RUNAWAY BOND YIELDS Dominique Marchese, 2026-10-01

Key words: Interest rates, Artificial intelligence, Oil, Financial markets.

Version PDF

While the conflict in the Middle East remains entrenched despite unsuccessful attempts at dialogue between the United States and Iran, fueling tensions in oil markets, long-term interest rates continue to climb, reaching record highs not seen in two decades. The nominal yield on the 10-year US Treasury note has reached 5.25%, its highest level since 2006, well before the subprime mortgage crisis. This surge in bond yields is global, even as the world economy demonstrates resilience (with volume growth at an annual rate of 3%), driven by the investment cycle in artificial intelligence (AI).

Interest rates are flirting with dangerous levels

The surge in long-term interest rates worldwide is, of course, the dominant theme this fall. Since the United States steers the world's financial markets, it's useful to briefly revisit dollar-denominated bond markets. To begin, without getting bogged down in the details, let's review what defines the equilibrium nominal interest rate (with economic activity and inflation stabilized), according to academic literature. Long-term (sovereign) interest rates can be broken down as the sum of a real rate and expected inflation. This real rate is itself the sum of the expected growth rate of economic activity in volume terms, the lender is thus compensated for its role in financing the economy—and a term premium. This premium is supposed to compensate for the duration risk borne by the investor, which corresponds to the uncertainty surrounding future inflation, the absolute enemy of the bondholder. This risk premium is also influenced by the degree of confidence in the debtor—the U.S. Treasury in this case—and the guarantor of monetary stability, namely the central bank—here, the Federal Reserve (Fed). Today, the 10-year Treasury yield breaks down as follows: 5.30% (nominal rate) = 1.8% (real growth) + 2.4% (expected inflation) + 1.10% (term premium). The 10-year real economic growth rate is the one estimated by the Congressional Budget Office, a highly reputable bipartisan agency that provides Congress and the public with various reports on the U.S. fiscal path and key macroeconomic indicators. When nominal and real interest rates rise due to improved economic prospects (for example, from expectations of productivity gains linked to AI), the situation is not problematic: the increase in the cost of capital, which affects all market participants, is offset by higher profit prospects for businesses and increased tax revenues for governments. However, a generalized rise in long-term interest rates due to creditors demanding an additional risk premium becomes more problematic. Before the summer, long-term dollar interest rates corresponded to an equilibrium level with virtually no term premium. Yet, it is precisely the reconstitution of this premium that is the real issue, since it does not stem from the expectation of increased economic activity in the future.

The reasons for the reconstitution of the term premium seem clear. First, the proliferation of shocks since the pandemic (supply chain crisis in 2021, war in Ukraine and energy shock in 2022, trade war in 2025, conflict in the Middle East in 2026, tensions fueled by the AI infrastructure investment cycle) makes the occurrence of shocks of comparable intensity and the subsequent rise in inflation volatility more likely in the eyes of investors: inflation will not only be higher compared to the 2010s, but also more volatile and less predictable. Second, questions about the credibility of the United States, which is allowing its federal budget deficit to balloon (-6.3% in 2026), and of the Fed are central to this term premium. Indeed, Treasury Secretary Scott Bessent's clumsy attempt during the summer to calm markets by announcing a manipulation of the yield curve (purchasing long-term Treasuries in exchange for issuing short-term debt), and the missteps in communication by the new Fed Chairman, Kevin Warsh—despite his more recent efforts—have contributed to rebuilding the risk premium on Treasury bonds. Finally, and no less fundamentally, US and global markets are subject to a growing ex-ante imbalance between available savings and the massive investment needs in AI (and more generally in digital technologies), electrification, the energy transition, sovereign infrastructure, defense, and the aging population. The convergence and scale of all these transitions are unprecedented in human history. The competition induced between economic actors (for example between the US Treasury and AI actors) to attract available savings naturally leads to an increase in the cost of capital which penalizes the financing of the global economy and the valuation of financial assets, in particular that of shares.

