Where Comes the Sun?

Monday’s deal that involved Nvidia (NVDA) investing up to $100 billion in OpenAI got me thinking about how interconnected the AI industry has become.  It is tempting to think of the AI interconnections like the food web or carbon cycle.  But while those are enabled by plants’ ability to convert sunshine into energy, AI’s ability to turn electrons into intelligence will require money.  Lots of it.  The sun is a much more dependable, inexhaustible resource than investors’ willingness to continually provide funding. 

Just prior to that news, we pondered whether the current rush to build data centers to power artificial intelligence was reminiscent of the late-1990’s race to lay bandwidth to expand the internet.  We noted that even though the bandwidth eventually proved necessary, some of the key companies involved in that effort went out of business.  The reason for the bankruptcies was neither about the ultimate utility of the internet – that turned out to be everything we’d hoped for, if not more – nor the need for bandwidth to facilitate its growth.  It was instead about investor willingness to finance the short-term overcapacity that was being created.

It should be obvious that investors require a return on their investment.  The returns from investing in AI have proven to be phenomenal, as of course they were in the early days of the internet.  As long as the key stocks moved higher, whether NVDA now or AOL then, investors large and small were eager to participate.  Life-changing technologies don’t come around every day, so the enthusiasm was both understandable and justifiable.  No one wants to miss out, and riches have been created.

Yet we might now be at an inflection point regarding investor attitudes toward AI, with the math behind Oracle’s (ORCL) deal with OpenAI causing some reconsideration.  A $300 billion contract for computing power was indeed cause for celebration by ORCL investors.  But then we needed to face an inconvenient question: how is OpenAI going to pay for it?  According to a July 30^th article in Reuters, Open AI reached $12 billion annualized revenue in the first 7 months of the year.  That is double its prior year’s result, which is phenomenal, but a looonggg way away from the $60 billion that is committing to spend annually at ORCL.  That article also reports that OpenAI increased its cash burn projection to roughly $8bn, waaayyy above the $1bn burn that it had earlier projected.

Thus, where is the source of the necessary funding for the ORCL deal, let alone the NVDA deal?  For NVDA, some of it will come from NVDA itself, which may seem a bit incestuous, but it is not unusual for large manufacturers to finance its customers’ purchases.  Think of how car companies will finance sales at their dealers.  But ultimately, one of two things need to occur for these deals to come to fruition:

1. OpenAI needs to grow very quickly.  Growing from $12bn to $60+bn in revenues in just a few years would require some wild exponential growth – especially if 2X revenues requires 8X cash burn
2. Investors need to be willing to continue investing in OpenAI to cover the cash burn and contractual outlays

We need to wonder how long #2 can remain feasible until or unless there is a huge improvement in #1.  Taken to an extreme, we may be looking at the bubble that OpenAI’s Sam Altman mused about in August.  Less catastrophically, we should at least concern ourselves with whether investors should continue to reward companies simply for spending money on AI or announcing contracts with increasingly interconnected parties. 

In the food web, sunshine provides the necessary catalyst to make it self-sustaining.  A side effect is that sunshine is also said to be the best disinfectant.  In the AI web, money might be the catalyst, but it is hard to imagine that money flowing reliably for millennia.   Perhaps some sunlight shining on that investment cycle would be valuable too.