It’s getting pretty tough keeping all these AI deals straight

The role of “customer” and “investor” are usually pretty distinct: customers buy products. Investors provide money companies need to make products.

But it’s not always this clear, especially if you happen to be one of the big AI companies.

Just look at this morning’s megadeal between Advanced Micro Devices and OpenAI, in which AMD agreed to provide multiple generations of its Instinct processors to OpenAI — the fast-growing, but cash-incinerating, maker of ChatGPT.

Traditionally, when a company agrees to “provide” something, the entity that gets that something provides something in return, typically money.

However, in this deal, the putative buyer of the chips, OpenAI, seems to be the one getting compensated. AMD will issue warrants for up to 160 million shares of its stock — structured as it reaches certain milestones — and fork them over to Sam Altman’s firm. In theory, that would be enough for OpenAI to end up with a 10% stake in AMD. So in this case, is the customer is also becoming an investor?

Wait, it gets more confusing, because The Wall Street Journal reports an unattributed fact of some importance: that OpenAI “will buy the chips either directly or through its cloud computing partners.” In other words, OpenAI might not be the one buying these AMD chips.

This makes sense. OpenAI does not actually currently have a business that could be expected to generate tens of billions of dollars to buy AMD chips over the next few years, which AMD executives said this deal was supposed to do.

(OpenAI’s revenue, according to a report in The Information, is on track to be just $12 billion this year. The company is also making large losses that put it on track to burn through $115 billion through 2029.)

So does that mean Oracle, which has likewise signed an enormous deal with OpenAI to provide it with data center infrastructure, will actually be the one buying the chips? If so, will OpenAI still get the warrants if it isn’t the corporation writing the check?

And doesn’t it matter to anyone that AMD has potentially just given away 10% of the company? The warrants for OpenAI are priced at $0.01 a share. That ownership stake itself was worth “tens of billions” before the deal was announced — roughly $27 billion.

Apparently it does not!

The market loves this deal like a Labrador retriever loves a fresh new tennis ball. Advanced Micro Devices shares soared by more than 25%, the most since early 2016, creating $75 billion in market value.

But while the deal seems to make sense to the market, there is growing discomfort among Wall Street analysts about the recent spate of deals that companies have signed with OpenAI, even if they’ve generated sometimes massive market gains.

It was essentially an announced deal between OpenAI and Oracle, in which OpenAI agreed to buy some $300 billion in computing power from Oracle — OpenAI does not have this $300 billion — in the coming years, that lit the fuse on Oracle’s 36% price surge on September 10.

That surge created more than $250 billion worth of stock market wealth in a single day.

“We need to start being cautious about the promises OpenAI is making all over the place without being able to really have the capital to fulfill those promises,” tech analyst Gil Lauria, of brokerage firm DA Davidson, said last week during a discussion on the “Prof G Markets” podcast.

And on Monday, analysts at Goldman Sachs issued a note on Nvidia saying that its deal to invest some $100 billion into OpenAI, along with other deals, “have sparked investor debate around the nature of the deals and the extent to which Nvidia’s equity investments could be recycled by investees as GPU spending, recognized by Nvidia as ‘circular’ revenue.” They wrote:

“When equity investment comes from a supplier, we believe additional scrutiny is warranted given the ‘circular’ nature of the revenue because of the investor’s dual role as investor and supplier.”

Elsewhere, RBC software analyst Rishi Jaluria recently wrote an interesting note spotlighting “the growing interconnectivity and potential ‘round tripping’ of revenue” with Oracle, Nvidia, and OpenAI.

“I’ve obviously got a lot of questions about those sorts of deals,” Jaluria said in a phone interview. “The commentary people make is, you know, Oracle is buying business and this is all just circular, and it’s actually doing nothing valuable.”

But he argues that these deals, if they do succeed in supercharging the AI industry by enabling bigger, faster build-outs of data center infrastructure, may actually create significant value.

“If it purely is money going from, you know, Nvidia to OpenAI to Oracle back to Nvidia and nothing else, then 100%, that would be pure round-tripping of revenue and a closed loop system,” he said.

“However, if we actually start to get a result out of this, and OpenAI is able to develop better models faster,” he continued, “that’s where this benefits the broader global economy, even if it’s happening in this smaller sort of loop.”

The reason folks on Wall Street are so concerned about such loops is because they have played roles in a few market disasters in the past, from the collapse of the US energy trading market in late 1990s to the implosion of AOL Time Warner a few years later. And the practice tends to emerge in superheated markets where market prices become heavily dependent on maintaining superfast rates of revenue growth rather than profitability, a familiar environment for those following the AI industry.

Seasoned market observers seem to remember. On Monday, hedge fund manager Paul Tudor Jones told CNBC, “The circularity makes me nervous,” when asked about the dynamics of the AI data center build-out.

Other investors remain concerned as well. Jaluria says he tells those who’ve called that he has sometimes raised his eyebrows on the structure of these deals.

“I’m like, look, I get it. I get the criticisms. And that was probably my first instinct as well,” Jaluria said. “There might be some merit to that.”