Bandwidth Then, Data Centers Now?

One of the questions I’m asked most frequently is whether the current market environment reminds me of the internet bubble of the late ‘90s.  My short answer is “history doesn’t repeat but it often rhymes”.  The enthusiasm for all things related to artificial intelligence is akin to that of the dawn of the internet, though many of the circumstances are quite different.  That said, I am wondering if the frenzy to build ever-larger data centers is reminiscent of the race to add bandwidth in the prior era.

This is something I was pondering a few days ago when Oracle (ORCL) announced a huge boost to forward guidance that was largely propelled by a $300 billion, five-year data center commitment from OpenAI.  When we first wrote about it on the morning of September 10^th, I was not yet fully aware that the lion’s share of the guidance stemmed from that specific deal.  As I noted in an interview later that day,

… [Y]ou are correct in pointing out the risks inherent in the market’s complete revaluation of Oracle… Not only are Oracle stockholders crucially dependent upon the company meeting its guidance, but the broader market is, too.

This is indeed a monstrous commitment from a company (OpenAI) that burns prodigious amounts of cash and was reported in May to expect losses of $44 billion before turning profitable in 2029.  Remember, that was before they committed to spending $300 bn with ORCL.  Spending of this magnitude can only be achieved if investors are willing to finance it.  So far, that has hardly been a problem – investors have been eager to get a piece of the action from OpenAI and other companies in the AI space.  But it became one in 2000 when investors began to concern themselves with profits over promise. 

I was reminded of this today somewhat by chance.  I happened to be awaiting my turn for a media appearance when it turned out that the guest before me was the CEO of CoreWeave (CRWV).  His stock has been a phenomenon since going public earlier this year, but it too is a money burner.  On a rolling four-quarter basis, the company lost over $900 million, and analysts project losses for at least the next two years, though at smaller magnitudes.  Yet I believe that if CRWV announced a secondary offering tomorrow, it would be lapped up by eager investors. 

Will that be the case forever, especially after that CEO acknowledged that AI infrastructure will require trillions in investment?  Can we assume that investors will be perpetually willing to finance investments of that magnitude even if the near-term returns are scant?

I recall something similar with bandwidth during the internet rollout.  There was a huge rush to wire the country and the world for the coming internet era.  That bandwidth eventually proved necessary and profitable, but it took more time and money than most investors expected during the heady rally.  It also consumed some of the biggest names at the time.  Two of them proved to be outright frauds – Enron and Worldcom –but others like Global Crossing and Northern Telecom also fell by the wayside.  There was huge overinvestment that needed to be reckoned with – expensively.

Now, far be it from me to accuse any company of being an Enron-style fraud.  I have no evidence of that, nor do I want to portray even the slightest hint of casting that type of aspersion.  But the AI-ecosystem is tightly bound, with Nvidia (NVDA) as the nexus.  Their largest customers include names like Microsoft (MSFT), Amazon (AMZN), Alphabet (GOOG, GOOGL), and Meta Platforms (META), all of whom are indeed highly profitable and currently able to sustain a high level of investment into AI projects.  But will they want to do so indefinitely?  NVDA in turn is a huge customer of CRWV and Super Micro Computer (SMCI).  If there are cutbacks in NVDA spending from the big boys, that will spill back into the smaller beneficiaries.  There is more fragility than the market might recognize.

While I sincerely hope that there are no Enrons or Worldcoms lurking out there, there very well might be some Global Crossings or Northern Telecoms lurking.  It’s all great while the momentum is favorable, but it can turn ugly very quickly if not.  And for now, as I literally just saw a headline that NVDA plans to invest up to $100bn in OpenAI, the music is still playing – loudly.