Five Tech Giants Will Spend $1 Trillion on AI This Year. The BIS Just Compared It to Canal Mania.

The five companies building most of the world's AI infrastructure are set to spend more than a trillion dollars doing it, in two years. That figure alone would be striking. What makes it newsworthy is who just flagged it as a risk: not a short-seller, not a rival, but the Bank for International Settlements, the institution that exists to watch central banks watch everyone else.

In its Annual Economic Report 2026, released this June, the BIS didn't just note the size of the AI buildout. It lined it up against four historical investment manias, and none of the comparisons were flattering.

The Trillion-Dollar Bet

The report's starting point is simple: AI's productivity case is real but still unproven at scale. Task-level studies point to large efficiency gains, commonly in the 20 to 50% range for time saved on specific tasks like coding or customer support. But aggregate productivity growth estimates across the whole economy are far more modest, under 1% over a long horizon, reflecting how hard it is to adopt a new technology broadly and fold it into existing ways of working.

That gap between task-level promise and economy-wide delivery hasn't slowed the spending. The BIS puts a number on it: the five largest hyperscalers are on pace to spend over a trillion dollars on AI-related capital expenditure across 2025 and 2026 combined. Crucially, the report notes those commitments are outpacing the firms' own earnings and free cash flow, pushing some of them to issue debt just to keep the buildout going.

• The five largest hyperscalers are set to spend over $1 trillion on AI capex from 2025 through 2026

• That spending is outpacing earnings and free cash flow, driving some firms toward new debt issuance

• One cited industry projection puts cumulative AI capex commitments at $3 to 4 trillion through 2030

• Growing bottlenecks in electricity, advanced semiconductors, and grid equipment are already pushing up power prices and input costs

• US stocks, heavily weighted toward AI-linked names, now make up roughly 64% of the MSCI Global index

Why would firms with uncertain returns keep raising the stakes? The BIS points to a winner-take-most contest dynamic: with a widespread belief that only a handful of players with superior technology will end up dominating the market, competitors may be over-committing to projects with genuinely uncertain payoffs just to avoid being shut out. Modelling that contest structure, the report finds that as competitive capex rises, the net economic surplus for the AI sector as a whole, total payoff minus investment cost, actually declines, and could turn negative in adverse scenarios.

The Financing Is Getting Harder to See Through

Part of what worries the BIS isn't just the size of the spending, it's how opaque the financing behind it has become. Hyperscalers, chip makers, and AI labs are increasingly linked through private arrangements the report calls circular financing: chip makers and hyperscalers take equity stakes in AI labs or "neocloud" providers, who in turn commit to multi-year purchases of chips or computing power, sending the capital right back around as reported revenue.

Layer onto that a wave of data-centre construction that's increasingly outsourced to third parties and leased back to hyperscalers on long-dated contracts, sometimes with the same underlying assets pledged more than once, and the report says these arrangements now account for a sizeable share of sector-wide financing and forward revenue. Credit spreads on some AI-linked issuers have already started to widen, even as equity markets keep pricing in significant further upside.

The BIS also flags a second, less obvious channel: engineering, procurement, and construction contractors that build out the physical data-centre footprint. Their balance sheets are comparatively weak, and the report notes they would be among the first exposed if hyperscalers ever slowed or paused their capex pace.

Four Futures, and One the BIS Invented a Name For

Beyond the near-term financing risk, the report's Box C takes a longer view, asking whether AI could break the historical link between human idea-generation and long-run economic growth. It sketches four scenarios rather than a single forecast.

Under "business as usual," labour remains the binding constraint and growth stays near its historical trend of around 2% a year. Under a "bounded productivity boost," trend growth shifts up by a constant margin, something the report notes hasn't happened since the Industrial Revolution. Under "transformative AI," the technology becomes a self-improving, self-reinforcing growth engine, output expands exponentially, and labour's share of income falls toward zero.

The fourth scenario is the report's own contribution to the debate: a "demand bottleneck." As automation shifts income away from displaced workers and into further AI investment, the consumer base that would justify continued capacity expansion can erode. Forward-looking firms, seeing a shrinking future market for AI-produced goods and services, may find it unprofitable to automate the next task, not because the technology hits a wall, but because the demand to justify further investment simply isn't there. In this scenario, growth rises initially, then falls below its historical trend as the bottleneck takes hold.

The scenarios matter beyond growth forecasting; they shape the report's read on interest rates. The natural rate of interest, or r-star, rises under the productivity-boost and transformative scenarios, but under the demand-bottleneck scenario it rises at first and then falls below its pre-AI baseline. Which path plays out, the report says, depends on things nobody currently has firm numbers for: how reliant AI profits are on consumer demand, how fast competition erodes margins, and how quickly AI infrastructure becomes obsolete.

Why This Time Might Hit Harder

The BIS argues a correction today could travel further than in past tech cycles, for reasons that have little to do with AI itself. Household equity exposure has grown substantially as a share of both wealth and income over recent decades, meaning a sharp valuation reset would likely produce a more pronounced wealth effect and consumption pullback than in previous downturns. And with US equities, heavily concentrated in AI-linked names, making up roughly 64% of the MSCI Global index, a US-led repricing has more room to travel internationally than it once did.

There's also a credit channel. The report notes that broad indices of credit spreads tend to move against stock returns, more so in high-yield than investment-grade debt, and that large, synchronised corrections in both markets are rare but not unprecedented, pointing to the Great Financial Crisis and the March 2020 "dash for cash" as examples. A repricing of AI-linked risk, the BIS warns, has the potential to trigger a broader corporate credit freeze, one that could spill into the less transparent private credit market that has expanded rapidly among mid-sized and small firms. That segment, the report notes, already shows early stress signs, including rising redemption requests at direct lending funds.

AI Investment Sustainability: FAQ

Jans Bock-Schroeder

Publisher & Founder of AI Angst

Coming from the world of art, photography, and the luxury market, Jans launched AI Angst in 2025 to explore the cultural, ethical, and psychological impacts of artificial intelligence. His work bridges creative vision with critical technology analysis, offering clarity in an era of rapid technological change.