Can an AI Crash Save the Property Market?

The bubble is going to burst. Here’s where all that money actually goes — and why it isn’t into your mortgage.

There is a comforting story doing the rounds right now: when the AI bubble finally pops — and treat that as the working premise of this piece, not a hedge — the trillions of dollars sunk into GPUs and data centres will come crashing back to earth, dragging asset prices down with them and finally letting ordinary people buy a house. It has the shape of poetic justice. The money that priced you out of the market gets vaporised, and the market resets in your favour.

It’s the wrong question. The right question isn’t “will it burst” — the unit economics below make a fairly compelling case that it has to, in some form. The right question is: when it does, where does that money actually go? Because “it evaporates and lands in housing” isn’t one of the real options. Once you trace the actual plumbing — who spent it, who lent it, and who’s on the hook — you end up somewhere much less comforting than a discount on a three-bedroom house.

Option A: Where does the money go? Nowhere. It was never cash.

The first thing to understand is that most of the “trillions” isn’t sitting in an account waiting to be reallocated. The Big Four hyperscalers — Amazon, Microsoft, Alphabet, and Meta — are on track to spend somewhere between $725 billion and $760 billion on AI-related capital expenditure in 2026 alone, up 77% from an already-record $410 billion in 2025 (Yahoo Finance; Statista). Goldman Sachs has pushed its five-year forecast for the same four companies to $5.3 trillionthrough 2030 (Yahoo Finance).

That money isn’t parked, it’s spent — poured into concrete, silicon, and power contracts the moment it’s raised (Futurum Group). If the bubble bursts, this capital doesn’t “move” anywhere. It’s already a data centre in Texas or a GPU cluster in Ohio. The bursting doesn’t free the money up — it just reveals that a lot of it was mispriced. A written-down asset isn’t a liquidity event for anyone except the people forced to write it down.

Option B: It goes to the lenders — and this is the part people miss

Here’s the detail that actually matters for the “where does it go” question. A huge share of this build-out isn’t funded from corporate cash reserves at all — it’s funded with debt, and increasingly debt that doesn’t sit on the hyperscalers’ own balance sheets.

Morgan Stanley estimates global data-centre capex will hit roughly $2.9 trillion through 2028, while Big Tech’s own operating cash flow covers only about $1.4 trillion of that. The remaining $1.5 trillion gap is expected to be plugged mostly by private credit (Lynk Capital Markets):

Hyperscalers issued roughly $121 billion in bonds in 2025 alone — more than four times their five-year average — and total data-centre debt issuance nearly doubled to $182 billion (Quinn Emanuel client alert). Apollo’s chief economist estimates the “AI ecosystem” could ultimately need to support more than $2 trillion in debt, with roughly $1 trillion of that potentially too large for public bond markets to absorb — meaning it falls to private credit instead (Forbes).

This is the actual answer to “where does the money go” if the bubble bursts: it goes into losses on the balance sheets of the private credit funds, insurers, and pension vehicles that financed it. Private credit funds — Blackstone, Blue Owl Capital, Apollo, PIMCO, and BlackRock among them — originate most of this lending, often through opaque, off-balance-sheet vehicles (Quinn Emanuel). The US Federal Reserve has found that up to a quarter of bank loans to non-bank financial institutions now go to private credit firms, up from just 1% in 2013 — and that major life insurers now hold nearly $1 trillion tied up in private credit (Quinn Emanuel). In January 2026, four US senators formally warned regulators that the sector’s reliance on “complex and opaque debt markets” could trigger destabilising losses across an interconnected set of financial institutions (Quinn Emanuel). By mid-2026, 34% of fund managers surveyed by Bank of America named hyperscaler capex the single most likely source of the next systemic credit event — double the share who said so a month earlier (Bloomberg).

That is structurally the same shape as 2008: hard-to-value debt, distributed through insurers and non-bank lenders, that looks fine until the underlying collateral (in this case, AI revenue that doesn’t yet exist) fails to show up. If it goes bad, the money doesn’t reappear in Sydney or St Louis as cheaper housing credit. It disappears as a write-down on an insurance company’s balance sheet, or a pension fund’s private-credit allocation — assets that are, if anything, adjacent to yourretirement savings, not your deposit.

Option C: It goes to Nvidia and a handful of chipmakers — who don’t build houses

Some of it genuinely does convert into revenue, just not anyone’s mortgage fund. Nvidia’s data-centre business posted $89.0 billion in a single quarter in mid-2026, up 117% year-on-year (The Motley Fool). Broadcom, AMD, and Intel are all posting AI-driven data-centre revenue growth of 60–180% (The Motley Fool). That’s real money — but it flows to shareholders, executive comp, and further chip R&D, not into the residential construction or credit markets that set home prices.

Underneath this sits a genuine circularity problem. OpenAI is projected to lose somewhere between $14 billion and $33 billion in 2026 on roughly $20–25 billion of revenue (AI Business Weekly). Internal documents reported by the Wall Street Journal reportedly show its losses swelling toward $74 billion by 2028, before a hoped-for pivot to profit only from 2030 (Fortune, reporting WSJ).

