Is the AI Bubble Bursting? Big Tech's $725B Reckoning

Is the AI bubble bursting in 2026? The honest answer is that nobody knows yet, but the next two weeks will tell us more than the past year of speculation. Big Tech is on track to spend roughly $725 billion on AI infrastructure this year, the Magnificent 7 shed about $2.3 trillion in market value in a single month, and Q2 earnings from Alphabet, Microsoft, Meta, Apple, and Amazon are expected between July 22 and 30. This is analysis, not investment advice — but here is the numbers-first case for why this earnings season is the first real test of the biggest capital bet in corporate history.

The reckoning: why the next two weeks matter

The AI bubble debate has been mostly words for a year. It is about to meet hard numbers. Q2 2026 earnings are expected from Alphabet around July 22, Microsoft and Meta around July 29, and Apple and Amazon around July 30, per earnings trackers — dates can still shift. Nvidia reports later, in late August, on its offset fiscal calendar. These calls matter because they are the first clear chance to check whether AI revenue is catching up to AI spending or falling further behind.

The number that started the panic: $2.3T erased in June

The immediate trigger was a brutal June. The Magnificent 7 — Apple, Microsoft, Alphabet, Amazon, Meta, Nvidia, and Tesla — lost roughly $2.3 trillion in combined market value in June 2026, the group's largest single-month loss on record, according to CNBC and 24/7 Wall St. Roughly speaking, Microsoft fell about 19% (its worst month in years), Nvidia about 11%, Amazon about 9%, Meta about 6%, Apple about 5.5%, and Alphabet about 5%, per an aggregation from 24/7 Wall St citing CNBC data. Treat those single-month percentages as approximate.

How much Big Tech is really spending: the $725B question

Here is the number underneath the anxiety. The four hyperscalers — Amazon, Microsoft, Alphabet, and Meta — are projected to spend roughly $725 billion on capital expenditures in 2026, up about 77% from around $410 billion in 2025, according to Statista and industry trackers. The exact per-company split varies by source, so treat these as rough estimates only:

• Amazon: ~$200 billion
• Microsoft: ~$190 billion (roughly a 61% jump over 2025, by one estimate)
• Alphabet: ~$185 billion
• Meta: ~$125 billion (some sources put its range at $115–145 billion)

The headline — about $725 billion, up roughly 77% — is well-sourced. The individual company figures are estimates that move around, so nobody should treat them as exact.

Where the money goes: chips, data centers, power

Capex on this scale is not abstract. It buys three things: AI chips (mostly Nvidia's), the data centers to house them, and the electricity to run them. That is why a handful of hyperscalers reportedly account for around half of Nvidia's data-center revenue — a concentration bears flag as a risk, and a figure that comes from bearish analysis rather than company disclosure. Capex has also climbed to a projected ~92% of operating cash flow for major hyperscalers by 2026, by one estimate, meaning most of the cash these businesses generate is going straight back into the buildout.

The bear case: the ROI gap, depreciation math, and Michael Burry

The bearish argument has three legs.

First, the ROI gap. Bears note that a large share of the population does not yet pay for frontier AI. One figure cited in coverage holds that less than 2% of people pay for frontier AI models — a single-source claim worth treating skeptically, but it captures the worry that monetization lags spend.

Second, depreciation. Michael Burry, the investor made famous by "The Big Short," has publicly warned of an AI bubble in 2026. He argues hyperscalers extended the assumed useful life of AI hardware, which he estimates could understate depreciation by as much as ~$176 billion across 2026–2028 and flatter reported profits. This is Burry's own contested analysis, attributed to him and not established fact — other analysts dispute both the method and the number.

Third, concentration. The same Burry-linked analysis points to hyperscalers making up roughly half of Nvidia's revenue, arguing the entire trade leans on a small number of buyers.

The bull case: cloud growth and the cost of underspending

The other side is not just hope. Alphabet's cloud unit reported roughly 63% revenue growth by one account, evidence that AI is already monetizing for some players. And the bull case has a counterintuitive kicker: for these companies, underspending may be the bigger risk. If AI turns out to be as important as the CEOs believe, the company that skimped on data centers loses the platform war of the decade. That framing — better to overbuild than to miss it — is why capex keeps rising even as the stocks wobble.

There is also a real distinction between price and product. Whatever the market does, the models keep getting cheaper and more capable to use. You can see the spread across vendors on our LLM pricing page: the tools that run on all this infrastructure are getting cheaper for users even as the infrastructure bill climbs.

Is this a "bubble"? What the word actually means

"Bubble" gets thrown around loosely. Strictly, it means asset prices detached from fundamentals — paying dot-com prices for a company with no path to profit. That is different from a heavy, risky capital cycle where real revenue exists but may not justify the spend. The defensible read in 2026 is that the AI trade sits somewhere on that spectrum, not obviously at either end. Calling the top with confidence is a guess; so is declaring the spending clearly justified. The word "bubble" is doing a lot of work in headlines that the data does not yet settle.

The tell to watch on the July earnings calls

If you want to judge for yourself, watch one thing on each call: capex guidance versus AI revenue disclosure. If a company raises capex again but stays vague about AI revenue, bears get ammunition. If it raises capex and shows concrete AI-driven revenue — cloud growth, ad gains, paid AI seats — the bull case strengthens. The gap between what these companies spend and what AI actually earns is the whole debate, and these calls are where it finally gets measured. For a sense of what the models cost to actually run, our LLM cost calculator puts real per-token numbers next to the billions in the headlines.

2027 and the road to $1 trillion

The spending is not projected to stop. Analysts at Evercore and Bank of America forecast hyperscaler capex topping $1 trillion in 2027, with Goldman's base case around $1.1 trillion and a bull case near $1.4 trillion, per CNBC and Fortune — the "above $1 trillion" consensus is firmer than any single bank's specific number. Whether that figure is visionary or reckless depends entirely on the revenue that shows up to meet it.

What it means for your portfolio — measured, not advice

Handle this part plainly: this is not financial advice, and nothing here is a recommendation to buy or sell anything. What is fair to say is that if you hold an index fund or a typical 401(k), the Magnificent 7 is likely a meaningful chunk of it, so this debate is not only for day traders. The measured takeaway is awareness, not action — understand that a large slice of the market's value now rests on an AI bet whose payoff is still unproven, and make any decisions with a licensed advisor rather than a blog post. This piece does not predict where any stock goes next.

Dot-com déjà vu? Where the analogy holds and breaks

The comparison everyone reaches for is 2000. It holds in one way: massive infrastructure spend ahead of proven demand, much like the fiber buildout of the late 1990s. It breaks in another: the companies doing the spending today are highly profitable, cash-generating incumbents, not cash-burning startups. The fiber glut eventually got used; the pets.com equity did not. Which precedent fits AI is genuinely unknown, and that uncertainty — not a confident crash call — is the honest position.

So, is the AI bubble bursting in 2026? The defensible answer is that it is neither a scam nor free money. It is the largest capital bet in corporate history, and this earnings season is its first hard test. The roughly $725 billion in spending, the $2.3 trillion June drawdown, and the trillion-dollar 2027 forecasts are real; whether they add up to a bubble depends on revenue that has not been reported yet. Watch the capex-versus-revenue gap on the July calls, keep both the hype and the doom at arm's length, and remember that the loudest voices on either side are, for now, still guessing.