Sedenir.dev | Has the AI bubble already burst? The viral video summarizes four sources

A video with a blunt title — "A bolha da IA já estourou | E os CEOs só não te contaram ainda" ("The AI bubble already burst | And the CEOs just haven't told you yet") — has been circulating widely over the past few days. Its central argument: big tech companies are already quietly pulling back from billion-dollar AI infrastructure bets, even as the public narrative keeps pointing to unrestrained expansion. Rather than just commenting on the video's opinion, we went straight to the four sources it cites as its basis — and they tell a consistent story, even coming from very different places: a Fortune report, a Forbes article, the Bank for International Settlements' (BIS) annual report, and an independent quantitative analysis published on Substack.

The lead that started it all: Microsoft canceling data centers

The oldest reference in the group is a Fortune report from February 2025: according to an analysis by brokerage TD Cowen, Microsoft had begun canceling leases for a substantial amount of US data center capacity, which analysts read as a possible sign the company was building more AI computing capacity than it would actually need in the long run.

OpenAI, Microsoft's biggest partner, had leases totaling hundreds of megawatts of capacity voided, and Microsoft stopped converting so-called "statements of qualification" — agreements that typically evolve into formal leases — a tactic rivals like Meta had already used when cutting back on spending.

Note

Microsoft, at the time, publicly reiterated its target of spending more than $80 billion on AI data centers for the fiscal year, saying it would "strategically" pace some areas without slowing growth elsewhere. The episode is more symptom than cause: it shows cracks in expansion plans were already appearing even while public messaging stayed optimistic.

A more recent lead: Starbucks retiring its inventory AI

The video's title also mentions Starbucks, and the case illustrates a different kind of problem: real-world AI adoption outside the infrastructure world. In May 2026, the chain quietly retired an automated inventory counting tool built by NomadGo, which used LiDAR sensors and cameras to count items like milk and syrups on shelves — just nine months after it had rolled out across more than 11,000 North American stores as part of CEO Brian Niccol's turnaround plan.

According to employees who spoke to Reuters, the tool frequently miscounted and mislabeled items, confusing milk types often enough that staff ended up manually re-verifying what the system was supposed to have already counted — erasing the efficiency gain the tool existed to deliver.

Not a wholesale rejection of AI

Starbucks hasn't abandoned AI as a strategy: it still runs Green Dot Assist, an assistant built on Azure OpenAI that helps baristas with recipes and troubleshooting, and it's building internal software using AI coding tools to replace Microsoft and IBM systems. What failed was specifically computer vision applied to a chaotic physical environment — reorganized shelves, changing lighting, misplaced products — where error rates above a certain threshold destroy users' trust in the system.

The gap between spending and revenue

The centerpiece of the financial argument comes from a Forbes article published in June 2026. The five largest hyperscalers — Amazon, Microsoft, Alphabet, Meta, and Oracle — are on track to spend between $700 billion and $900 billion in capital expenditures in 2026, a 36% increase over 2025, according to CreditSights estimates. Amazon alone guided for $200 billion, more than double its prior-year outlay.

The problem, per the article, isn't a lack of revenue — AWS is running at roughly $150 billion annualized, growing 28% year-over-year, and Microsoft's AI business crossed $37 billion in annualized revenue, growing 123%. The problem is relative speed: investment is growing faster than revenue, which keeps pushing the payback period on that capital further into the future.

The divergence between AI capex growth and revenue growth is already running at around 46%, according to Allianz Research — exceeding the 32% divergence observed during the telecom sector's excess in 2001, a period that preceded a brutal, multi-year correction in tech markets. As a result, companies are increasingly turning to debt: the group raised $108 billion in new debt in 2025, and projections from banks like Morgan Stanley and JPMorgan suggest the sector may need to issue $1.5 trillion in debt over the coming years just to finance construction already underway.

The tremor markets already felt

Meta shares fell 9.25% in a single session after the company raised its capex guidance, described by Mark Zuckerberg as investment in "personal superintelligence for billions of people." For the article's author, that reaction may have been the first real tremor of a broader repricing across the sector.

