Crossed Wires: The biggest, baddest, riskiest tech deal in history

The problem with large numbers (dollars, gigawatts, speed, size, etc.) is that they often stare uncomprehendingly back at you from the page until a comparison to something else can be made to give them context – perhaps something simpler or smaller to give the numbers relative colour. So, bearing this in mind, consider these two statistics, which, even when contextualised, fall into the WTF category.

The first, which I came across in a blog called the Power Law, compares a recent deal in the world of AI with the Manhattan Project – the US-funded WW2 project to develop the atom bomb. At the time, this was the largest and most expensive scientific project ever undertaken, and it lasted from 1941 to 1945. The total spend in today’s dollars was ~R648-billion ($37-billion).

Over the last couple of weeks, a series of connected deals has been announced between OpenAI (mother of ChatGPT), Nvidia (AI chips) and Oracle (cloud infrastructure). When unpacked (and there’s the rub), it is worth around R7-trillion ($400-billion) – 10 times the size of the Manhattan Project. This will be the biggest commercial deal in history, if it all pans out (which we will get to).

The deal itself is a bit of a headscratcher, at least at first reading. It looks like sleight of hand, a piece of magic – it’s not, but it is a vertiginous high-wire act.

It goes like this: Nvidia (the largest company in the world with a market cap of R73-trillion ($4.5-trillion – another big number) is investing R1.75-trillion ($100-billion) in OpenAI equity. This is small potatoes for Nvidia. For OpenAI, a company worth R8.8-trillion ($500-billion), it is substantial – it is 20% of its value.

What does OpenAI do with the money? Well, they are not actually getting cash. Their goal is to build the baddest-ass AI before anyone else (call it AGI, if you must), and Nvidia is going to assist by building and deploying R1.75-trillion worth of AI hardware for them (chips and racks and power connectors and the like). This is not that unusual – it is called vendor financing, in which the makers of stuff provide the funding for their customers to buy their stuff. It is a little risky because, if the customer goes belly-up, the equipment maker is out of luck. Nvidia is clearly confident that OpenAI will stay in business.

Here is where it gets more interesting. Having a bunch of expensive AI hardware is not enough to build an AI megaproject. You also need somewhere to put it. And that means “Tier 4” data centres – industrial, global, energy-hungry behemoths with scads of redundancy, floor space, comms and other bells and whistles. Oracle has lots of them. That’s its main business these days (it used to sell database software only, but the cloud infrastructure business has grown from sideline to top dog over the past two decades).

So it’s in Oracle’s data centres that Nvidia is going to deploy the AI hardware it built for OpenAI and for which it has received OpenAI shares.

OpenAI has committed to buy R5.3-trillion ($300-billion) worth of cloud services from Oracle for five years starting in 2027. It is a promise, an IOU, a signed contract which will be paid for with, um, profits? Stock sales? Loans? Like Nvidia, Oracle appears to be pretty confident that OpenAI will meet their obligations.

At the risk of oversimplification – Nvidia buys OpenAI shares and in return builds some AI stuff for them. OpenAI commits to buying data centre space from Oracle, and that’s where the stuff will be housed. Everybody is bound by commitments, interlinked, entangled and circularly chained, and based on their commitments, all their forecasts look rosy. Oracle books R5.3-trillion in new revenue; Nvidia books a nice hunk of equity and a strategic relationship with the world’s hottest AI company; OpenAI underpins its forecasts with the comfort of having secured the underlying hardware and services – assets which they will now endeavour to sweat.

What could go wrong?

For those who enjoy examining the emperor’s clothes or pointing out the lack thereof, I wanted to find out how much money OpenAI would have to make to justify the financial commitments of this megadeal. To do this, I asked ChatGPT for help (assuming it would not lie to me just because its mummy is OpenAI). I fed in all the public parameters of the circular deal and asked:

“How much revenue does OpenAI have to generate over the next five years to make this commercially viable? Think long before you answer.”

ChatGPT indeed thought long and finally spat out a financial model which I duly checked.

Here is the bottom line:

OpenAI will have to generate an average of R1.75-trillion a year in revenue (at 15% margin) for the next five years if the deal is to make sense. OpenAI is currently generating R175-billion in annual revenue. Do the maths. It has to grow its revenue by 10× in a very short space of time.

Place your bets, folks. Are businesses and citizens going to find uses for AI that will cause OpenAI revenue to scale up that dramatically? Alternatively, and perhaps more worryingly, will Nvidia chips continue to reign supreme in a world where everyone – including the Chinese – wants to take them down?

Remember the size of this deal: R7-trillion, the largest in history. If this goes pear-shaped, there is going to be, er, big trouble for many companies, their stock prices and their investors. 

Finally, I promised you two WTF statistics. Here is the second.

In justifying the deal, OpenAI CEO Sam Altman recently said: “Our vision is simple: we want to create a factory that can produce a gigawatt of new AI infrastructure every week.”

A gigawatt. A billion watts. That is equivalent to the amount of energy required to power 750,000 homes. Altman wants to roll that out every single week. He wants to consume as much energy annually fulfilling his AI aspirations as 10% of the US generating capacity for the entire country’s needs.. In response, let me simply quote Macbeth:

I have no spur
To prick the sides of my intent, but only
Vaulting ambition, which o'erleaps itself
And falls on th'other

DM

Steven Boykey Sidley is a professor of practice at JBS, University of Johannesburg and a partner at Bridge Capital and a columnist-at-large at Daily Maverick. His new book, It’s Mine: How the Crypto Industry is Redefining Ownership, is published by Maverick451 in SA and Legend Times Group in UK/EU, available now.