In October the AI risk moved from interest rates to the financing of the builders

The third quarter ended with a clear ranking of risks. On 30 September 2026 the ten-year US yield stood at 5.29%, close to its high for the year, Brent had gained 42% over the quarter, and the risk premium on high-yield bonds had jumped from 2.66% to 3.12% in the last week of September. The AI Macro Report Q3, published on 31 October, summed this up in one sentence: the AI boom had become a matter of rates and credit. In October the balance shifted. The yield fell to 4.98% by 29 October, while Brent rose by a further 13.6%. The S&P 500 ended the month almost unchanged, the Nasdaq 100 lost 6.7%, and the shares of data centre operators, the SOXX semiconductor index fund, Alphabet and Meta fell by between 15% and 41%.

The obvious reading of the month is that pressure from interest rates has eased and things are calming down. The AI Macro Cockpit uses eight signals to track how the AI build-out moves rates, credit and equity markets. Its data support that reading only in part. Yields fell worldwide, but the inflation expectation implied by inflation-protected US Treasuries (TIPS) stayed at 2.35%, practically where it stood at the end of September (2.36%). Only the real rate fell, and that is the pattern of a market that buys government bonds as a safe haven because geopolitics and technology stocks are unsettling it at the same time. The risk has therefore moved from the price of money to the question of how the builders pay for their investment. Meta announced a capital increase, Alphabet surprised the market by stepping up its investment, and in the credit market the bonds of smaller infrastructure providers reached yields that are otherwise reserved for weak borrowers. This is not a diagnosis of crisis: according to the providers, demand for computing power still exceeds supply. The finding concerns financing, and financing is where it becomes clear who will need the capital markets for the next stage of the build-out.

This Monthly first looks back at the position at the end of September and then follows October in six steps: the monthly scorecard, rates and oil, hyperscalers, credit, IT services, and banks and the car industry. It closes with the status of the four scenarios from the quarterly report, the dates to watch in November and Palantir’s quarterly results.

At the end of September the AI boom had reached the limits of rates and credit

The quarterly report set out three findings against which October can be measured. First, the Federal Reserve (Fed), the European Central Bank (ECB) and the Bank of Japan (BoJ) had tightened in response to the oil price, and ten-year yields in the US, Germany and Japan had risen by 37 to 85 basis points over the quarter. Second, the equity market had narrowed: the ratio of the equal-weighted to the cap-weighted S&P 500 (RSP ÷ SPY) fell by 4.3% and reached its low on the last trading day, and semiconductors lost almost 12% against the Nasdaq 100. Third, the capex of the ten large AI builders tracked by the cockpit (Microsoft, Alphabet, Amazon, Meta, Oracle, CoreWeave, TSMC, Micron, Constellation Energy and IREN) absorbed 79% of their operating cash flow, and the companies relied increasingly on bonds and loans.

From these findings the report derived four scenarios. The base case assumed that demand would hold despite expensive financing. It named two risks, a credit event at an AI infrastructure borrower and a second oil wave that forces central banks into further rate rises. The upside lay in diplomacy that lowers oil prices and interest rates at the same time. October did not settle any of these scenarios, but it did show the channel through which they operate.

October turned the sector rankings upside down

The monthly scorecard shows a pattern that cuts across sectors. The losers were the companies that build, operate or equip data centres. CoreWeave fell by 37%, the data centre operator IREN by 41%, Alphabet by 19%, the US semiconductor exchange-traded fund (ETF) SOXX by 17% and Meta by 15%. The winners, remarkably, were the IT services firms, a sector written off as an AI loser more often than almost any other in 2026. At the end of June Accenture traded on a price-to-earnings ratio of 10.0, its lowest valuation since its 2016 financial year. Accenture rose by 41%, and Wipro, EPAM and Capgemini gained between 22% and 31%. Microsoft, the hyperscaler with the lowest capex-to-cash-flow ratio in the cockpit, ended the month slightly higher.

