Seven hundred billion dollars is a number that stops meaning anything. So here is a way to feel it. Amazon, Microsoft, Alphabet, and Meta together are expected to spend between $700 billion and $725 billion on capital projects in 2026, mostly data centers, chips, and the power to run them. According to a Forbes analysis published October 7, 2026, that is up about 75% from a record $410 billion last year.
First, What Does “Bubble” Actually Mean?
A bubble is not just “a lot of spending.” It is spending, and the prices attached to it, running well ahead of the future earnings that could justify them. A big build-out can be a bubble, or it can be a good investment that arrives early. The difference only shows up later, which is why the argument never ends.
So the useful question is not “is AI real?” It clearly is. The useful question is whether the money going in can plausibly earn its way back, and how fast.
Exhibit A: The Spending
The Forbes analysis puts the 2026 guidance like this:
- Amazon: about $200 billion
- Microsoft: around $190 billion
- Alphabet: $180 billion to $190 billion
- Meta: $125 billion to $145 billion
Several Wall Street firms project that combined spending will top $1 trillion in 2027. CreditSights estimates that roughly three-quarters of it funds AI infrastructure. Zoom out further and a Brookings estimate, presented in late September, puts infrastructure investment through 2032 at about $10.3 trillion, or roughly 3.6% of US GDP each year, as reported by Axios.
On its own, this exhibit proves nothing. Large spending is exactly what you would expect if demand were enormous. It only becomes worrying next to the revenue.
Exhibit B: The Revenue
Here the picture is mixed, and that is the honest summary.
On the encouraging side, Microsoft’s AI business passed a $37 billion annual revenue run rate earlier in 2026, more than doubling from a year earlier. Google Cloud’s backlog, meaning signed commitments not yet delivered, swelled past $460 billion. Those are real signals that customers are buying.
On the worrying side, a Stanford economists’ analysis cited by Axios on September 30 finds a shortfall of nearly $1 trillion between what the largest spenders have put into AI and the AI revenue they have earned since 2024. To make the spending pay off, those companies would need to triple or quadruple AI revenue next year and keep growing at that pace for about a decade. Goldman Sachs, according to the same report, says hyperscalers’ AI revenue is still below what they need to break even on their capital spending, though it is more optimistic that returns will come.
One more wrinkle: Meta does not sell cloud capacity, so its AI payoff has to arrive indirectly through advertising and engagement. That makes its returns harder to measure from outside.
Exhibit C: The Financing
Who pays for all this matters as much as how much. Forbes notes that capital spending is on track to consume nearly all of these companies’ operating cash flow this year, and some of it is financed with debt.
Other reporting from early October points the same way. Morgan Stanley has estimated that AI infrastructure could need about $1.5 trillion in external financing by 2028. SpaceX was reported to be pursuing around $40 billion to buy Nvidia chips. And Firmus Technologies, an Nvidia- and Blackstone-backed data center company, withdrew a planned IPO of about $5 billion after Reuters reported that only 42 megawatts of its planned 1 gigawatt of capacity was operational.
Debt is not a warning sign by itself. It becomes one when revenue growth slows and interest costs rise. The Axios report notes that the Stanford analysis assumes the cost of capital will not rise meaningfully, while interest rates are now climbing.
Exhibit D: The Chips Wear Out
This is the least discussed piece of evidence. A railroad lasts decades. A warehouse of AI chips does not. Axios reports that much of the spending goes to chips that lose value after about five years, which shortens the window in which the investment has to pay for itself. Microsoft also attributes about $25 billion of its 2026 budget to higher component pricing, so some of the spending is simply things costing more.
The Defense
A fair trial hears the other side.
The most common defense is historical. Electrification, the Stanford economists note, took decades to benefit factories, so AI payoffs may simply take longer than investors want. Forbes makes a similar point, comparing AI to earlier technology build-outs that produced long-term value alongside “painful periods of overbuilding,” and saying AI will likely be no different.
Both can be true at once: the technology can change the economy and some of the money spent on it can still be lost. Stanford economist Ryan Cummings, quoted by Axios, says he thinks these companies will eventually turn the spending into profit, though he is unsure which ones will, and he expects the big profits later than investors need them.