Microsoft owns a large stake in OpenAI and is also its biggest cloud supplier — roughly 45% of Microsoft’s $625 billion cloud backlog is tied to OpenAI commitments (RD World). Nvidia has pledged up to $100 billion toward bringing Stargate’s capacity online — capacity that will largely be filled with Nvidia’s own chips (AI Magicx / OpenAI stats roundups). Oracle’s Stargate exposure has stretched its balance sheet enough that Barclays has flagged its credit rating could edge toward junk status if spending keeps outrunning revenue conversion (RD World). Money moves in a tight loop between a handful of counterparties, each booking revenue somewhere along the circuit, without a proportionate amount of genuinely new capital entering the system. Tech commentator Ed Zitron has spent two years arguing publicly that the compute demand implied by these valuations doesn’t exist at the scale claimed, and that unprofitable labs are being propped up by hyperscalers with pockets deep enough to absorb the losses — for now. The financial facts above (the losses, the backlog concentration, the credit-rating warnings) stand on their own in company filings and analyst notes regardless of how you weigh his framing.

Option D: It goes into job losses — and that’s the part that actively hurts housing

If there’s a channel connecting an AI bust to the housing market, it runs through employment, not liquidity — and it moves prices the wrong way for buyers.

US tech layoffs passed 165,000 through mid-2026 on some trackers, with AI cited as a factor in roughly half of Q1’s cuts (Tom’s Hardware; TechCrunch). Meta cut roughly 8,000 roles even while reporting strong earnings; Oracle disclosed a 13% headcount reduction over twelve months (TechCrunch). White-collar sectors — finance, IT, professional services — have shed jobs on net for three straight years despite continued GDP growth (Metaintro). This is precisely the buyer pool that has underpinned demand in mid-to-upper-tier housing for the past decade.

A serious correction plausibly accelerates this — more layoffs, tighter corporate spending, and, if the private-credit exposure above starts to sour, tighter bank lending standards generally, including mortgage underwriting. That’s not a lower bar to homeownership. It’s a higher one: shrinking incomes and shrinking credit availability, at exactly the moment asset prices might be soft. A recession that hits the exact demographic that currently qualifies for a mortgage is not an affordability fix. At best it’s a stall. At worst it’s a broader credit crunch.

Why none of this touches the actual housing problem

Even setting the debt structure aside, real estate runs on a completely different balance sheet to the AI trade, for reasons that predate the AI boom by well over a decade.

The US is short roughly 1.2 million homes (NAHB), and the number of listings affordable to a household earning $75,000 or less has fallen 60% since March 2019 (Harvard JCHS). At a median new-home price of roughly $413,600 and a 6% mortgage rate, about 88 million US households are priced out entirely — and every extra $1,000 on the median price prices out another 156,000 households (Eye on Housing / NAHB). More than half of American households earn under $80,000 a year — well short of what’s needed to qualify for a median-priced mortgage (Eye on Housing).

None of that is downstream of Nvidia’s share price. It’s zoning, construction labour shortages, materials costs, and a decade of under-building. A crash in tech valuations doesn’t pour concrete, approve a rezoning application, or train a carpenter. If AI-linked equities lose value, that wealth destruction happens almost entirely within capital markets — venture funds, index funds, corporate treasuries, sovereign wealth funds — with no mechanical link to what a house costs in a supply-constrained metro.

The math the “AI crash will save housing” theory always skips

This is the core of it. Home prices are, in the end, a function of what buyers can borrow and repay. An AI bust destroys paper wealth and jobs; it does nothing to fund the wage growth needed for existing borrowers — let alone new ones — to service loans at 6%-plus rates on $430,000+ medians. Even in the scenario where a crash did knock 15–20% off home prices via a broad demand shock, it would do so by making the remaining buyer pool poorer and more credit-constrained, not by making housing attainable in real terms. You cannot fix an affordability crisis rooted in stagnant wages by triggering an event that stagnates wages further. The bubble bursting doesn’t create a single dollar of new wage capacity anywhere in the economy — it just relocates who’s holding the loss.

The quieter, more likely version of this

There’s a case that this doesn’t even need a dramatic “burst” to matter. MIT’s Media Lab, through its NANDA initiative, found that 95% of enterprise generative AI pilots were failing to deliver measurable financial return, with only a narrow slice of tightly integrated, operational use cases showing clear payoff (Healthcare IT News; Medium). That gap between capital expenditure and demonstrated enterprise value is arguably the more likely trigger for a slow deflation — boards asking harder questions about a trillion-dollar bet — rather than a single dramatic collapse.

It’s worth stating the counter-case fairly, too. Some economists caution against reflexively blaming AI for white-collar layoffs, arguing companies are using it as cover for cuts they’d have made anyway after pandemic-era over-hiring — a point OpenAI’s own leadership has echoed (Tom’s Hardware). Infrastructure bulls also note that inference demand — running AI models day-to-day, as opposed to training them — continues to outstrip available supply, which is a genuine argument for continued build-out rather than pure speculation. The debate over how much of this spending is rational versus overextended is real and unresolved.

The bottom line

If the AI bubble bursts, the money doesn’t flow anywhere near a mortgage. It’s absorbed as equity losses for institutional and wealthy shareholders, as credit losses for the private funds, insurers, and pension vehicles that financed the debt, and as job losses for the white-collar workers who make up a large share of the current buyer pool. None of those three outcomes manufactures the one input that would actually fix affordability: incomes rising faster than debt service costs. A financial correction that makes buyers poorer and lenders more cautious at the same time isn’t a rescue plan for the property market — it’s a second crisis, stacked on top of the one nobody has fixed yet.

Not financial advice. This article is a factual and analytical piece for general information purposes only. It is not personal financial, investment, or property advice, and nothing here should be treated as a recommendation to buy, sell, or hold any asset. Markets, forecasts, and figures cited below can and will change. Speak to a licensed financial adviser before making decisions based on any of this.

Sources

Note: this piece is anchored in US data, which is where the AI-capex and credit discussion is most concentrated. The same debt and labour dynamics are increasingly relevant to other developed housing markets, including Australia’s.