Boards have stress-tested the AI upside harder than anything on the risk register. Almost none have modelled the downside with the same rigour. The real governance question is not the AI you run. It is your exposure to an AI market that could correct, and that exposure is a number your CFO owns.

Many boards I sit in front of have modelled the AI upside. Productivity. Margin. Growth. Competitive position. The deck is thorough, the numbers are ambitious, and the enthusiasm is genuine.

I have yet to see one model the downside with the same rigour.

That is the asymmetry that should concern you, and your CFO.

I run agentic AI in production at Cyber Impact. I believe in the technology. I have written about where the AI money is actually moving, and why the operators capturing it have treated governance as the unlock rather than the restraint. None of that changes the question in this piece. If anything, it sharpens it. The more real the boom, the more real the correction, and the correction is the part nobody has put in front of the board.

The board's attention is pointed one way

Look at how AI arrives in a board pack. It arrives as opportunity. Time saved, roles augmented, revenue defended, a competitor being outrun. The business case assumes model capability keeps improving, pricing stays where it is, and access holds. Every one of those assumptions is doing quiet work in the financials, and none of them is owned by anyone in the room.

The downside, when it appears at all, appears as an operational footnote. Model risk. Data handling. A line about vendor management. Useful, but small.

That is not a risk register. It is a sales document with a compliance paragraph stapled to the back.

The upside has been over-modelled and the downside under-modelled. That gap is not an AI problem. It is a governance one.

Boards are good at this discipline everywhere else. They stress-test currency exposure, supplier concentration, refinancing risk, a key-customer loss, a commodity shock. They know how to ask what happens if the thing they are counting on stops being true. On AI, that muscle has not switched on.

What a correction looks like on your balance sheet, not on the news

An AI correction is not an abstract macro event that happens to other people. It shows up in four concrete places, and most boards have mapped none of them.

The capex you have committed on an AI thesis. Your investment case assumed a trajectory: capability up, cost down, availability stable. If the cycle corrects, the pricing you modelled and the roadmap you relied on may not hold. The question is simple and rarely asked. If the market re-rates AI tomorrow, does the business case you approved still clear your hurdle rate, and does your CFO own that number when it moves?

Operational dependence on a handful of frontier labs. A small number of providers now sit underneath a growing share of real work. APRA said this out loud in April, naming concentration risk where entities depend on a single provider for multiple use cases with thin contingency planning. Concentration is not only a resilience issue when a provider goes dark. It is a commercial issue when a provider reprices, because you have very little leverage over a dependency you cannot easily replace.

Supplier and sector exposure you did not choose. Your own AI spend is the visible part. The invisible part is everyone around you. Suppliers whose margins assume the boom continues. Customers whose budgets are propped up by the same sentiment. If you are ASX-listed, an index and a valuation multiple carried in part by a cluster of AI-exposed names you do not control. The exposure is in your ecosystem whether or not you bought a single token.

What re-rates if the thesis softens. This is the one to put in front of the board and the CFO. If AI-related valuations reprice, what actually moves in your P&L, in your balance sheet, in your supplier base and your customer base? If the answer is "we have not modelled that," you have found the gap.

The BIS put a number on the wiring

On 28 June 2026, the Bank for International Settlements released its Annual Economic Report 2026. The BIS is the central bank for central banks, and it has a track record of flagging systemic risk before it crystallised, including in the run-up to 2008. It does not write for headlines.

It wrote this: "Disappointment in returns could trigger a sudden pullback in financing and turn the capex boom into a protracted investment bust, with potential knock-on effects on financial conditions." It added that a major equity-market correction "could have larger macroeconomic consequences today than in the past," and warned that a repricing of risk, whether triggered by higher rates or an AI bust, had the potential to be as disruptive to credit markets as the 2008 global financial crisis.

The detail that should hold your attention is the plumbing. The five largest hyperscalers are on track to spend more than a trillion US dollars on AI-related capex across 2025 and 2026, a pace the BIS notes is already outrunning their earnings and free cash flow, with some issuing debt to cover the gap. Underneath sits what the BIS calls circular financing: chipmakers and hyperscalers take equity stakes in AI labs or neocloud providers, which then commit to multi-year purchases of chips or computing power from those same investors, with data-centre construction leased back on long contracts. The terms, in the BIS's words, are "typically poorly disclosed, with risks of the same asset being pledged multiple times." Concentration compounds it. On market reporting around the report's release, the ten largest companies in the S&P 500 now account for roughly 36 to 40 per cent of the index.

I am not forecasting a crash. Neither is the BIS. The point is narrower and more useful. The financial architecture under the AI boom carries risk that has not been fully mapped, and if you have not mapped your own slice of it, you are carrying an exposure you have not priced.

This is the same thesis, one layer out

I have argued before that frontier AI has moved from a technology choice into an operating dependency, and that nobody had properly solved AI governance until it stopped being a policy document and started being a control. This is the same argument at the level of capital.

Governance is not just controlling the AI you deploy. It is governing your exposure to an AI market that could re-rate. The dependency map I keep asking management to build should not stop at "which workflow uses which model." It should extend to "what in this business assumes the AI investment cycle keeps running, and what happens to us if it does not."

What your board and CFO should ask this quarter

None of these are theoretical. Each has a named owner or it does not count.

  1. If our AI business case assumed capability, pricing or availability the market no longer supports, does the investment still clear the hurdle rate, and who owns that number when it moves?
  2. If a primary frontier provider repriced, degraded or withdrew for ninety days, which critical operations stop, slow or cost materially more?
  3. What is our exposure through suppliers, customers and index, not just our own spend, and what moves in our P&L and balance sheet if the sector re-rates?
  4. Have we committed capex, debt or long-dated contracts on the assumption the boom continues, and are there embedded exit clauses or concentration we have not disclosed to ourselves?
  5. Who at this table owns the downside case, and when did we last stress-test it with the same rigour we gave the upside?

If the answer to the last one is "we have not," that is the work. Not because a bust is certain. Because the exposure is already on the balance sheet, and the only choice you have is whether you understand it before the market makes you.

The board that can answer these questions is not the one with the longest AI strategy. It is the one that treated its AI exposure like every other material risk, and priced both sides of it.

If your board has modelled the AI upside but never priced the downside, that is a conversation worth having.

Sources

  • Bank for International Settlements, Annual Economic Report 2026, "Progress and peril", 28 June 2026, bis.org (report and press release).
  • Fortune, "The central bank of central banks just released its flagship annual report and it sees a $1 trillion AI investment boom headed for a reckoning", 29 June 2026.
  • The Next Web, "The BIS warns an AI bust could hit credit markets as hard as the 2008 financial crisis" (S&P 500 concentration and 2008 comparison), June 2026.