AI Bubble: What the Spending, Revenue and Debt Data Shows

Last Updated: October 4, 2026

Parts of AI look like a bubble and parts look like earnings. Nvidia reported $96.2 billion in revenue for the quarter ended July 2026, while the biggest cloud companies spend hundreds of billions and borrow more. The Bank of England flags financing, more than demand, as the weak link, as of October 2026.

Alphabet's Google Cloud grew 82% in the second quarter of 2026 to $24.8 billion. In the same quarter, Alphabet's free cash flow was negative $5.9 billion, according to its Q2 2026 earnings call. Both numbers are true, and together they explain why the AI bubble debate has no easy answer.

Part of the confusion is that nobody agrees on what the term means. Some people mean stock prices, some mean spending, and some mean the debt behind it.

This guide tests the AI bubble claim one layer at a time, using company filings, central bank reports, and government data. It separates what looks like real demand from what looks like risk.

💡 Not sure which AI tool is actually right for your business?

Get the free guide, Which AI Tool Should You Actually Use? — a straightforward breakdown of the leading AI tools to help you pick the right one for your needs.

Subscribe to AI Business Weekly for the guide, plus daily coverage of AI trends, acquisitions, and product launches.

Get the Free Guide →

The AI bubble is the idea that prices, spending, and borrowing around artificial intelligence have run ahead of the profits the technology can deliver. A bubble forms when expected future earnings are priced in before the cash arrives. Whether AI qualifies depends on which part of the industry you measure.

Even the people building it use the word. MIT Technology Review reported in December 2025 that Sam Altman said "when bubbles happen, smart people get overexcited about a kernel of truth." OpenAI chairman Bret Taylor put both halves in one sentence: AI will transform the economy, and "we're also in a bubble, and a lot of people will lose a lot of money."

A cleaner way to read the debate is to split it into three questions. Are stock prices ahead of earnings, is spending ahead of revenue, and is the financing fragile? The answers differ, which is why "tech bubble or not" rarely resolves in a headline, whether it is framed as a tech stock bubble or an AI industry bubble.

Partly, and unevenly. Chipmakers and cloud providers report record earnings and backlogs, which looks like real demand, while spending, private valuations, and borrowing are growing faster than revenue in other layers, which looks like bubble risk. A 2026 research preprint reached the same split verdict.

The preprint's authors, Qianan Wang and Zen Chen, concluded that AI is "a real technological revolution with localized bubble dynamics". It found stronger fundamentals in semiconductors and more fragility in applications and private model companies. It is a preprint, so treat it as a framework rather than a verdict.

The closer a company sits to selling chips or cloud capacity, the stronger the reported cash. The closer it sits to the model lab or the lender, the more the story depends on revenue that has not arrived yet. Nvidia's figures come from its fiscal Q2 2027 release, and its Nvidia statistics page tracks the longer run.

Alphabet, Meta, and Amazon alone are guiding to roughly $545 billion to $570 billion in 2026 capital spending. Alphabet raised its range to $195 billion to $205 billion in July, Meta guides $130 billion to $145 billion, and CNBC reported Amazon raised its 2026 figure to $220 billion. Microsoft reports on a fiscal year ending June.

The table collects the latest quarter from each company's own results, all reported in late July 2026.

The sources are Microsoft's fiscal Q4 results, Amazon's Q2 release, Meta's Q2 release, and CNBC's report on Amazon's capex increase.

One detail in the Microsoft row matters more than the headline: its backlog excluding OpenAI grew 25%, against 84% including it. That gap suggests much of the growth is commitments from a single lab that is itself burning cash, the circular exposure regulators keep flagging.

Nvidia's $96.2 billion quarter is, in part, these same companies' capital spending arriving as revenue. The chip layer's strength and the cloud layer's cash strain are two views of one flow of money. For the longer run of numbers, see our AI spending statistics.

Not yet at the scale the spending implies. Bain & Company estimated that AI compute needs $2 trillion in annual revenue by 2030, and even generous assumptions left an $800 billion yearly shortfall. Anthropic and OpenAI, the two largest labs, report run rates of roughly $65 billion and $40 billion.

Bain's September 2025 technology report reached the $800 billion figure even if US companies moved all on-premise IT budgets to the cloud and reinvested AI savings. Axios reported in August 2026, citing Bloomberg, that Anthropic's run rate passed $65 billion in late July, up from about $9.3 billion at the end of 2025. It noted the two labs may measure run rate differently.

Add the two run rates and you get about $105 billion, roughly a twentieth of Bain's 2030 requirement. That is a floor, not a gap measure, because the labs are only one slice of AI revenue and Bain counts savings as well. It still shows how far the funding model has to travel, and our OpenAI statistics page follows the lab-side numbers.

On the customer side, the Census Bureau's Business Trends and Outlook Survey found 17% to 20% of US businesses using AI between December 2025 and May 2026, rising to 37% among firms with 250 or more employees. A Federal Reserve note from July 2026 found that labor-market effects remain concentrated and have not broadened across the economy.

MIT Technology Review also cited a 2025 MIT study finding that 95% of organizations investing in generative AI were getting "zero return". That study predates the 2026 revenue figures above, so read it as a snapshot of early deployments rather than the current picture. Our AI adoption statistics and AI ROI statistics pages cover the return question in more depth.

In some ways it is more extreme and in others less. Valuation and concentration measures rival or exceed March 2000, but today's leaders earn real profits, while many dot-com leaders did not. The Bank of England said in July 2026 that a key valuation gauge for the S&P 500 is approaching dot-com levels.

The table uses figures from a TNW comparison published April 25, 2026. They cover US equities, and the 2026 values were snapshots from that spring.

