3BQRIIINO. INDEPENDENT TECH JOURNAL
← All stories

AI Bubble / Nvidia / OpenAI / AI Industry / Markets · November 10, 2026

Is AI a Bubble? The Crash, the Money Loop, and the Revenue Gap, Explained

After the worst stock-market day since April 2025, 'AI is a bubble' is everywhere again. The claim bundles three different questions — and the evidence gives a different answer to each. Here is the money loop, Goldman's $230 billion gap, and the demand numbers on the other side.

Is AI a Bubble? The Crash, the Money Loop, and the Revenue Gap, Explained

Is AI a Bubble? The Crash, the Money Loop, and the Revenue Gap, Explained

"AI turned out to be a bubble" is suddenly everywhere — in videos, in group chats, in market commentary. The trigger was real: on Thursday 5 November 2026, US stocks had their worst day since April 2025. The S&P 500 fell about 4%, the Nasdaq about 5%, and chip stocks dropped roughly 10% intraday; estimates of the value erased that day run from $3.2 trillion (FactSet figures reported by the Financial Times) to $3.7 trillion (Bloomberg). Nvidia alone has lost about $1.4 trillion in market value since its October record, according to market coverage of the sell-off.

A one-day crash, however, is not an analysis. "AI is a bubble" bundles together three separate claims, and the evidence points a different way on each one. This article separates them — because for anyone who uses AI tools to make a living, the difference matters.

Classification: the market events and company figures below are Confirmed News from the cited reporting, checked 10 November 2026. Whether they add up to "a bubble" is a contested characterisation, and it is treated as one throughout.

Claim one: "The technology itself is a fad"

This is the weakest version of the argument, and the demand data contradicts it.

None of this proves the valuations are right. It does disprove the claim that nobody is paying for the product.

Claim two: "The financing has bubble mechanics"

This version is much harder to dismiss — and it is where the serious critics, including Michael Burry and analysts at Goldman Sachs itself, are actually aiming.

The spending is enormous and still accelerating. Goldman estimates the big five's AI infrastructure spending at roughly $800 billion in 2026, rising to $1.2 trillion in 2027 — above the Wall Street consensus. Alphabet alone has signalled more than $300 billion for 2027.

The revenue does not yet cover it. Goldman's September note put the break-even point for the build-out at about $300 billion of annual AI revenue (and roughly $417 billion for a 10% return on invested capital). Against that, the cleanest measure of genuinely new AI demand — cloud revenue above its pre-AI trend — was about $70 billion annualised. That is a gap of roughly $230 billion a year that has to close. Bain's September 2026 Global Technology Report reached a similar shape of conclusion, estimating an $800 billion annual revenue shortfall by 2030 on current trajectories.

Some of the money moves in a circle. The most-cited example: Nvidia's announced investment of up to $100 billion in OpenAI, alongside OpenAI's purchases of Nvidia systems; OpenAI's roughly $300 billion cloud deal with Oracle, which in turn spends billions on Nvidia chips; Nvidia owning about 7% of CoreWeave and committing to buy back $6.3 billion of its cloud capacity. The London credit fund Sona Asset Management mapped $3.6 trillion of AI financing across 176 deals and flagged about 120 as "highly circular" — the same companies appearing as investor, supplier and customer at once. Both the IMF and the Bank for International Settlements have named circular AI financing as a risk to the wider financial system. The deals are disclosed and are not evidence of wrongdoing — but circular demand is harder to trust than independent demand, which is precisely the dot-com era's vendor-financing lesson (the comparison hedge-fund veteran George Noble draws explicitly with Cisco, whose peak index weight Nvidia's roughly 8% of the S&P 500 now doubles).

The guarantees are piling up behind the scenes. Reporting based on Nvidia's own filings describes equity investments of about $99 billion plus $25 billion in commitments, and — most strikingly — credit support of up to $105 billion for the SB Energy data-centre campus in Ohio being built to serve OpenAI. When a supplier guarantees its customer's landlord, the supplier is exposed if the customer stumbles.

The accounting is disputed. Burry argues hyperscalers depreciate AI chips over five to six years when their realistic useful life is nearer two and a half — which, if he is right, understates depreciation by roughly $176 billion across the industry and flatters current profits. Nvidia disputed his maths in a memo to analysts. The dispute itself is the point: reported AI profits depend on an assumption nobody can yet verify.

And the biggest customer in the loop is not profitable. OpenAI is reportedly on track to lose around $14 billion this year. Meanwhile borrowing for the build-out has risen sharply — Evercore noted hyperscalers and AI companies seeking hundreds of billions in financing within a two-week span, and Bank of America economists have linked part of the rise in interest rates to build-out borrowing. AI investment now accounts for roughly a quarter to a third of US economic growth this year, by estimates reported in the Wall Street Journal — which is why this is no longer just a tech-sector question.

Claim three: "It has already burst"

Unproven. A bubble bursting is a process measured in months — a collapse in financing, defaults among the circular deals, capex guidance being slashed — not a single brutal session, however historic. Two facts argue against calling 5 November the end:

What would actually settle the argument is knowable, and some of it is dated: Nvidia reports earnings on 17 November, the single most-watched checkpoint. Beyond that: whether AI revenue keeps compounding toward Goldman's $300 billion break-even line; whether any circular deal defaults and pulls a guarantor with it; and whether 2027 capex guidance (already pencilled at $1.2 trillion) gets confirmed or cut.

What this means if you create with AI tools

You do not need a position on Nvidia's stock to be affected by this. Three practical consequences:

  1. Today's AI prices are partly subsidised. Venture money, circular deals and a capex arms race are why frontier models are sold below their true cost. If the financing tightens, API and subscription prices rise or free tiers shrink — the bubble's bursting would reach your tools bill, not just portfolios.
  2. Keep your workflow portable. If your process depends on one vendor's below-cost pricing or one model's quirks, a repricing becomes your emergency. Export your assets, know your second-choice tool, and avoid annual lock-ins you cannot justify at unsubsidised prices.
  3. A correction is not a shutdown. Even in the harshest scenario above, the data centres exist, the backlogs are contracted, and the revenue is real. The dot-com crash bankrupted the financing layer while the fibre it overbuilt carried the next two decades of the internet. AI's version would rhyme: the tools keep working and keep getting cheaper to run; it is the money around them that gets repriced.

The verdict

Three claims, three answers. The technology is not a fad — the revenue and backlog data kill that version. The financing layer genuinely shows bubble mechanics — circular deals, supplier guarantees, a $230 billion annual revenue gap against break-even, and accounting assumptions doing heavy lifting — and the critics documenting that are citing the companies' own numbers, not vibes. And "it has burst" is, so far, a headline in search of evidence: 5 November was a violent correction after an enormous run, with the first real test — Nvidia's 17 November earnings — still ahead.

If the bubble story has a centre of gravity, it is this: the risk was never that AI does not work. It is that it might not earn back $800 billion a year fast enough for the people who financed it. Those are very different problems — and only one of them should worry the people who merely use the tools.

Sources (checked 10 November 2026)

Market figures are as reported by the cited sources on their publication dates; several (OpenAI's losses, Anthropic's annualised revenue, the day-loss totals) are estimates or reported figures rather than audited disclosures, and are flagged as such in the text.

Prefer reading in Arabic? You can translate this article into Arabic with Google Translate.