Circle Twerk

Sam Altman, Mel Brooks and Comic Timing

The number that stuck in my head the other week while chewing on massive OpenAI mega-deal with AMD - or any of the other OpenAI + Insert Tech Giant Here deal – was an almost quaint $13 billion. Not the $78bn in AMD’s MI450 chips, or the warrants on 160 million AMD shares that OpenAI is getting in return (roughly 10% of the company). It isn’t even the 40% pop in share price for the chipmaker. If all goes according to hype, AMD’s stock would triple what it was trading just a few week ago.

The deal was touted as part of OpenAI’s strategy to unseat Nvidia’s dominance of high-end semi-conductors. Ok fine, this is a technological revolution, baby. Other actors like Google and Amazon, are also gunning to break Nvidia’s 70% marketshare on the chips that make the AI magic happen. On the surface the AMD deal was just a plucky billion dollar start up trying in to secure future computing power. Step back, and the entire ecosystem starts to look a bit bent. The AMD deal was announced just weeks after Nvidia’s announcement that it would invest up to $100bn in OpenAI so that it could buy Nvidia’s chips and forget about designing its own. The tentative Nvidia/OpenAI arrangement was announced just days after Nvidia declared that it would plow $5bn investment into its struggling competitor, Intel, to diversify away from Taiwan’s TMSC. Nvidia has the cash and these investment make a certain amount of Machiavellian sense.

OpenAI is a different story. It’s a privately held non-profit that is deeply entangled with Microsoft - the world’s second largest company - in a way that neither company can fully explain. It has announced a $300bn partnership with Oracle for the White House’s Stargate project. None of it changes the fact that the crucial number here is still that paltry $13bn getting lost in the drum beat of new, huge OpenAI commitments.

That is OpenAI’s revenue. It has brought in $13bn while paying out $16bn on renting computing services alone. An expense that is expected to hit $400bn by 2029. And that assumes that the energy infrastructure can supply to needed juice to run all those data centers. Broadly, AI doesn’t make money yet and, per MIT, some 95% of AI projects are failing to result in any ROI. About half the cost of these data centers are the chips, not the infrastructure - which need to be upgraded to stay competitive. Unlike laying railroad tracks or fibre optic cable, these are not cases of massive upfront costs that drop off after the bild out. With AI the buildout may never stop outpacing revenues. Which raises the question, just how in the hell is OpenAI going to meet these commitments as its expenses increase exponentially while it’s revenue climbs linearly? The privately held firm is starting to use its own shares as currency to make these vendor financed mega-deals happen.

A quick rundown of the less glorious parts of my CV include a getting laid off when the advertising agency I worked for got clobbered in the dot com crash, then jumping ship from the bond business at the 2008 financial meltdown. These circular dynamics strike a cynical watcher as a little like Mel Brook’s brilliant The Producers, about two broadway producers who figure out that they can use sell shares in the production as cash, provided that the play never makes money. I’m not implying a deliberate fraud, but messianic self-belief in your own genius Even AMD said that the financing was creative - or words to that effect.

Vertical integration makes a certain sense, even if it comes with its own blind-spots. This circular investing with competitors, while trying to unseat them, makes less sense. There is an incoherent disconnect between short-term tactics and long-term strategy that is being obscured by razzmatazz. The fundamentals have been replaced with hype and a velocity that is creating its own g-force. You can only ignore gravity, or reality, or returns for so long. Technology aside, future revenues will be constrained by the capacity of energy infrastructure: we simply haven’t got it and that will be a long, expensive build out. Financing can get very creative, the immutable laws of the physical universe tend to stay put.

The danger here, if you are inclined to take a look, isn’t so much that investors will be blindsided when the whole inbred thing goes sideways. Pretty much anyone can see that the AI bubble is filling. I’ve called two bubbles, I knew what was coming and I wasn’t alone. Most of us knew the correction is coming. It’s a matter of timing that moment when exuberant greed turns to fear.

Which is just not something AI can predict.

___________________

For more insights to tear things up, head to the 4717:

AI: Hype & Glory : A trillion dollar answer in search of a trillion dollar question

Big Theater, Small Door : 4717’s Artificial Intelligence Insight