The greatest investment risk in today's market may not be a data surprise. It may be a change in how investors interpret the data they already have. That is the provocative central argument of Carlyle's latest whitepaper, Rules of Relevance1, authored by Jason Thomas, Head of Global Research and Investment Strategy. Drawing on literary theorist Paul Fry, Thomas argues that human minds build "rules of relevance" that render certain information "functionally invisible" before it can be interpreted. In markets defined by rapid technological and geopolitical change, that blind spot may be the biggest risk of all.
Two Observers, One Boom
The AI compute capital expenditure explosion is Thomas's clearest illustration. A scaling engineer at a frontier lab and a time-series statistician look at identical data and see entirely different objects. The engineer sees a rational effort to relax a technically binding constraint. The statistician sees an accelerating series without precedent in the sample. Both are right. They simply notice different things. Market pricing has sided with the engineer. As Thomas states, "Asset prices do not represent a weighted average between these two perspectives; they reflect the framework operative at the clearing price." Risk premia have declined steadily throughout the buildout even as the range of probable outcomes widens exponentially. The statistician's concern remains functionally invisible.
External Finance, Internal Fragility
The stakes rise as the buildout shifts from internally to externally financed. More than 80% of anticipated AI capex through 2030 will require external financing through equity, debt, and off-balance-sheet structures. Thomas's analysis of the $2.5 trillion in contracted compute backlogs reveals only roughly $375 billion in net economic value, or about 15% of the nominal contract base. Credit enhancements have obscured the market's view of the true economics. The parallel Thomas draws is pointed: the mid-2000s, when creditors underwrote guarantors like Citi rather than the collateral backing their CDOs.
Interest Rates: The Savings Glut That Was
The same perceptual shift is underway in fixed income. Bond yields spent years anchored by expectations of sub-3% normalization, a framework built on a corporate savings glut that no longer exists. Net corporate sector savings are down 86% relative to the equivalent point in the previous cycle, as AI data centers now absorb savings once channeled to banks, governments, and shareholders. "There's been nothing special, or duration-specific, about the rise in Treasury yields," Thomas states. "They've simply responded to the rise in forward rates." The old lens has cracked.
Globalization's Hidden Subsidy
The unwinding of global trade carries a cost few have priced. Dollar-invoiced global production networks created sustained demand for Treasuries that suppressed US borrowing costs for decades. With trade volumes reversing, Treasury issuance now outpaces external demand. Thomas is direct: "With the funding benefits of 'globalization' occluded from public perceptions, it's been easy to recast the whole era as a massive loss for US interests." Should foreign reserve managers become net sellers of Treasuries, today's rate complaints may seem modest by comparison.
Europe's Perspectival Mirage
Europe's peace dividend, Thomas argues, may be a perspectival illusion. Domestic value-added accounts for barely €20 of every €100 of hardware acquired. Defense investment in the US, China, and Israel generated significant commercial technology spillovers while Europe's defense-industrial base atrophied alongside its global technological relevance. Higher domestically produced defense output may ultimately reveal the peace dividend for the mirage it has always been.
The animating risk across every one of these themes is the same: "when the framework itself is at issue, valuations disperse and asset prices become susceptible to 'jumps' as the same phenomenon gets viewed through a new lens."
5 Key Takeaways for Advisors and Investors
- Direct AI exposure is likely sufficient. AI-exposed stocks account for over two-fifths of US market capitalization. Incremental direct exposure should offer upside convexity to offset repricing risk, not deepen an already-dominant consensus.
- Technology revolutions reward users, not builders. From railroads to the internet, economic value accrued downstream. In ten years, AI gains may appear in unexpected places across portfolios.
- Fiscal risk surfaces as inflation, not default. The Treasury borrows in a unit it manufactures. Duration in long bonds is not protected by credit quality; it is compounded by the inflation channel that fiscal pressure opens.
- Credit beats duration. USD cash yields sit at 3.65%. Reaching beyond that requires duration or credit risk. Inflation benefits credit issuers mechanically; it erodes fixed principal.
- Defense and energy are the deglobalization plays. Economies must power and defend themselves. Supply shortfalls in fuel, electrons, and defense hardware carry consequences no price adjustment resolves quickly.
Footnote:
1 Thomas, Jason. "Rules of Relevance." The Carlyle Group, Sept. 2026, https://www.carlyle.com/sites/default/files/2026-09/Rules-of-Relevance.pdf.
