Is the AI Bubble Bursting?

Friday 30 October Amazon says it will spend even more on AI next year. Intel warns it will burn cash.

The Nasdaq hit its lowest level in nearly three months.

One day's damage

Intel's fall was its steepest one-day drop on record.

Wiped off the Nasdaq 100 in October alone, down more than 5.8% for the month.

2026 spending on AI data centres ($ billion)

Both Amazon and Alphabet say 2027 will be even bigger.

J.P. Morgan's estimate of global AI spending in 2026 alone. It sees $5.5 trillion by 2030.

AI is just hype. Nobody is actually earning from it.

Microsoft's cloud business, Azure, grew 48% in the July–September quarter, and it expects 49–50% next quarter. Azure crossed $100 billion a year for the first time. Its Copilot AI assistant has 30 million+ paid seats.

Some companies are clearly earning. The worry is that revenue is growing slower than the spending bill.

This is exactly the dot-com bubble. Everything will crash 80%.

In 2000–02 the Nasdaq fell about 78%, because many dot-com companies had no profits at all. Today's big spenders are some of the most profitable companies on earth. But they're now borrowing to spend: Meta plans to raise $97.5 billion.

The selloff means companies are giving up on AI.

The opposite. Amazon's CEO told shareholders to expect it to "spend meaningfully more in 2027". Alphabet said the same. Morgan Stanley expects AI infrastructure spending to pass $1 trillion next year.

Investors aren't selling because AI is dying. They're selling because it's getting more expensive.

A US tech crash doesn't touch Indian investors.

Indians hold US tech through mutual funds. The Mirae Asset NYSE FANG+ fund was up about 21% in 3 months, India's top performer. Friday's fall will show up in its NAV this week.

If one of your SIPs is a US tech fund, last quarter's star may be this quarter's biggest drop.

An AI selloff is always bad news for TCS, Infosys and Wipro.

It's more twisted. Indian IT is hurt by AI succeeding, since it shrinks their outsourcing work. In late July, a global AI selloff actually sent Nifty IT up nearly 3% in a day, because Indian IT has little AI hardware exposure.

Weight of the five big IT firms in the Nifty 50: the lowest since at least 2002.

The US Fed raised rates in September for the first time since 2023, then held at 3.75–4.00% on 28 October.

Watch rates as closely as earnings. They decide how much the market pays for the future.

The basic maths A data centre is like a new hospital. You borrow to build it, the machines wear out, and you need enough patients every year to pay for it all, plus a profit.

Rule of thumb from Sequoia's David Cahn, who has tracked this gap since 2023.

The gap keeps growing

Cahn's estimate: the hole grew about 15 times in three years.

One real example

Microsoft's AI revenue (about $37 bn) against its 2026 spending plan (about $190 bn), as of April 2026. The best-placed company is earning back roughly a fifth so far.

If you own a US index fund or a "global" fund, about a third of your money may sit in just seven companies. Check the fund factsheet.

Spending is now faster than cash coming in, so Big Tech is borrowing like never before.

• Oct 2025: Meta sells $30 bn of bonds, the biggest such sale ever.
• Jun 2026: Nvidia borrows $25 bn, its first bond in five years.
• Jul 2026: Amazon's $25 bn sale has to offer extra interest to attract buyers.
• Alphabet even sold a 100-year bond.

J.P. Morgan's forecast for tech-sector bond sales in 2026, with about $317 bn from the cloud giants alone.

The circle problem

Ratings agency Moody's has warned that a handful of AI labs make up a large share of the cloud giants' order books, and that these "circular" deals add concentration risk.

Railway mania, Britain, 1846

Parliament approved 263 new railway companies in one year, for about 9,500 miles of track. Many investors were ruined. Later, about a fifth of US railway mileage went into receivership. The railways still built the modern world.

The fibre bubble, 1996–2001

US telecom firms spent over $500 bn, mostly on bandwidth. In 2001, only about 2.6% of the fibre laid was actually in use. By 2002, 23 telecom companies had gone bankrupt.

The twist

The "dark fibre" was later bought cheap by companies like Google, and became the backbone of search, streaming and the cloud.

The technology usually wins. The people who paid to build it often don't.

World data-centre power use (IEA)

By 2030, more than all of Japan uses today. AI's own share is expected to triple.

India's gap

India creates about a fifth of the world's data but has only about 3% of its data-centre capacity.

India's build-out (Wood Mackenzie)

AI-dedicated capacity: 275 MW to about 6,500 MW. The government says AI data centres could add 26.3 GW to power demand by FY32.

The good ending. AI revenue catches up with spending over 2–3 years, like Microsoft's cloud is showing. This dip becomes a buying chance people later wish they'd taken.

The likely ending. Real technology, overheated prices. Stocks swing hard for a year or two as each earnings season tests the spending, then the winners separate from the losers.

The bad ending. Borrowed money keeps funding data centres that don't pay off, rates stay high, and a 2000-style 40–60% fall hits the most expensive names.

With money

• Check how much of your money is in US tech or AI-themed funds. Keep it a slice, not the meal.
• Don't stop SIPs because of one scary week.
• Don't pile into "AI funds" or NFOs because of the hype either.

With forwards

• "It's crashing 80%!" and "It's going to the moon!" are both forwards, not facts.
• Ask: is the company earning, or only spending?

With your career

Whatever happens to AI stocks, AI tools are already in every office. Learning to use them well is the safest bet on the board.