A few remarks on future inflation: while current prices are driven by oil prices, long-term expectations reflected in financial products (swaps, inflation-linked bonds) remain firmly entrenched on both sides of the Atlantic. Five-year forward inflation swaps have barely fluctuated in recent months, standing at 2.15% for the euro and 2.45% for the dollar (compared to inflation indices slightly above 3% over the past twelve months). One might think that this stability primarily reflects the credibility of central banks, which have decided to gradually tighten their monetary policy (the European Central Bank for the second time this year, with its deposit facility rate at 2.50%, a 0.25% increase in September; the Fed for the first time with a target range for the federal funds rate of 3.75%–4.0%). In reality, this is primarily a consequence of the stability of energy prices observed on the futures markets, which reflect a scenario of a gradual return to normalcy in the oil and gas markets starting in 2027, thanks to a satisfactory resolution of the conflict in the Middle East. We must be wary of granting the ECB any credibility whatsoever in the fight against inflation induced by an external energy shock. Its insistence on the risks of second-round effects (a price-wage loop whose signs are absent so far) demonstrates a dogmatism that it has always been adept at during times of crisis—apart from the blessed but bygone era of Mario Draghi's presidency. It is the reopening of the Strait of Hormuz that will bring prices down, not Frankfurt. The current crisis primarily highlights our loss of control over production and prices in entire sectors of industry (petroleum products, diesel, kerosene, etc., in this case). Long-term interest rates in euros now exceed the long-term nominal potential growth of the euro area, which is rather weak, barely above 3% per year (3.5% according to the ECB), including anticipated inflation, while the investment needs to restore Europe's competitiveness are gigantic.

A few final words on the oil market situation: the freeze in the Middle East conflict and the continued restrictions on traffic in the Strait of Hormuz justify crude oil prices above $100 a barrel, which is still manageable for the global economy. However, if the alternative route of the Saudi East-West pipeline is not quickly restored to operation and the security of the Bab-el-Mandeb Strait is no longer guaranteed, the total supply shock would likely be closer to 9 million barrels per day (approximately 9% of daily crude demand), which would quickly become unsustainable given the very low level of global strategic crude oil and petroleum product stocks. The question of the availability of refined products would become even more critical. A return of Brent crude to $120-130 a barrel (peaks reached during the 2022 invasion of Ukraine) cannot be ruled out. Washington is still betting on Iran's economic collapse, while Tehran is counting on the political pressure of the upcoming midterm elections in Congress. Which of the two adversaries will give in first?

AI: Manipulation and pressure on open weight models

September was a busy month for AI news. Faced with the explosion in token consumption (the smallest unit of data processed by AI), accelerated by the rapid spread of agentic AI, the apocalyptic pronouncements of proponents of closed models (Anthropic, OpenAI), security concerns, and the ever-fiercer competition from open-weight models, particularly Chinese ones, investors are struggling to understand the sector's dynamics. However, some certainties seem to emerge from what increasingly resembles the "fog of war" described by Carl von Clauswitz in a different context. To begin with, the AI revolution is an undeniable reality that is no longer confined to a simple closed circuit between chip manufacturers (Nvidia, Broadcom, Micron, etc.), hyperscalers providing computing power in the cloud (Amazon Web Services, Microsoft Azure, Google Cloud, Meta, etc.), and model developers (themes of financial circularity in the AI ecosystem and the risk of a bubble). Recent publications from major software vendors (Microsoft, SAP, Salesforce, etc.) demonstrate the ever-accelerating adoption of AI-powered applications by end users. Agentic AI (the automation of complex tasks) is taking off (the recent, acclaimed launch of MetaMuse is a prime example), leading to an exponential increase in inference (the use of trained models) and token consumption. Asset manager Schroders estimated in a recent report that AI-related services should now generate around $300 billion in annualized revenue. This amount may seem significant for a sector that was virtually nonexistent before 2022, but it is actually still far from sufficient. Investments in AI infrastructure (data centers, computing power) are in the order of $1 trillion this year and are expected to increase by at least 30% annually until 2030, reaching an annual total of $2.9 trillion. Nvidia, the market leader in GPU chips, estimates that investments will then be in the range of $3 trillion to $4 trillion annually. Expressed as a percentage of gross domestic product (GDP), these staggering sums are comparable in magnitude to those seen during the major industrial revolutions of the past (railways, electricity, the internet). Even more interesting is the theoretical calculation of the return on investment needed for AI to create sufficient value for all players in the chain (hyperscalers, AI platforms, end users). The total accumulated capital would thus be around $4.5 trillion by 2033-2034 (after depreciation). According to Schroders analysts, $5.4 trillion in annual revenue would need to be reached within seven years to justify the enormous efforts being made today. According to our calculations, this would represent approximately 3% of the estimated global GDP in 2033! This brief analysis helps to better understand the stakes and the enormity of the challenge. In any case, investors will need to be patient: it takes approximately three years from the establishment of a data center dedicated to AI to the full utilization of its computing capacity. It is worth noting, however, that the sectors currently benefiting most from AI deployment are software (coding), IT services, customer relationship management, and back-office operations. Ultimately, and depending on ongoing experiments, the deployment of AI in these business segments could generate between $2.2 trillion and $3.7 trillion in revenue annually, which is not insignificant.

