What a Chola-era Dam Can Teach Us About AI’s Future

Somewhere around the second century CE, near present day Thanjavur, a Chola king named Karikala put thousands of labourers to work sinking unhewn stone into the sandy bed of the Cauvery to make a dam, in a hope that it could supply water to his kingdom. To everyone’s surprise, no mortar was used to hold it together; there were only rocks laid against rock for more than three hundred metres, across a mighty river that flooded every year without fail.

I’ve often wondered what his ministers muttered in private. Though the Chola empire was one of the richest in ancient India, its treasury was not unlimited, and it was being emptied into a delta that was still mostly scrub and wilderness. There was no farmland in existence near that delta and any promise for great harvests was a delusion. Any prudent and senior adviser in the Chola court would have called it recklessness dressed up as a grand vision.

Nineteen centuries later the Kallanai dam is still stands. It controls the river’s floods and provides irrigation water to lakhs of acres of land in the delta districts into what became the rice bowl of South India. The great Karikala never saw most of the returns on his stone; his grandchildren's grandchildren were the main beneficiary and collected grand returns.

While Karikala Chola built the original structural core around c. 150 CE, the Kallanai we see today is the result of continuous enhancements. It was reinforced, raised, and remodelled over generations, most notably by British engineers like Captain Caldwell and Sir Arthur Cotton during the 19th century.

Not a Bubble

I keep coming back to that dam every time someone tells me the AI boom is just like the Internet bubble, only with better graphics. Critics argue that the AI boom resembles the dot-com bubble: technology firms may spend about $635 billion on AI infrastructure in 2026 while AI software companies like OpenAI, Anthropic and the rest, generate only about $35 billion in revenue. To draw a rough analogy, its 18 dollars of stones poured in river for every dollar of rice harvested, so to speak.

American billionaire investor Ray Dalio says today’s AI investment levels look like the dot-com era to him, which anyday will burst and will impact millions of small and big investors.

Michael Burry, the man who called the 2008 housing crash, has taken fresh short positions against Nvidia and Palantir, and he describes the circular funding among the big players as “a picture of fraud, not a flywheel.” The Bank for International Settlements, the institution that banks the central banks, has warned that the bubble could pop and drag the world economy down with it.

These are well-built cases for the prosecution. But I think they fall short on three counts. Firstly, in 1999, the money came from IPO proceeds and retail speculation, poured into companies that had never earned a profit and never would. Pets.com lost money on every bag of dog food it shipped, and the market cheered anyway. ‘Ask Jeeves’ popularity peaked in the pre-Google era, owing largely to its novel “Jeeves” persona, an online butler who would fetch users the information they’d requested.

Launched in 1996 and went public in July 1999, it tanked from $190 to $0.86 per share by 2002. It tanked because the founders never had a pragmatic monetization plan in a world where digital ecosystem (Smartphones) and digital supply chains were not invented yet.

Today’s spenders are Microsoft, Alphabet, Amazon and Meta, four of the most profitable enterprises ever built, funding their data centres largely out of operating cash flow rather than borrowed hope that one day they will be able to make profit.

Nvidia’s quarterly revenue rose 85 percent year-on-year to about $81.6 billion, driven by a 92 percent jump in data-centre sales. While its valuation may be debatable, its cash flows certainly are not.

Secondly, the dot-com crash arrived while the US Federal Reserve was raising rates hard, draining money out of markets at the precise moment the Nasdaq-100 was trading near sixty times forward earnings.

Today the vector points the opposite way. The US Fed has cut rates three times since September 2025, bringing its target down to 3.50 to 3.75 percent, and technology valuations, stretched as they are, sit at less than half that 2000 peak multiple. Could stocks still fall from here? Of course. But the trigger that detonated 2000, tightening money colliding with delirious pricing, simply isn't loaded right now.

Visionary Infrastructure

The third count is the one Great King Karikala would have recognised. When the dot-com bubble burst, telecom firms like WorldCom and Global Crossing went bankrupt sitting on millions of miles of ‘dark fibre’ laid for internet traffic that hadn’t yet arrived.

Bought for pennies out of bankruptcy courts, those same cables went on to carry broadband, YouTube, and the cloud on which today’s AI models run. Even the waste of that bubble became the plumbing of the following two decades.

Viewed through that lens, today’s spending looks different. The money is flowing into durable assets - data centres, power capacity and advanced chips - that support a technology already embedded in everyday work. By 2025, McKinsey found that 76 percent of employees were using AI at work, up from 30 percent just two years earlier.

The dot-com boom was fuelled by cheap money and IPO speculation. AI investment, by contrast, is following adoption rather than anticipating it.

In 2026, the retail and commercial usage is running ahead of the infrastructure being built to serve it. That inversion is the single most under-discussed fact in this whole debate. Should one conclude that nothing can go wrong? No. Cisco earned genuine, growing profits in 2000, and its shares still fell 85 percent from the peak.

Burry’s point about circularity also deserves a straight answer rather than a dismissal. When a chipmaker invests billions in a model company that turns around and spends those billions on the chips, revenue and investment start chasing each other’s tails, and OpenAI’s commitment to spend $1.4 trillion on data centres over eight years, against $13 billion of revenue, is being financed substantially with debt. If credit markets seize up, that chain gets tested and definitely some links will snap.

If you’re an Indian investor reading this over your morning tea, here’s a number worth keeping in your pocket. When the Nasdaq-100 collapsed 75 percent after 2000, the Nifty fell 39 percent, and in the recovery from 2003 to 2007 the Nifty returned 557 percent against the Nasdaq’s 88. Distance from the epicentre changed everything for the Indian saver last time and it might again.

A bubble is a price with no cash flow underneath it, and by that definition this isn’t one. While AI has plenty of froth, some of it absurd, it is riding on top of real technology and its demand. Profitable firms are still financing durable infrastructure with real earnings. And while some fortunes may vanish, the build-out will remain for the next generation of advancement.

The Cauvery flooded every single year of Karikala’s reign, and every year, I imagine, someone at court waited for the king’s folly of stone to wash away. It never did, because the stone had been laid to work with the flood rather than to defy it. Two thousand years later, the builders of AI infrastructure are making a similar wager, while sceptics once again predict ruin.

The next time someone tells you it’s 1999 all over again, ask them who’s paying for the build-out, and with whose profits? The answer will tell you whether you’re looking at a mania, or at a dam.

(The writer is a strategy and transformation leader. He can be reached out at abhisheknjain5@gmail.com. Views personal.)