Monthly scorecard October 2026, change from 30 September to 30 October 2026 in per cent

The first cockpit signal, index concentration, more than made up its September loss in October. The ratio RSP ÷ SPY rose by 5.9% and reached its high for the month on the last trading day, and smaller companies (IWM) even gained 7.7% against the Nasdaq 100 (QQQ). A single month’s rotation does not make a trend, and part of the move is a rebound after a quarter in which the mega-caps did all the running. But the timing coincides with the results season. The two strongest days for breadth were 22 October, when Meta lost 7% after its results, and 29 October, the day Alphabet’s shares slumped. On both days RSP ÷ SPY gained more than 1.5%.

Market breadth, indexed 30 September 2026 = 100

In October the same oil shock triggered a flight from risk

In the third quarter the oil price and interest rates moved closely together. The correlation of the daily changes in Brent and the ten-year US yield was plus 0.66 in September, so a higher oil price almost reliably meant a higher yield. In October the relationship turned to minus 0.28. Brent rose in bursts, driven by attacks on tankers in the Strait of Hormuz on 8, 16 and 30 October and by reports that the US and Israel might resume strikes on Iran after the midterm elections on 3 November. In the January contract, the benchmark from November, the rise was 19.3%. Yields fell nonetheless, by 31 basis points in the US, 22 in Germany and 24 in Japan.

Ten-year yields in per cent and Brent in USD per barrel

This is a contradiction only at first sight. In the summer the bond market read every attack in the Gulf as an inflation risk. In October it also read it as a risk to growth and share prices. On a rates desk you learn this distinction early, and it can be read off a single number. Breakeven inflation, the difference between the nominal and the inflation-protected ten-year US yield, stood at 2.36% on 30 September and at 2.35% on 29 October. The entire fall in the yield was a fall in the real rate. The market expects unchanged inflation and accepts a lower real return on the safe asset. The two-year yield also fell, by 27 basis points. Over the next two years the market therefore expects a flatter rate path, even though it still expects a rise in December.

This became clearest on 22 October. The ten-year yield fell by 17 basis points to 4.96%, according to CNBC the largest daily decline since 10 March 2023, on a day with a new oil high and a sell-off in AI stocks. CNBC named technology stocks, the Middle East and the oil price as drivers and quoted Mitch Goldberg of ClientFirst Strategy: “People are all of a sudden terrified of the amount of bond issuance by AI-related companies.” As early as 15 October investors had sold technology stocks after a higher producer price reading and, according to CNBC, switched into bonds. The cockpit’s correlation module shows the change. The correlation between the Nasdaq 100 and the ten-year yield was minus 0.72 in September: rising yields pushed share prices down. In October it was plus 0.65: share prices and yields fell together.

Day-to-day relationship, correlation of daily changes by month

The central banks were not the trigger. On 28 October the Fed kept the target range for the federal funds rate at 3.75% to 4.00%, Lisa D. Cook and Beth M. Hammack voted for a further rise of 25 basis points, and the statement noted: “Inflation remains elevated.” According to the Bureau of Economic Analysis the US economy grew at an annualised 3.4% in the third quarter, and inflation was 3.7% in September. On 29 October the ECB left its deposit rate at 2.5%. Admittedly, the weak jobs report of 2 October, with only 29,000 new jobs, gave a further reason for falling yields. But yields still rose afterwards, to the high for the year of 5.31% on 5 October. According to market reports, a rise in December was largely priced in by the end of October. The inflation risk has therefore not receded. It is being overshadowed by a different concern.

The market punishes hyperscalers that need fresh capital

The quarterly results of the large cloud providers (hyperscalers) confirmed demand. Google Cloud grew by 103% to USD 30.7 billion, Microsoft’s Azure by 48% and Amazon Web Services by 39%, its fastest growth in almost five years. Microsoft’s chief financial officer Amy Hood said: “We still have demand that exceeds supply.” The market takes demand as given and judges the build-out by how it is financed.

