When the bubble pops, the only AI spend that survives is the kind that redesigned actual work

Moody's counts $662 billion of data-centre lease commitments that aren't on a balance sheet yet. When they land, the correction deletes the spend that never became capability — and can't touch the work you actually redesigned.

Andrew Ramsden 11 August 2026

You don't say "bubble" in the exec meeting. Half the room holds the stock, the other half wrote the strategy, and the word costs more than it clarifies. And yet on the drive home you have run the sum more than once. If this corrects, this year or next, what happens to our programme? Which lines of our own AI budget would look embarrassing in the write-down year?

Most of us suspect the answer is "most of them". You can find out precisely in one afternoon, and it is much better to know before anybody else does.

The numbers behind your arithmetic

Moody's analysts counted about $662 billion of future data-centre lease commitments across Amazon, Meta, Alphabet, Microsoft and Oracle. They are signed but not yet started, so under current accounting rules they sit off the balance sheet until the leases commence.

The Bank for International Settlements, the central bank of central banks, used its 2026 annual report to warn about an AI bubble bursting, debt-funded capital spending, and contagion through private credit.

The demand side is where it gets uncomfortable. MIT researchers analysed 300 public enterprise AI deployments and found that 95% of pilots delivered no measurable impact on profit and loss. Their report also leans on 52 executive interviews and a survey of 153 leaders, and its critics say it defines a pilot loosely, so I would discount the figure rather than quote it as gospel. Even discounted heavily, it describes the same gap the BIS is pricing: an infrastructure build-out of historic scale, resting on enterprise returns that mostly have not arrived.

So yes, the shape of a bubble. The debate about when it pops is fully staffed without us.

The wrong layer

That debate runs entirely at the market layer: valuations, capital spending, circular financing, who is holding whose equity. The market layer prices companies. You run one.

At your layer, the history of bursts says something almost nobody quotes. A bubble bursting does not delete capability. It reprices it. The correction destroys spend that never became capability, and it leaves redesigned work almost untouched, because redesigned work does not trade.

The dot-com crash is the clean experiment. Amazon's shares fell from about $106 in 1999 to $6 in 2001, a drop of roughly 94%, and its revenue grew more than 140% from 1999 to 2002, straight through the crater. The crash repriced the stock. It could not reprice the warehouse network, the one-click checkout or the rewired retail operation. Those were the work. Pets.com died in the same crash for the mirror-image reason: it was market story all the way down, with no redesigned work underneath to survive the repricing.

The companies that had genuinely rewired their operations for the web walked out of the wreckage owning the next two decades. The ones that had bought web presence, which is spend wearing capability's clothes, wrote it all down.

That is the morning after. It does not ask what you spent. It asks what the spend became.

The correction has already started — inside

We don't have to wait for the market to watch the sorting begin. It is running now, quietly, inside budgets.

S&P Global Market Intelligence surveyed more than 1,000 organisations across North America and Europe. The share of them abandoning most of their AI initiatives rose to 42%, up from 17% the year before, and the average organisation scrapped 46% of its proofs of concept before they reached production. Gartner, polling more than 3,400 organisations that are actively investing in agentic AI, predicts more than 40% of those projects will be cancelled by the end of 2027.

The private correction runs ahead of the public one. Finance teams are already doing to AI line items what the market has not yet done to AI stocks. The write-down year starts one budget cycle at a time.

Which puts us in a squeeze. Keep spending the way most programmes spend, tools first, adoption dashboards, workflow redesign left to emerge, and we manufacture the very line items a correction deletes. Freeze instead, and we hand the redesigned-workflow advantage to whichever competitor kept going through the winter, which is exactly when capability is cheap to build. Both roads are expensive. Waiting is the first road, taken slowly.

Three columns

So run the sorting yourself. Every line of AI spend in your company lands in one of three columns.

