The trillion-dollar question
Artificial intelligence (AI) has become the market’s biggest investment theme, with strong earnings and seemingly reasonable valuations convincing many investors the rally still has further to run.
But beneath the surface, several warning signs suggest the picture may not be as strong as it appears.
Big claims, bigger questions
Elon Musk says SpaceX’s Total Addressable Market (TAM) in Enterprise AI is US$26.5 trillion. Outside estimates suggest $US50-$300 billion. This leaves the ‘buyers’ and the ‘sellers’ with some important questions.
Sellers: SpaceX insiders have invested a total of US$11 billion since the Series A in 2002. With the valuation now at US$2 trillion, do you reckon they’re thinking of selling a some shares?
Buyers: SpaceX has been paraded before investors since its Series A funding round in 2002. If hedge funds, sovereign wealth funds, and pension funds had the chance to invest at much lower prices but declined, why would they now buy at US$2 trillion?
Even AI pioneers are cautious.
Yann LeCun; a pioneer of AI, founder of AMI Labs, and former chief AI scientist at Meta, criticised the business models and technology of major AI firms, warning of a potential industry correction. He specifically highlighted Elon Musk’s xAI as a company that might encounter significant issues, saying, “xAI is kind of a failure, frankly, because the founding team has departed.”
The economics don’t add up
Even though the cost of running AI systems is falling, it’s not falling fast. At the same time, the price AI companies need to charge to make a buck has to rise, but customers won’t stomach it. The consequence is large ongoing losses, meaning our enjoyment and use of AI tools is being funded by investors.
When an AI hardware producer records a sale they book huge revenue and profits. When their customers record the purchase, however, they book it as capital expenditure (capex) rather than as an expense. So aggregate earnings for the AI group goes up, but cash flow deteriorates. Aggregate free cash flow among the hyper-scalers is now collapsing but reported profits are surging. No wonder S&P500 margins are at record highs.
In fact, for MSFT, AMZN, GOOGL, Meta and ORCL, aggregate trailing annual cash flow was US$250 billion in Q4 2024. In Q1, 2026 it’s US$150 billion and for Q1 2027 is forecast to fall to US$25 billion.
If the ‘E’ in the price-to-earnings (P/E) is wrong, does it mean market valuations are more stretched than they appear?
And even if the modest P/Es for hyperscalers and AI players today are right, does that mean the market can’t crash? Many analysts believe that because P/Es are much lower today than during the tech bubble, the market is a long way from crashing.
But U.S. bank stocks traded at a P/E of 11-13 just before the Global Financial Criss (GFC), homebuilders traded below 10x forward earnings for most of the pre-GFC bubble, and shipping stocks traded at a P/E of 4 before they plummeted in 2022. If low P/Es are based on an ‘E’ that’s too high or unsustainable, the market can crash, and has crashed, from a low PE level.
What if the bubble isn’t in share prices at all, but in the earnings themselves?
Earnings are cyclical for many industries, including mining, retail, airlines…and…semiconductors. So, when sales go hyper-parabolic it’s rarely sustainable. Between 1990 and 2025 semiconductor sales rose from US$5 billion to US$60 billion – that’s annual growth of 7.35 per cent. Between 2025 and 2026 sales have jumped from US$60 billion to nearly US$120 billion, or 100 per cent growth in less than a year. Not sustainable. And remember, the earnings aren’t all real.