It is still early days, but things are not looking rosy for AI
Michael Roberts is an Economist in the City of London and a prolific blogger.
Cross-posted from Michael Roberts’ blog
It seems that Asian AI bubble has burst. It all started in South Korea with the KOSPI stock market index plunging over 8% on 15 October. That just added to a huge fall in the Korean index of 41% from a peak of the Korean AI investment boom last June. The share prices of the two large Korean tech companies, SK Hynix and the mega Samsung Electronics, fell over 10% on that one day.
Now the bursting of the bubble has spread to China. The STAR50 stock index, tracking key AI and chip firms, has dropped 40% since its July high, while the broader CSI 300 entered bear market territory after a 20% fall from June. When placed alongside Seoul’s $900 billion seven-day wipeout on October 15, nearly $1.4 trillion in ‘fictitious capital’ has been vaporised across the East Asian hardware supply chain in October alone. By fictitious capital, Marx meant the market value of stocks and bonds, which are claims on the value of real assets like machinery, software and components etc that companies have. It’s fictitious because it can be well end way out of line with the real value of assets owned and employed by companies to make profits.
In Asia, it has been retail investors (that’s households) who have been pitching into the stock market. Household savings rates are high, especially in China, and the previous investor boom in Chinese real estate collapsed years ago. Buying AI stock is the latest craze to make a fortune. But now it seems that investors have decided that the price of investment in AI-driven Asian companies is well over the top.
SK Hynix and Samsung are key companies that supply the the semi-conductor chips required to run advanced AI accelerators and data centres. For the past two years, Samsung and SK Hynix appeared to be making huge profits on chip sales exploiting extreme global demand for chips from cash-rich US tech monopolies badly needing more server memory for their AI projects. But these super profits now appear in jeopardy, especially after China’s CXMT unveiled a data centre server memory dramatically reducing the cost of chips. Korean stock market investors saw the writing on the wall – no more super profits.
CXMT’s break into the Korean share of the chip market doubled its global DRAM market share from 4% to 10% year-on-year in Q2 2026). But that had its own consequences for the Chinese stock market. Chinese and Asian AI labs have shifted towards so-called “open-weight” models (free to use). Falling chip costs has also meant falling prices for AI companies in China, where the competition has been intense. Capital had flooded so much into chip design, memory, and robotics that profit margins are disappearing.
So China’s speculative investors have got worried and CXMT itself has lost more than 25% of its stock market value. According to Goldman Sachs, AI-related stocks had swollen to 42% of China’s total equity market capitalisation and 57% of total trading turnover across China mainland A-shares and Hong Kong H-shares. Even after losing 40.4% of its value, the STAR 50 still trades at roughly 100 times trailing earnings (down from nearly 250x at its June/July peak), more than triple the US Nasdaq tech stock index.
As commentator Saad Zaidi explained, China’s rapid horizontal replication of AI hardware capacity has compressed profit margins across the entire East Asian supply chain, with loss of pricing power amid overcapacity. Caught in a pincer movement between faltering returns on downstream AI capital expenditure on one side, and China’s state-directed tech industry lowering profit margins on the other, the speculative bubble in East Asian hardware is visibly fracturing.
As Zaidi says, this follows Marx’s law of profitability, namely that any rapid rise in investment in technology will eventually weigh down profitability, unless it drastically raises the productivity of living labour across the wider economy. And so far, as I have shown before, total factor productivity (economic gains from AI innovation) have hardly risen. “The market is now waking up to the reality that upstream hardware valuations cannot stay detached from downstream profitability forever.” (Zaidi)
As I have shown before, this AI mania mirrors that of the 19th century investment in burgeoning railway technology. Then railways were also seen as a powerful new technology that could transform transport and travel, and so boost productivity. This led to massive speculation in railway shares as company after company launched a new rail lines across Britain during the 1840s and in the US in the 1870s. Eventually, investor concerns about overcapacity rail projects, along with increasing distrust of railroad securities, triggered crashes.
In the UK, the mania reached its zenith in the 1840s. Marx noted then that, “in the railway swindle from summer 1844, railway investors apparently expected much higher than the average profit rate.” Marx called it the ‘great railway swindle’ to emphasise that the claims made by the railway companies for huge profits to be made were deliberately overstated (shades of US AI companies now). Investors naïvely poured capital into schemes that were significantly less profitable than they expected and and rail company directors had promised.
The question now is whether investor fears about overinvestment and declining prospects for profits from AI will spread to the heart of the AI mania: the US. Competition from China’s free or very cheap to use and every improving AI models is eating into the potential profits of the US AI labs, OpenAI and Anthropic. All top five most popular models are now Chinese and, according to a WSJ report published last week, “approximately 45% of all tracked tokens now flow through Chinese models.”
As Torsten Slok at Apollo Capital has pointed out, despite massive investments and widespread hype around artificial intelligence, the technology has delivered zero financial payoff to corporate America outside of tech. “The bottom line is that the AI story will be decided in the S&P 493, not the Mag 7, and so far, there is no evidence that revenue growth or margins are improving outside the tech sector.”
And the profitability of the mega hyperscalers that are driving AI investment has been falling: in return for over a trillion dollars of capital expenditures they have seen their return on invested capital drop on a quarterly basis.
And there are signs that the US AI investor bubble may be leaking. Yesterday, the Alphabet (Google) share price fell 10% after it boosted its AI spending to more than $50bn in the third quarter, leading to a record ‘cash burn’ that spooked investors. Free cash flow (ie money available to invest) fell to minus $9.1bn in September compared to minus $5.9bn in the prior quarter. To back its spending spree, Alphabet has taken on tens of billions in debt and in June announced plans to raise about $85bn in its first share sale in more than two decades — a sharp reversal after years of buying back its own shares.
But the big test is coming: Anthropic is about to release shares to the public in order to raise $150bn, making the company’s market value equivalent to $1.5trn – a truly staggering market value for a company that makes no profit yet. As Anthropic put it in the prospectus to potential share holders: “We are at an early stage of our revenue growth, we have experienced rapid growth in recent periods, and our historical growth rates are not indicative of future performance or our ability to sustain or forecast future revenue growth. We do not expect our revenue will continue to grow at comparable rates, and we cannot assure you that it will continue to grow at all.”! It’s one big bet.
Be the first to comment