Credit: iStock
Altman has been engaged in a frenzy of high-stakes deal-making in recent weeks that adds up to mind-boggling sums.
Apart from the Nvidia and AMD deals, he has committed to buying $US300 billion of computing power from Oracle and is partnering with Oracle, Microsoft, Softbank, Nvidia and others in the $US500 billion Stargate data centres project. He’s also negotiating with Broadcom to develop custom chips for its next generation of AI models.
OpenAI, via the deals with Nvidia, Oracle, the Stargate partners and AMD, has committed to acquiring the equivalent of about 23 gigawatts of computing power at a cost that it estimates will top $US1 trillion. The Nvidia and AMD deals alone will cost OpenAI about $US800 billion over time.
The company that ignited the AI boom has been funding itself by raising equity and debt. It will have revenues of about $US13 billion this year, but will lose money because of the scale of the investments it is making. It expects to be profitable in 2029 – after burning more than $US100 billion of cash.
OpenAI has more than 700 million users, but only about five per cent of them pay for its services. That’s reflective of the entire sector, where companies are committing tens of billions of dollars each this year in the hope that eventually there might be a massive commercial return.
AI-related investments in the US this year are likely to approach $US400 billion and estimates of the investment required by the end of the decade range from about $US4 trillion to as much as $US7 trillion.
The “hyperscalers” – companies like Google’s parent, Alphabet, or Meta Platforms, or Microsoft and Amazon – have the vast cash flows from their existing businesses to fund their investments in data centres and chips, but even they are showing signs of the strain of funding such large non-cash-generating investments.
It is improbable that all those chasing a position in whatever the AI sectors looks like in future will survive, given the financial stresses involved.
The market’s enthusiasm for AI stocks is now being widely compared to the dot-com and telco bubbles a quarter of a century ago. When the dust settled, only a handful of the players survived, but Google and Amazon emerged with the foundations for today’s dominance established.
Something similar could be expected if AI is as transformative as it promises to be, although the increasingly incestuous nature of the sector, the exposures to each other’s balance sheets, the need to continually access equity markets for capital consumed in recurrent spending – GPUs have a useful life of between one and three years before they are made redundant by the next generation of chips – and the increasing use of debt makes the whole structure of the sector vulnerable.
Open AI and Nvidia are both allies and competitors in the quest for AI at, or superior to, human levels of intelligence. They are now formally linked via OpenAI’s commitment to buy Nvidia’s chips and Nvidia’s shareholding in AI.
The deal with AMD, which is developing a chip it believes will rival Nvidia’s next generation of chips, diversifies OpenAI’s supply chain and helps a Nvidia competitor which, in turn, might generate some price competition (Nvidia’s Blackwell chips can cost up to $US70,000 each) and lower OpenAI’s own costs over time.
Good bubble, bad bubble
It also, however, highlights the interdependencies that have developed in the sector, which is again reminiscent of what occurred during the dot-com boom and bust.
Loading
Amazon’s Jeff Bezos recently described the investment boom in AI and the stretched valuations afforded to anything AI-related within the sharemarket as a “good” kind of bubble, saying it was more of an industrial bubble than a financial bubble.
He referred to the vast amounts of fibre optic cabling deployed in the 1990s that survived even after most of the companies that laid it didn’t as an example of a good bubble.
It is almost inevitable that at some point, the gap between what companies are investing in AI, the cash they are generating from that investment and the need to continuously invest in the latest generation of chips will become too much for some to sustain. Even good bubbles eventually burst.
Investors and lenders will look at the returns required in future to make sense of the investments being made today and blink, and markets will shudder.
Whether that looks like the bursting of the dot-com and telco bubbles in 2000 is impossible to predict but, leaving aside the hyperscalers, the emerging structure of the sector and the demands being made on its finances by the scale of the investments required are making it increasingly vulnerable.
The Business Briefing newsletter delivers major stories, exclusive coverage and expert opinion. Sign up to get it every weekday morning.