The quantitative case: why the math doesn't add up

The densest numbers-driven analysis comes from Anomaly Investment Partners, published in June 2026, updating an earlier report from December 2025. The argument starts from a simple question: how much new revenue would the AI industry need to generate to justify the capital already invested?

According to J.P. Morgan, cited in the piece, the industry would need close to $650 billion in new annual revenue just to hit a modest 10% return on the infrastructure being built. Bain's projection, also cited, is even more demanding: $2 trillion in annual AI revenue by 2030. Against that requirement, revenue currently attributable to AI — even under generous assumptions crediting all incremental cloud growth to AI — sits between $50 billion and $150 billion a year. The gap ranges from 4x to 13x, depending on which end of the range is used.

The piece also details three structural fragilities that, according to the author, rarely get covered together:

Three blind spots the Anomaly Investments thesis highlights

Overly optimistic depreciation. Hyperscalers assume a five-to-six-year useful life for AI servers, while Nvidia's release cycle renders prior GPU generations obsolete for primary workloads within two to three years — inflating reported earnings.
Circular financing. Chipmakers invest in the very AI companies that buy their products, which in turn are funded by cloud providers that require the money to flow back as infrastructure spending commitments. A meaningful share of "new revenue" is capital recycling within the same small circle of companies.
Off-balance-sheet lease commitments. Moody's identified hundreds of billions of dollars in data center contracts already signed but not yet commenced, which don't show up in the capex figures investors typically scrutinize.

The article acknowledges the counterargument: part of the growth is genuine, and Anthropic is cited as the most financially solid of the three AI IPOs expected in the second half of 2026, with expanding gross margins and a projected positive cash flow as early as 2028. Even so, the author's conclusion is that the pace of capital deployment is running structurally ahead of what any reasonable revenue scenario can sustain — the same pattern, in his view, that characterized the 19th-century railway boom, the 1920s electrification wave, and the fiber optic buildout of the late 1990s.

The warning from the central bank of central banks

The fourth reference is the most institutional of the four: the Bank for International Settlements, the body that coordinates central banks around the world. In its 2026 Annual Economic Report, published in June, the institution compared the current AI investment boom to historical episodes — 19th-century canal mania, Britain's railway bubble, 1920s electrification, and the dot-com bubble — describing a common thread across all of them: genuine technological breakthroughs that attracted more capital than commercial returns could ultimately justify.

The BIS estimates the five largest hyperscalers will spend more than $1 trillion on AI capex between 2025 and the end of 2026, increasingly financed by debt and by circular financing structures — where chipmakers and cloud providers recycle revenue between themselves via equity stakes. The report warns that a significant repricing of AI-related stocks could produce sharper wealth effects and a steeper consumption pullback than in past cycles, with the potential to spread internationally given the weight of the US market.

The BIS doesn't call it a bubble — but doesn't rule it out either

The institution avoids categorically using the word "bubble," preferring the term "robustness" to describe what it recommends policymakers build. Even so, the report is direct in stating that disappointing AI returns could trigger a sudden pullback in financing, turning the capex boom into a prolonged period of investment retreat.

What to make of this

All four sources agree on one central point: there is a real and growing gap between the capital being invested in AI infrastructure and the revenue that infrastructure currently generates. Where they diverge is on the urgency and inevitability of the outcome.

Fortune documents a specific symptom (lease cancellations) that could reflect either genuine caution or strategic reallocation — the report itself notes analysts saw it as "net neutral" for third-party demand.
Forbes describes a market dynamic (the capex-revenue gap) without delivering a verdict on "bubble," treating it as something to monitor.
Anomaly Investments builds the most explicit, quantified case that this is a bubble, while acknowledging the technology's long-term potential is real.
The BIS sits in the middle: it declines to use the word "bubble" outright, but describes exactly the mechanisms — debt, circular financing, risk concentration — that have historically preceded corrections of this kind.

Where this leaves people working in tech

Regardless of which side of the debate one lands on, the IPOs of SpaceX, OpenAI, and Anthropic expected in the second half of 2026 will function as the first real public test of these assumptions — exposing, for the first time under open-market quarterly scrutiny, cost and financing structures that have so far stayed shielded behind private capital.

Has the AI bubble already burst? The viral video summarizes four sources — and they don't contradict each other