Treasury yields hit 4.94% amid Iran war and AI debt fears

Bank of Japan lifts rate to 31-year high as central banks tighten

The US 10-year Treasury yield closed at 4.94% on September 20, 2026, according to Federal Reserve data — near its highest level since 2007, according to The Guardian. Government borrowing costs have climbed as oil prices held above $100 a barrel and as fighting in the Middle East continued without a clear resolution. John Higgins, the chief economic adviser at Capital Economics, said the 5% level is “seen by some as a threshold above which financial markets might go into meltdown,” though he and his colleagues are not convinced that 5% is a “magic” number. Bloomberg macro strategist Simon White has calculated that trouble would “really start if the US 10-year Treasury yield rose above 5.25%,” calling that level the historical “inflection point where stocks and bonds have reinforced losses in one another.”

Government bonds have been sold off in recent weeks, pushing up yields. Higher bond yields make equities less appealing because equities carry more risk than government debt. The Guardian reports that the Iran war has stoked concerns about higher inflation and that President Trump’s tax and spending plans — driving Washington’s debt levels above $40tn (£29.9tn) — have also worried investors.

The Federal Reserve approved its first interest rate increase since 2023 — a decision The Guardian characterized as defying President Donald Trump despite the administration’s stated concerns about the trajectory of monetary policy. The European Central Bank raised rates the previous week, citing the hit to the eurozone from the escalating conflict. On Friday, the Bank of Japan raised its policy rate to the highest level in 31 years. Financial markets suggest the Bank of England will raise interest rates four times before the end of next year, even after it kept borrowing costs on hold this week. According to The Guardian’s reporting on the rationale, the move is intended to limit the potential for short-term inflation to become entrenched. The Guardian reports that job losses will probably rise, compounding the challenges facing governments already swimming in debt. The Bank of England reported this summer that the spread between the riskiest and safest high-yielding debt has widened since the Iran war began, indicating investors are warier of holding risky debt.

A US recession has historically followed between three and three-and-a-half years after the first rate rise, on average, Deutsche Bank’s Jim Reid has calculated. Markets have a habit of falling before a recession begins, and often start to recover before the economy does.

The cyclically adjusted price-to-earnings ratio — or CAPE ratio, a popular measure of whether a market is overvalued — has risen to its highest level since 2000. For the S&P 500, the ratio stands at almost 41 points, more than double its long-term average of about 17 points, and approaching the record high of 44.19 points in December 1999, just before the dotcom crash.

The big concern among investors is that the main hope of economic redemption — AI — could be a dud. Research by Fathom Consulting shows that for the multitrillion-dollar AI boom to turn a profit, AI-related sales of the tech companies involved would need to rise by between $600 and $800bn within two years. Against a backdrop of investor patience increasingly tested, the consultancy gives a 30% chance that the AI bubble pops next year. Brian Davidson, an economist at the consultancy, said: “For all the impressive advances in AI technologies in recent years, the economics behind the current capex boom do not work. Yes, recent AI advances could yet unlock huge productivity gains; but sales of AI models need to increase by hundreds of billions of dollars per year over the next two years to justify the current spend. Such growth appears unlikely.”

The huge spending plans announced by AI companies are causing concerns that they may simply borrow too much. More than 1,000 investors registered for an analyst call conducted by Jefferies this week into “AI Extinction Warnings,” after the bosses of the world’s top tech companies called for a slowdown in “reckless” development. Albert Edwards, a senior analyst at Société Générale, wrote in a note to clients: “These are febrile times. The key worry for investors and policymakers alike is the extent to which the current oil price ‘shock’ will ripple through the global economy and whether it will necessitate sharply higher, recession-inducing, interest rates.” For many investors, a slowdown in feverish AI investment will guard against the danger of a bubble.

The situation has parallels with the dotcom crash of 2000, when many internet companies billed as the next big thing dramatically tumbled in value. That was a reminder that even if a technology is going to be revolutionary, investors can still lose their money if they finance too much infrastructure too early. Adrian Cox of the Deutsche Bank research team said: “It took a decade or more for demand to catch up with the infrastructure laid down in the British canal and railway and US telecoms and fibre booms – and many investors never recovered their capital.”

There are also sobering comparisons with the buildup to the crash of 1929. A hundred years ago, many small US investors were buying stocks “on margin” — acquiring them with borrowed money — and they were wiped out in the market turmoil that preceded the Great Depression. This year, South Korea’s army of traders have been buying shares in AI-linked chip makers on margin, doubling the value of the blue-chip Kospi index. But once the market started to fall, they were hit by a wave of margin calls — when investors are told to stump up more cash to keep borrowing. According to Goldman Sachs, 1.2 million South Korean retail investors were hit by margin calls — the equivalent of one in 30 adults.

One firm with ambitious AI plans is Oracle, the database software vendor. Oracle’s shares surged a year ago after it announced a cloud computing deal with OpenAI, the company behind ChatGPT. However, the shares have since halved as investors have fretted that Oracle could be borrowing too much to fund datacentres.

Despite the turbulence, some analysts argue the picture is less alarming than headlines suggest. Andy Haldane, the former Bank of England chief economist, told LBC this week: “Do I think there’s a significant dose of reality though in that productivity miracle in AI? Yes.” He warned the situation is fragile: “I don’t think outright collapse in a dotcom bubble type fashion. But could I see a slow release of air that doesn’t collapse the world economy, but slows it down? Yes, I could.” Analysts at Oxford Economics wrote in a note to clients: “A major downturn would need a trigger. Further geopolitical instability could be the catalyst, but we’ve long argued that the impact of geopolitical shocks on economic activity is overstated. Resurgent inflation and policy rate hikes by the Fed and other central banks are another candidate. However, our view on inflation is less alarmist — we think market expectations overstate the risk of further policy tightening.”

There are signs that AI is beginning to power economic growth. The US economy has recorded a rise in productivity growth, while AI is among reasons the UK economy has beaten expectations to grow at the fastest rate in the G7 in the first half of 2026. Such a turnaround in productivity could help companies with sky-high share prices begin to justify their valuations, though it is a high-stakes play at a time of intense global volatility.

The S&P 500 index of leading US companies sits 3% below an all-time high, with a combined value of more than $20tn for the “magnificent seven” tech stocks — Nvidia, Apple, Google, Microsoft, Meta, Amazon and Tesla.