The AI bubble and it’s possible disintegration gives us some pretty grim insights into the question of what happens when the richest people in the world rebuild the economy in a way that it no longer needs working people, but still expects them to pay their bills.
Reuters reported this week that China’s AI and semiconductor stock boom has suffered a sudden and staggering reversal: their STAR50 index of Beijing’s AI and chip champions (like our NASDAQ) has plunged 40 percent in less than four months, wiping out roughly $480 billion in market value.
Their robotics company Unitree, whose IPO just three months ago was oversubscribed 8,000 times by Chinese small investors and has lost half its value. Dong Baozhen, a Beijing fund manager, told Reuters:
“And now we see the bubble bursting.”
The same day (Monday) here in America Microsoft offered voluntary buyouts to 60,000 employees, about 27 percent of its global workforce, as it’s planning to replace them with AI. And when Anthropic (Claude) filed to go public they disclosed an $80.7 billion net loss and some $657 billion in infrastructure deals with companies including Microsoft, Amazon, and SpaceX; fully $498 billion of that is noncancelable even if the company never uses the computing capacity it’s reserved.
Things got even weirder on Wednesday, when Microsoft and Alphabet reported some striking “earnings” numbers. Alphabet burned through $9.1 billion in free cash flow for the second straight quarter. NPR’s 1A tallied up Monday’s AI deals alone at roughly $235 billion, nearly one percent of American GDP, all in just one single day.
While Wall Street has been pouring massive sums into AI companies, data centers, chips, and power infrastructure, Economist Dean Baker reports at his weekly AI Bubble Monitor that CoreWeave’s debt (the company rents computing power to the AI giants) is now considered so deep in junk bond territory that they’re having to pay 16% interest on their long-term debt to get investors to bite.
In 1965, Intel co-founder Gordon Moore postulated that every two years the number of transistors in an integrated circuit would double. Not only was his “Moore’s Law” right, but as the power of digital hardware increases, costs also reliably drop by a similar factor.