Sand is spend that evaporates in a correction, because it never attached to any work. Licences bought ahead of need, pilots that ended in a demo, platform subscriptions bought to be ready, the innovation team's showcase, adoption numbers with no workflow behind them. Sand can be enormous, and green on every dashboard. Its tell is simple: if it stopped tomorrow, no process would run differently. Most of us have signed off a line like this, and signed it off because of the year it was rather than because of a job it changed.

Scaffold is spend that enables the building without being the building. Training, enablement, data clean-up, integration work, the champions programme. Scaffold is honourable, and nothing gets built without it. And yet scaffold left standing with no building inside it is only slow sand. It points at capability that is supposed to arrive later, and "later" is the whole risk.

Asset is a named workflow that runs measurably differently because of the spend. Claims triage cut from ninety minutes to nine. A month-end close that lost a week. An owner who can say so in numbers. Asset survives a correction, because a repriced model or a cheaper vendor does not un-redesign your work. Amazon's warehouses did not care what the stock did. An asset is a line you would keep paying for even if the AI market halved.

If you mapped your own budget while reading that, you probably know the problem already: honestly sorted, the asset column is thin. That thinness, rather than the Nasdaq, is your exposure.

The Morning-After Ledger

Pull the AI budget, every line over a threshold you choose, and put each line through nine questions, three per column.

The sand questions:

i) If this line stopped tomorrow, which process would run differently? No answer means sand. ii) Has it produced anything beyond a demonstration, a pilot report or a dashboard? No means sand. iii) Was it bought because of a named workflow, or because of the year it was? The second answer means sand.

The scaffold questions:

iv) Does it exist to enable a specific named build, or capability in general? A named build is scaffold. In general is sand. v) Does the thing it enables have a delivery date and an owner? Yes, hold it as scaffold. No, reclassify it as sand at the next budget. vi) Is what it was scaffolding still funded and alive? If the building died, the scaffold is sand standing up.

The asset questions:

vii) Name the end-to-end process that runs differently, and name its owner. Both names, or it is not an asset. viii) State the change in cycle time, cost or quality, in units. "Feels faster" is scaffold at best. ix) If AI prices doubled or the vendor vanished, would the redesigned process keep most of its gain? Yes is asset. No is scaffold wearing an asset's badge.

Then compute one number: the sand ratio, which is sand dollars over total AI dollars.

Under a third, the programme compounds. It survives a burst and comes out of it ahead. Protect the owners of the asset column above everything else.
Between a third and two-thirds, you are half-built, with real assets and real exposure. The move is not more spend. It is converting named scaffold into named asset before the cycle turns.
Over two-thirds, you are exposed. Most of the programme reprices to zero in a correction, whatever the dashboards say. Knowing that before the market does is the whole advantage the afternoon buys you.

One honest boundary: the ledger cannot time the burst. Nothing can, the BIS can't and Moody's won't. It can only make sure that whenever the morning after arrives, your budget is already sorted into what a correction deletes and what it cannot touch.

The winter advantage

For a company whose spend is mostly asset, a burst helps. Capability gets cheaper, talent becomes available, and competitors whose programmes were sand spend two years explaining write-downs while your redesigned workflows keep compounding, much as the web kept compounding after 2001 under new owners.

Start with the first question, on your three biggest AI lines, before Friday. If none of the three passes it, you don't need the other eight to know which column your programme lives in. And if the ratio comes out worse than you hoped, it is a map of where the redesign never happened, which is a far more useful thing to carry into a strategy conversation than a view on when the market turns.

Attribution

Written by Andrew Ramsden. AI tools assisted research and drafting; all outputs verified.

Accountable

Andrew Ramsden.

Limitations

The MIT 95% is quoted from a legal-industry write-up of the report; the underlying base is 52 interviews and 153 survey responses. Moody's figure is quoted from Fortune's report of the analysis, not the Moody's note itself. Amazon's share prices come from a book-summary blog and are rounded.

References

Eight sources, retrieved and hashed on 26 and 27 August 2026, each figure quoted with a locator in the SOURCED sidecar.