Definitions matter here. The Bank of England's July 2026 report said AI companies now account for around half of the S&P 500, up from a quarter in 2022, which uses a broader definition than the top-10 share above. The IMF's April 2026 stability report added that the S&P 500 would need earnings to grow at close to 30% a year to justify prices, well above analyst expectations.

Our reading: the AI boom is cheaper than 2000 on forward earnings but more concentrated, so fewer names carry more of the index. Our guide to the biggest AI companies shows how few names carry the weight.

Debt is where regulators see the most new risk. The Bank of England reported that AI hyperscalers went from 3% of US investment-grade debt at the end of 2025 to 15% of issuance by early May 2026. AI issuers also made up 41% of non-refinance US high-yield issuance so far this year.

The same Bank of England report said private credit's share of AI financing jumped from 9% in 2024 to 34% in 2025. It also warned of "self-reinforcing capital loops," where technology companies invest in AI firms that then buy those companies' products. A separate concern is timing: data centers financed with long-dated debt can lose value if newer chips make them obsolete.

A September 29 Forbes column by David Trainer, an opinion piece with a bearish thesis, collected specific lender data points: a CoreWeave term loan costing about 10.4%, an $11.1 billion SoftBank junk-bond offering at 8.625% and 9.75% that the column called the largest on record, and 5% withdrawal caps at Apollo, Blackstone, and Blue Owl funds. Axios reported on September 25 that Oracle's 2055 bonds traded at 77 cents on the dollar after it moved to delay lease payments on a New Mexico data center.

The counterweight is the IMF, which said the financial stability impact appears modest currently. Our reading is that debt changes the failure mode: a bubble funded with cash tends to fade slowly, while one funded with borrowing can force selling.

The IPO window is one place to watch, as covered in our OpenAI IPO and Anthropic IPO guides.

No one can say, and anyone who gives a date is guessing. What can be tracked are the signals that would come before a burst: spending guidance, credit costs, IPO access, and lender behavior. As of October 2026, spending and chip demand were still rising while funding signals had weakened, per the sources below.

The IPO rows come from the Forbes column above and a Financial Times-based report on OpenAI's projections, which noted Altman indicated a likely 2027 listing and the $278 billion cash-burn forecast. Those projections are internal and unaudited.

Demand signals are healthy and funding signals are strained. Our reading is that if the AI bubble pops, it is more likely to start in credit than in chip orders. Whether it pops or deflates slowly is the open question.

A sharp equity correction would come first, amplified by concentration. The Bank of England modeled a hypothetical 45% US equity fall over six quarters that would cut UK GDP by 2.2 percentage points. Google CEO Sundar Pichai told the BBC no company in the sector would be immune.

The Pichai comment was reported in November 2025. The Bank of England scenario is a stress test, not a forecast, and it shows how an AI-led drop would travel through credit spreads, which it modeled at 350 basis points wider.

What would survive is the hardware and the demand that already exists. Cloud revenue, chip sales, and lab run rates would not vanish in a correction, though weaker lenders and the most leveraged data center projects would be the first casualties. This is analysis of public data, not investment advice.

Partly. Chip and cloud companies report record earnings and backlogs, while spending, borrowing, and private valuations are growing faster than revenue elsewhere. The Bank of England said in July 2026 that AI valuations are concentrated and that sharp corrections remain a risk, while the IMF called the current stability impact modest.

For the numbers behind that split, see our AI spending statistics.

The AI bubble is the claim that prices, spending, and debt tied to artificial intelligence have outrun the profits the technology can produce. A bubble forms when future earnings are priced in before they arrive. The debate usually covers three things: stock valuations, capital spending, and how the buildout is financed.

Our AI ROI statistics page covers the return side of that question.

No one knows, and any specific date is a guess. Analysts watch spending guidance, credit costs, IPO access, and lender behavior instead. As of October 2026, hyperscalers were still raising capital spending while borrowing costs and private credit limits showed strain, according to Forbes and Axios reporting.

Our guides to the OpenAI IPO and Anthropic IPO track one of those signals.

In part. Spring 2026 data showed the top 10 companies at 36% to 40% of the S&P 500 versus about 27% in 2000, and a Shiller CAPE of 38 to 40 versus 44.19. Today's leaders, however, generate large free cash flow, while many 2000 leaders did not.

See our list of the biggest AI companies for who holds that concentration.

A sharp stock market drop would likely come first. The Bank of England modeled a hypothetical 45% US equity fall that would cut UK GDP by 2.2 percentage points, with credit spreads widening 350 basis points. Google's CEO said no company in the sector would be immune.

The IMF said the financial stability impact appears modest currently.

Nvidia is the clearest case of AI demand showing up as profit. It reported $96.2 billion in revenue and $59.7 billion in GAAP net income for the quarter ended July 26, 2026, and guided to $108 billion next quarter. Its results still depend on customers continuing to raise their own capital spending.

Our Nvidia statistics page has the longer record.

Alphabet, Meta, and Amazon alone are guiding to roughly $545 billion to $570 billion in 2026 capital spending, based on their July 2026 reports. Alphabet's range is $195 billion to $205 billion, Meta's is $130 billion to $145 billion, and Amazon's reported figure is $220 billion.

Microsoft reports capital spending on a fiscal year ending in June.

The AI bubble is better understood as three questions than one. Chips and cloud look like earnings, with Nvidia at $96.2 billion in quarterly revenue and Google Cloud up 82%, while spending, private valuations, and borrowing look riskier as the Bank of England flags debt and circular financing. No one can date a burst, but credit and IPO signals, not chip demand, showed strain first as of October 2026, and this is analysis of public data, not investment advice.

By Sameer Khan

AI-assisted. Researched, reviewed, and edited by Sameer Khan before publication.