AI's Potemkin Profits

It’s very easy to get attention in business news by printing a headline with words to the effect of “Leading Investor X Warns of Impending Recession”, “InvestWithUs CEO Thinks Current Market is A Bubble”, or my personal favorite: “Michael Burry Just Sold All His Stocks.” The fact you’re reading this is somewhat proof of this. 

In a shockingly high number of cases, these events never materialise, at least not in the way they were predicted to, and we’re rarely called out on our mistakes retrospectively. But while there will forever exist analysts willing to sell market doom and gloom, it doesn’t mean we should always turn a blind eye. 

The “magnificent 7”, the cohort of leading AI infrastructure and investment firms, have been responsible for 75% of the gains made by the S&P 500’s rally over the last few years. So, we all know AI is a hot product, but when Nvidia recently announced an eye-watering $100 billion investment into OpenAI – one of its own customers – then alarms began to sound. 

Telltale signs of bubbles, however much of a crapshoot it may be at assessing the size and nature of the bubble, do exist. And they often are observed through two angles: the quantity of money involved, especially investment on the back of predicted future demand, and the deliberately engineered obfuscation of where the money is actually going and coming from. 

I can only give a few simplified examples of the big movements. Any more granular detail and even I begin to lose the plot. 

Nvidia, now the world’s most valuable company by a country mile thanks to the AI rally, has invested in OpenAI, now to the tune of $100 billion. On the back of this investment, OpenAI has pledged to build data centers powered by Nvidia’s chips. Jensen Huang – Nvidia CEO – insisted that OpenAI is not obliged to buy Nvidia tech with the money Nvidia gave them. Alright. But OpenAI is not the only customer Nvidia has given money to. 

A chart pieced together by one of Morgan Stanley’s analysts illustrates some of this complex web neatly.  

Source: Morgan Stanley, through Barrons 

Despite its colours, it’s still an absolute eyesore. 

A lot of this might be best described as Vendor financing (depending on the legal agreement). It's technically an entirely legitimate form of business. That is, a company lends to its customer, often with the intention to finance that customer’s purchases from that company. But such agreements make it increasingly difficult to understand how much value is actually being put toward the underlying product, which, crucially, is not Nvidia’s chips. The underlying product is the AI software. You need to ask: if AI is ready to capture all this value, then why is Nvidia essentially selling its product at a discount?

An extract from this Fortune article articulates the justified worry of a bubble rather well:

“According to a story in The Information, Nvidia agreed this summer to spend $1.3 billion over four years renting some 10,000 of its own AI chips from Lambda, which like Coreweave runs data centres, as well as a separate $200 million deal to rent some 8,000 more over an unspecified time period. Those Nvidia chips Lambda is renting time on back to Nvidia? It bought them with borrowed money collateralised by the value of the GPUs themselves.” 

Good grief. Admittedly, for an American economic appetite, a few billy here or there is a drop in the ocean. But arrangements such as Lambda’s are clear signals of froth. To get a better sense of the hype and scale, I’d encourage a search on Thinking Machines Lab, a company from ex OpenAI CTO Mira Murati. It obtained a record-breaking $2 billion for its first seed round. It did not have a product. 

These so-called “hyperscaler” AI ecosystem firms “are now spending close to $400 billion annually on capex supporting AI infrastructure”, that is, data centers. We spent less than the equivalent in inflation-adjusted dollars on the entire Apollo programme over a decade. Morgan Stanley researchers estimate this alone has contributed to a whole percentage point of quarterly GDP growth – blowing the actual rate of consumer spending growth out of the water.

But no better example of the hype exists than the Oracle deal. In this case, superlatives just don’t work. A week before the Nvidia deal, OpenAI struck a $300 billion commitment with Oracle, another data centre and cloud computing giant, for its services.

Put another way: OpenAI, who only turns over about $12 billion a year, has pledged (but not guaranteed, mind you) $60 billion every year for the next 5 years to a company who does not have the money or time to build the data centres it would need to fulfil that contract. 

The $300 billion takes the form of an RPO, or remaining performance obligation, which is not guaranteed cash for Oracle. The deal is only set to begin in 2027. OpenAI doesn’t turn a profit. Oracle currently does not have the capital to service the data centre capacity required for the contracted services: it is going to have to borrow to build them. Oh, and one of the key reasons Oracle was able to secure this deal with OpenAI? Oracle ramped up purchases of Nvidia’s GPUs. 

You can read that all again if you like.

Computing ability fuelled by Nvidia’s chip deals is no longer a proportional proxy to the actual value that AI can currently add. The moment debt payments cannot be made, there will be questions to answer. If Morgan Stanley Wealth Management’s CIO begins her day by opening her Bloomberg terminal to check CDS spreads on Oracle debt, then we do need to pay attention. This is the other side of bubble town: investment is overcommitted. 

People see bubbles at risk of “bursting” when the firms who receive oodles of cash cannot make clear exactly how they will develop a sustainable enough competitive advantage to earn that investment back. In this case, for zero economic re-adjustment to take place, it would require most of the major AI ecosystem to figure out how to correctly and fully utilise the money they have given to develop a sustainable revenue with AI’s current capabilities.  

From a macro lens, inflation, jobs, and interest rates will have a negligible impact on our ability to assess the weight of the financial anvil we’ve placed ourselves under. Whether the bubble pops catastrophically or not is up to your doom and gloom analyst to forecast, but investors and onlookers alike are craving – and deserve – a serious, transparent readjustment.

Illustration: Will Allen/Europinion