The Verdict: Not Provable Yet, But Not Ignorable
The evidence supports a more careful statement than either camp offers: AI looks like a real technology boom with credible overbuilding risk. Whether it is a bubble depends on whether revenue grows fast enough, for long enough, before financing costs and chip replacement cycles catch up. Nobody knows that yet, and anyone who claims certainty is selling something.
What you can do is watch for the signals that would settle it.
Five Tripwires to Watch
- AI revenue growth versus capital spending. If revenue keeps compounding, the gap closes. If it flattens while spending climbs, worry grows.
- The cost of borrowing. Rising interest rates hurt debt-funded projects most.
- Funding stress. Withdrawn IPOs, delayed projects, and more debt-financed chip purchases are early warnings.
- Chip economics. Faster obsolescence or pricing swings change the math on payback.
- Backlog turning into revenue. Signed commitments only matter when customers actually pay.
What This Means for You
If you use AI tools every day. The tools will likely stay, but pricing and free tiers can change. Treat today’s generous terms as not guaranteed.
If you run a small business. Avoid building a process that depends on one vendor. Picture an online seller in Ohio who runs all customer replies through one paid tool. If the price doubled next year, she should be able to switch within a week, which means keeping her templates and policies in her own documents.
If you are job hunting or building skills. Skills hold value whatever markets do. Knowing how to use AI well is useful in a boom or a correction, as the guide to AI skills employers are actually hiring for explains.
If you invest. Forbes flags concentration risk, since many index and tech funds are heavily weighted toward the same few companies. This post is not advice. If you are unsure how your holdings are exposed, a licensed financial professional can help.
Common Problems and Practical Solutions
Problem: I cannot tell whether the headlines are hype or warning. Solution: Ignore the adjectives and track the numbers: revenue, spending, borrowing costs. Check them quarterly rather than daily.
Problem: I am afraid my AI tools will disappear or get expensive. Solution: Keep your prompts, templates, and data in files you own, and test a second tool now so switching is easy later.
Problem: I want to invest in AI but worry about timing. Solution: Understand your total exposure first, including funds you already hold. A professional can help you decide, and this is a decision for you, not a headline.
Problem: I worry a crash would cost jobs. Solution: Build skills that transfer between employers and industries. This explainer on AI versus human jobs is a useful starting point.
Problem: The topic feels too big to follow. Solution: Pick one source type, such as company earnings reports, and read just the AI-related lines each quarter.
Where to Read Next
If the hardware side of this story interests you, see why AI chip demand is booming and what comes beyond GPUs. For the investor angle, our guides to AI for stock market analysis and AI in trading show what these tools can and cannot do. And for where the technology is headed next, read what’s next in AI: 7 trends to watch in 2027. To keep following practical AI guides, you can learn more about us on the AI Overview Search about page.
Frequently Asked Questions
Is there an AI bubble right now? The evidence is mixed. Spending is growing far faster than revenue, but revenue and customer commitments are also growing quickly. Analysts disagree, and the answer will only be clear in hindsight.
How much are tech companies spending on AI? Forbes reported on October 7, 2026 that Amazon, Microsoft, Alphabet, and Meta together are guiding to roughly $700 billion to $725 billion in 2026 capital spending, with several Wall Street firms projecting more than $1 trillion in 2027.
How is this different from the dot-com bubble? Reporting highlights that AI spending is led by very large, profitable companies, but also that it is increasingly debt-financed. Both earlier build-outs and AI involve long-term value alongside the risk of overbuilding. The comparison is useful but not exact.
Will the AI bubble pop? Nobody can say. Watch for slowing revenue growth, rising borrowing costs, funding stress, and faster chip obsolescence as early warnings.
Should I stop using AI tools if there is a bubble? No. A market correction would not remove useful technology. But avoid depending on a single vendor, and keep your own copies of your work.
Should I sell my AI stocks? This post cannot tell you that. It is general information, not financial advice. Consider your overall exposure and consult a licensed professional.