To briefly revisit the controversies surrounding security issues and apocalyptic rhetoric, we emphasize the danger posed by the heightened competition from open-weight AI models. This clearly threatens the profit trajectory of Anthropic and OpenAI but simultaneously boosts the end-user demand for AI services. Open platforms, particularly those from China, have the advantage of offering performance close to that of closed models (sometimes only a few months behind frontier models, according to experts), but at significantly lower token costs. The availability of these open platforms largely explains the explosion in their use. Furthermore, unlike closed models designed to operate in the cloud (with architecture and parameters not made available to the user), an open-weight AI is free and fully downloadable (except for its training data) on the infrastructure of companies concerned with controlling the use of their own critical data. This sheds light on the panic among American private companies, who, just months before their IPOs, are attempting, through dubious means (security blackmail by arsonist firefighters?), to legally block the dissemination of open-weight models. It's worth remembering that China considers large-scale language models a commodity, far removed from artificial general intelligence. Beijing aims to create value by bringing together AI, robotics, and cutting-edge industries. It should be noted that the warnings about generative AI—which have flooded social media and mainstream media —are not shared by all experts (see, for example, the less-than-friendly comments of Yann LeCun, Turing Award winner), and that open-weight models are championed by Nvidia, Microsoft, IBM, and more generally by all players who don't want to be tied to a single platform.

In conclusion, we can be confident that the investment cycle is far from over. In the United States, according to a recent Federal Reserve report, it accounts for 30% of real GDP growth over the past eighteen months. Furthermore, domestic consumption is driven by the wealthiest 10% of American households (who own approximately 90% of personal stock holdings and account for 50% of consumption), who benefit from the wealth effect stemming from Wall Street's robust performance. The shift from model training to inference and agentic AI will further boost computing power requirements. Physical constraints (access to the electricity grid, water, and the most advanced electronic chips, etc.) and regulatory constraints (growing opposition from local communities and the public to the inconveniences caused by data centers; paralyzing European directives) will remain the only obstacles to this industrial revolution for a long time to come—a revolution destined to succeed, as the reader will have surely understood.

Conclusion

We can only reiterate our advice on prudence and sound diversification, which we issued during the summer. Equity indices, supported until now by AI and the strong financial health of companies (consensus 2026 profit forecasts of +29% and +16% respectively for the US and the Eurozone), are vulnerable to a potential continuation of the bond market correction and a rise in the cost of capital. The ex-ante imbalance between the massive investment needs and available global savings, in a context of intense competition between indebted states and the private sector, can only exacerbate tensions on the cost of credit, further aggravated by international crises.

Disclaimer :

The above content was produced and is distributed by Pure Capital SA. It is provided for informational purposes only and aims to present the management activities of Pure Capital SA (hereinafter "Pure Capital") and to provide information about its investment strategies. The information and data (including text and photographs) contained in this document are protected by copyright, and any reproduction or distribution thereof to third parties, in whole or in part, without the prior consent of Pure Capital, is prohibited.

Pure Capital reserves the right, at any time, to modify or change the data contained in this document without prior notice.

This document should in no way be construed as an offer to buy or sell any financial instrument or as any form of solicitation or inducement to buy or sell financial or investment instruments. All information published herein is provided for informational purposes only and does not constitute investment advice. Nevertheless, no guarantee can be given as to its accuracy or completeness. This information has not been prepared in accordance with legal requirements promoting independent investment research and should therefore be considered marketing material. Although this content is not subject to any restrictions on its use prior to its dissemination (e.g., for executing orders), Pure Capital does not seek to profit from it.

For further details concerning the investment products, the official sales documents (the Key Investor Information Document, the prospectus, the half-yearly report and the annual report) are available free of charge upon request from Pure Capital SA (hereinafter “Pure Capital”) (tel.: +352 26 39 86) or by consulting the website www.purecapital.eu.

The information contained in this document is provided in good faith and does not constitute a commitment on the part of Pure Capital.

There can be no guarantee that investment objectives will be achieved. Past performance is not a reliable indicator of future results. Performance may vary over time. The net asset value of the portfolio depends on market trends. Therefore, all investment funds are subject to market fluctuations, and investors may receive less than their initial investment.

Annual custody fees, or deposit charges, may be levied by your account holder. These vary from one institution to another. Please inquire with them for details.

This document is provided to you on the condition that it will not constitute a primary basis for any investments. Any investment should be the result of a fully informed and considered decision.