The Tail of Two Risks

 

One enduring lesson from history and human evolution is this: We spend far more time imagining what can go wrong than what can go right. Yet risk has two tails. Elroy Dimson reminds us that risk means "more things can happen than will happen"— and yet investors naturally fixate on what they can see rather than what they cannot imagine.

 

AI illustrates this well. As the debates during the internet era (which I remember) and the mainframe revolution (which I don't), much of today's debate has centered on job destruction rather than job creation. Yet history suggests these fears rarely survive contact with reality. The Economist has found that AI has created around one million new jobs in America. They suggest the jobs apocalypse is postponed. An AI jobs boom is underway. Conveniently timed for the midterms, where AI and data centers are clearly on the agenda.

 

The same two-tailed logic applies to bonds and the broader macro picture.

 

What Might Go Right with Bonds

 

We remain cautious about bond yields — they are a concern, and we continue to monitor positions for signals warranting tactical adjustment. At current levels, however, our base case is the market can live with a higher cost of capital if it also believes productivity and future cash flows are accelerating. Bond volatility remains well-behaved. The left tail risk — the case for significantly higher yields — is clear and well-articulated by many capable investors. The right tail, though, receives less attention.

 

A plausible right-tail outcome might echo the late 1940s: a period when debt burdens grew sufficiently large that debt management began to matter as much as inflation and employment, leading to greater coordination between monetary and fiscal authorities. A modern version need not involve explicit yield curve control (though such arguably remains within the Federal Reserve's legal remit under the Third Mandate to maintain moderate long-term interest rates), but could emerge through a combination of indirect and direct debt management — buybacks, twists, foreign exchange interventions, closer Fed-Treasury coordination focused on bills, improved private-sector absorption of government bonds through banking deregulation, hedge funds and stablecoin issuers and most critically, stronger productivity growth.

 

The Congressional Budget Office has shown that productivity improvements can dramatically alter debt trajectories. Should AI generate productivity booms comparable to the 1950s or internet era, debt-to-GDP ratios could collapse well below 100%. In other words, investors may be overestimating the risk of uncontrolled yield rises while underestimating the possibility of greater Treasury-Fed coordination that prioritizes productivity growth, debt sustainability and financial stability.

 

Perhaps the greatest risk today is not that investors are underestimating the left tail. It is that they have stopped looking for the right one. One found in AI, which is precisely why we're cheering for it. We all should be.

 

Collateral as Canary

 

Another metric worth celebrating — and arguably the core of the AI bull case — is the current value of a graphics processing unit (GPU). Why? Because GPUs represent the financiability and underlying collateral behind compute, and ultimately, the credit in the broader ecosystem. As recent data from Silicon Data suggests, these remain stable. The A100 GPU, launched in May 2020, has hovered near $5,000 since late 2025. The H100 and B200 have actually appreciated. This is not the steep depreciation curve that sceptics would expect from a three-year technology cycle. Instead, compute has pricing power, cash flows are likely to accelerate, and Jensen Huang's observation that "compute is revenue" may be the most important insight from this entire earnings season. It's difficult to have a systemic credit crisis without a collateral crisis — and for now, the collateral is holding and in some cases, rising.

 

The Weakest Link

 

One thing is clear: If some form of financial repression ultimately emerges, the adjustment mechanism is more likely to be a weaker U.S. dollar and higher hedge ratios than widespread liquidation of U.S. assets. We simply don't see evidence of U.S. asset sales — quite the opposite. As the quantity of money increases, its quality decreases — a fact that supply-constrained real assets like gold and bitcoin should celebrate. Indeed, the performance of both since Treasury Secretary Scott Bessent’s buyback announcement (which some view as the opening signal for financial repression) is instructive and may foretell what's ahead.

 

Reconsidering Real Assets

 

As we’ve long discussed, as bonds' role in de-globalising markets comes into question, real assets deserve strategic reassessment. Bonds look less capable of providing protection in a world where supply-chain and trade shocks are becoming more structural, and where sovereign debt itself becomes a source of risk. Norway's Sovereign Wealth Fund — the world's largest — recently noted that "the degree to which bonds will dampen volatility in future crises depends on the nature of the crisis." Their defensive quality is regime-dependent. Should sovereign debt concerns become systemic, bonds may not diversify equities as reliably as investors assume. We may see higher yields demanded as compensation, or alternatively, real assets gaining greater strategic value. We remain constructively positioned toward real assets, with the view that diversification in a regime-shifting world may benefit from assets less correlated to fiscal stress and currency debasement.

 

Should History Rhyme?

 

Midway through the dotcom boom in 1997, Fed Chair Alan Greenspan tapped the brakes with a single 25-basis-point rate rise. Only one was needed because productivity improvements offset inflation pressures. The economy kept expanding. Equity markets kept rising. Interestingly, the ten-year Treasury yield peaked around 6.9% at that time, and by September 1998 it was below 4.5%, according to Bloomberg.

 

The fiscal situations between then and now bear little resemblance, and it's seldom wise to reason purely by analogy. That said, it's difficult to argue the internet created more efficiencies than AI: The internet was about communication; AI is about cognition itself. Should Fed Chair Warsh and team choose to tap the brakes in coming weeks — a possibility that's being discussed — it's unclear much would change. It might even be constructive, reinforcing Fed independence and inflation credentials.

 

Currency Dynamics and the Carry Trade

 

The yen is back in focus. Many macro clients are positioning for yen strength, not from repatriation fears (which appear overdone) but from the more forceful tone of Bank of Japan Governor Kazuo Ueda, who told Reuters at the G20 that "a rate hike will be discussed thoroughly in every meeting, including the next one." This followed a meeting with Bessent himself — information asymmetry worth noting. James Aitken of Aitken Advisers suggests this needn't portend a meltdown in the yen carry trade or mass dollar asset sales. Japanese investors may simply lift their dollar-hedge ratios. But the carry trade worth watching is the U.S. dollar itself. As discussed, most roads lead to dollar weakness and, by extension, greater global liquidity. That’s not bad news for risk assets.

 

Watching China's Move

 

Keep a close eye on China. Oil prices often signal regime shifts. As discussed weeks ago, the signal for an oil rally may not come from tankers transiting the Strait of Hormuz (which have recovered roughly three-quarters of pre-war traffic) but when China meaningfully re-enters the market. Well, China is back. Shanghai crude has recently traded above levels hit during the Iran war onset, breaching $110 per barrel. What's curious, at least to me, is the timing ahead of the late-September Xi-Trump summit—one where President Xi Jinping is rumored to be bringing a large business delegation, suggesting readiness for trade deals. I’m seeing little focus on this, although I suspect that will change.

 

And Finally

 

Tesla launched its fully autonomous Cybercab robotaxi the same week Volkswagen announced 100,000 job cuts. These headlines are surely linked. Thirty-seven is the average age at which people accomplish the feat that earns their obituary. And Nvidia, fittingly, is named after the Latin word for envy.

 

Less said about the rugby the better.

 

Have a great week.

 

Gary