OpenAI just announced a $300bn cloud deal with Oracle to power it’s AI solution with advanced infrastructure. The announcement doubles the annual spend from a previous announcement from $30bn a year, to $60bn a year. Pundits aren’t certain why the increase is necessary, and some question if that level of compute capability can be built by Oracle by 2028 when the deal starts.
In many ways, AI infrastructure is a hotter topic than AI itself, which is starting to cool. Billions of dollars of paper are already exchanging hands. Altman reckons he’ll need trillions in infrastructure just to support OpenAI. Multiply that by 10 several times if all the AI firms out there reach scale—which they almost certainly won’t.
The real questions is where this cash will come from. OpenAI’s revenues are growing—some estimates say it’s now banking $1bn a quarter—but reaching in excess of $60bn annually to cover just one of the infrastructure deals it has on the cards seems far out of reach. The ubiquitous Amazon, which also has interests in AI and Cloud, pulls in $638bn a year. And they famously have their fingers in almost every pie.
There’s also the question of how much private capital is around to keep companies like OpenAI solvent. Estimates of global private markets peg assets under management at just under $12 trillion. At this rate, handing all of that to AI firms so they can partner with infrastructure companies might not do the trick. To put it in perspective, back in 2024 Altman chased $7 trillion to tackle chip building challenges, more than the market cap of several of the world’s biggest companies combined. At this rate, there’s every possibility the cash will run dry long before our AI ambitions are met.
