Artificial Intelligence – Bubble Scenario pt. II

The present is rhyming more with the past.

• Three ground-breaking IPOs, a massive equity offering and record-breaking debt issuance demonstrate that the AI industry is moving to address the seemingly insatiable demand for paper, making the AI Boom look more like the Internet Bubble with every passing transaction.
• One of the reasons why internet stocks ran up so aggressively in 1999 and 2020 was that there was not enough equity available to meet demand, and so the price of the paper went sky high.
• The internet industry started issuing paper in the form of IPOs to meet that demand, and after a while, the demand did not materialise, and the market collapsed, with many businesses going bust and the Nasdaq 100 falling by about 80% over the following two years.
• The AI Boom has one saving grace, which is demand, but this has to be profitable to avoid a collapse, and here there are question marks.
• The last three years have seen a shortfall in the supply of AI-related paper, but in 2026, it very much looks like this shortfall is being more than addressed.
  • First, 3 gigantic IPOs: where SpaceX will be first out of the gate, raising a record-breaking $ 75 bn – $86bn, giving the company a valuation of $1.78tn.
  • The valuation is also outlandish, with a 2026 EV / Sales of roughly 75x, but to be fair, the company is unique.
  • I have been positively surprised by the economics of putting data centres in space (see here), and if the other problems can be overcome, SpaceX will be in a class of one.
  • However, even with this, this IPO leaves very little on the table for investors.
  • By being first, SpaceX is likely to see the best demand, meaning that both Anthropic and OpenAI, who are also likely to ask for crazy valuations, will have to contend with what is left.
  • I think that Anthropic may do quite well, especially as my estimates indicate that it should be making money by now, but I see a high risk that the market refuses to pay for OpenAI, which could trigger a valuation reset.
  • This would take the pressure off of the race to be first to super-intelligent machines, causing growth to slow or reverse and the share prices of everyone involved to correct.
  • Second, Google secondary offering: where Google is now raising $85bn to pay for its AI investments, which is $5bn more than anticipated.
  • This is a clear signal that the supply of AI-related paper still lags demand, and this deal is greatly supported by Berkshire Hathaway taking 10% of the new paper, lifting its holding to $32bn.
  • Berkshire has the reputation of being the ultimate value investor, and to be fair, Google’s valuation does not look that stretched until one looks at free cash flow, which has fallen substantially as a result of capex.
  • In my opinion, equity issuance is better than debt as equity does not have to be paid back, but existing shareholders are going to be diluted as a result of this deal.
  • Third, debt issuance: where Morgan Stanley estimates that $450bn of debt will be raised in 2026 to pay for AI investments in infrastructure.
  • This is highly problematic because if the economics of AI compute do not improve, there will be no money with which to pay back the debt.
  • This is precisely what happened in 2001, and as debt levels and paper supply rise, the more the AI Boom looks like the Internet Bubble.
• The one thing that separates the two “bubbles” is demand, and this is crucial to the debate.
• The Internet Bubble popped because demand did not show up to make use of the infrastructure that had been built, and as a result, the debt could not be paid back, and there was a large valuation reset.
• Given the scale of things today, this now looks like a blip when one looks at the Nasdaq 100 chart, but those that ignore history are often doomed to repeat it.
• The difference with the AI Boom is that demand for AI compute remains very far above the industry’s ability to supply it, meaning that demand is not the problem.
• This time around, it is the profitability of the compute that is being sold that is the problem, and the vast majority of the industry appears to be selling compute at break-even.
• Break-even is not nearly enough to pay back the debt or satisfy investors who at some point will ask where their returns are.
• This is why it is the profitability of compute that I am looking at very carefully, as this will determine whether we are at the dawn of a new age or the top of another investment bubble that typically precedes a new age.
• I have no idea when the market is going to care, but at some point it will, and at that time those still losing money will be severely punished by the market.