When ChatGPT went live on November 30, 2022, Nvidia was best known for gaming chips. The company was worth $420 billion and Jensen Huang under $20 billion. This month, Nvidia closed at a record $5.77 trillion, and Forbes puts Huang at $199 billion.
The builders of Babel, Genesis 11:4 records, set out to “make us a name”. Today, billionaire lists are where names get made. Forbes’s 2023 annual census of the Richie Riches counted 2,640 billionaires worth $12.2 trillion. The 2026 edition found 3,428 worth over $20 trillion. Elon Musk, Larry Page, Sergey Brin, Jeff Bezos and Mark Zuckerberg went from $514 billion between them to $1.78 trillion.
The divergence between AI and not-AI tells the post-ChatGPT story. Value exchange-traded funds have struggled returning about 47 percent in three years, whereas the VanEck Semiconductor ETF returned 327 percent. Below the billionaires there is the salaried millionaire gentry. At OpenAI, more than 75 employees sold $30 million each in one tender. SK Hynix is on course to pay its average worker $547,000 for 2026. In Bengaluru, India’s five biggest IT firms shed 57,891 jobs in two years. Icheon gets rich; Bengaluru gets thin.
India’s family offices are funding homegrown AI, but it has yet to show up on the global radar, and fund managers lament India’s underperformance and thin AI presence. Distance from AI costs money. Bernard Arnault, the richest man alive in 2023, is poorer than he was. In 2023, the richest man in China sold bottled water. Zhong Shanshan is still worth $68 billion. ByteDance’s Zhang Yiming has gone past him, from $45 billion to $69.3 billion. Tencent’s Ma Huateng has climbed from $35.3 billion to $53.8 billion.
Wall Street has always toned up its walk and talk for the season: Atkins and power yoga in the dot-com years, the Peloton through the pandemic. This era’s diet is a weekly GLP-1 jab that kills the appetite, though not for AI-adjacent stocks. The with-it crowd has moved from Zone 2 cardio to tai-chi, which seeks stillness in motion. The talk has followed the body into verbal calisthenics. Huang says compute is now revenue. Alphabet spends and warns, in one breath, that cash flow will stay under pressure. In March, with Nvidia down 11 percent for the year, the question was whether AI would pay. It is back in circulation.
Consensus has it that AI is real. There is no consensus on whether the valuations are. The result is babel about the bubble. This week the IMF warned AI could threaten financial stability. It counted 14 publicly disclosed circular deals worth $1.4 trillion and flagged “wrong-way risk”: the report warned that stakes that lose value just as the revenue tied to them is impaired. At best it signals scepticism; at worst, trouble ahead.
Gravity arrived soon enough at the stock markets. The S&P500 notched its worst week since June and the Nasdaq fared worse. The core of the AI build-out, the weave that ties Nvidia chips to labs, is stitched with borrowed money. The fabric has begun to unravel. The fall of CoreWeave symbolises what could be—its bonds have sunk below 80 cents on the dollar, yields are at 15 percent and its shares tumbled 17 percent in four days. Bonds of data centres leased to Meta and Microsoft slipped and Nvidia shed about $270 billion of market value in a week. The index is still near its record, and opinion is stranded between wobble and correction, between correction and crash. John Templeton said memorably, “Bull markets are born on pessimism, grow on scepticism, mature on optimism and die on euphoria.”
There is rationale scaffolding the AI story. Gordon Moore saw in 1965 that chips would keep doubling in capacity; the corollary was falling cost. Stanford’s AI Index says running a model with GPT-3.5-level capability fell from $20 per million tokens in November 2022 to $0.07 by October 2024. Jevons named the paradox in 1865: the more efficiently Britain burned coal, the more coal it burned. Ditto with tokens. ChatGPT had 900 million weekly users by February. Nvidia booked $215.9 billion in revenue in the year to January; Anthropic was billing at more than $30 billion a year by April. Somebody is paying these invoices.
The earnings are visible. Are they enough? Gartner reckons AI data-centre owners need $8.5 trillion of revenue by 2030 to cover their costs, and forecasts no more than $3.6 trillion. OpenAI has pushed its listing to next year, after its annualised revenue was pegged $20 billion short of what it had signalled. Nvidia held $99 billion of stakes in AI companies by late July, against about $7 billion a year earlier.
At a macro level the promised land of productivity comes at a cost. Loss of jobs equals loss of consumption and growth. The ILO estimates unemployment among the 15-24 age cohort is at 12.4 percent. It is over 9 percent for the youth in the US, 16.1 percent for the 16-24 group in China and 9.9 percent of the 15-29 cohort in India are out of work. The white-collar worker forms the spine of the middle class—in India the IT is the largest exporter, the largest formal employer and a multiplier of consumption. AI is already busting business models.
Jules Winnfield, in Pulp Fiction, quotes Ezekiel 25:17 on “the inequities of the selfish”. Ezekiel did not quite say that; Quentin Tarantino wrote most of it and probably meant iniquities. The slip fits. A growing share of retirement savings sits in index funds, and 10 companies make up 37.8 percent of the S&P500. The consequences will be K-shaped either way. If it isn’t a bubble, the gains stay with a few. If it turns out to be a bubble, billionaires lose paper; everyone else feels the pain.
Read all columns by Shankkar Aiyar
Shankkar Aiyar
Author of The Gated Republic, Aadhaar: A Biometric History of India’s 12 Digit Revolution, and Accidental India
(shankkar.aiyar@gmail.com)