AI infrastructure, indexed 30 September 2026 = 100

The reactions differed by company, and negative free cash flow alone does not explain them. Amazon reported negative free cash flow of USD 42.4 billion for the last twelve months, the weakest in the group, and lost only 3.5% in October. The market punished companies that signalled additional capital needs. On 21 October Meta reported negative free cash flow of USD 2.12 billion, investing more than its operations generated, and narrowed its 2026 capex range to USD 135 billion to 145 billion by raising the bottom end. On 26 October the company announced that it would raise around USD 35 billion through new common shares and mandatory convertible preferred shares, and held out the prospect of a further programme of around USD 60 billion, which will only become possible once a new class of non-voting shares has been introduced. On 30 October, according to Reuters, the rating agency S&P Global Ratings lowered the outlook on Meta’s rating to negative. Alphabet invested USD 51.4 billion in the third quarter, reported negative free cash flow of USD 9.1 billion and announced significantly higher capex for 2027. The shares fell by 12.6% on 29 October, according to market reports wiping around USD 480 billion off its market value, the largest one-day loss of market value in the company’s history. Microsoft invested less than expected and its shares were little changed the day after its results.

The cockpit’s AI financing module shows the scale. The ten AI builders it tracks invested at least USD 255 billion in the third quarter, and only seven of them have reported so far. Their capex-to-cash-flow ratio, capex over the last four quarters as a percentage of operating cash flow, rose from 79% to 85%. Granted, 15% of operating cash flow is still left over, and the hyperscalers are highly profitable groups with investment-grade ratings. However, Amazon, Alphabet, Meta and Oracle each invested more in the third quarter alone than their operating business generated, and on a four-quarter basis this is already true of Amazon (128%) and Oracle (161%). The gap is closed by bonds, leases and, since October, equity as well. When a group with Meta’s earning power issues shares rather than buying them back, it shows how far capex needs have outgrown cash flow.

Capex and capex-to-cash-flow ratio of ten AI builders

The credit market draws the line at smaller borrowers

The second cockpit signal tracks the price of credit. The risk premium on US high-yield bonds (HY spread) rose from 3.12% at the end of September to 3.40% on 22 October and stood at 3.26% on 29 October. For the market as a whole this is a moderate move. According to market reports it was driven by the weakest credits, whose spreads rose above ten percentage points for the first time since 2022. There is no clear explanation for the narrowing after 22 October.

US high-yield spread (ICE OAS), per cent

At the centre was CoreWeave, a provider of computing capacity that funds its build-out mainly with debt. On 14 October the shares lost around 15.5% after the yield on its bond maturing in 2032 had risen to around 15%, dragging Oracle and Nebius down with them. According to CNBC, the credit default swap (CDS) on CoreWeave was quoted at 1,145 basis points on 21 October. Chief executive Mike Intrator countered: “We are confident in our backlog which more than supports all of our debt obligations.” The backlog is real. But according to CNBC Microsoft accounted for more than two thirds of the previous year’s revenue, and around USD 20.7 billion of its debt falls due by 2030. Even Oracle, with an investment-grade rating, traded “more like a high-yield credit” in the view of the research firm CreditSights.

In its Global Financial Stability Report of 13 October, the International Monetary Fund (IMF) described the channel through which the issue reaches banks: lending by US banks to non-bank financial institutions rose by 18% in 2026, and exposures to private equity and private credit firms by 30%. The IMF does not classify this as a systemic risk in itself, but as a possible source of losses under severe stress. In Bank of America’s fund manager survey for October, according to summaries of the results, 39% of participants named hyperscaler capex as the most likely source of a systemic credit event. No credit event has occurred. But the price at which the market provides new AI capital has risen, and it rises first where equity is thinnest.

The supposed AI losers rallied despite falling fees

Accenture opened the month with a share price jump of 15.8%. On 1 October the company reported 7% growth in local currency for its fourth fiscal quarter and an outlook of 3% to 6% for the new fiscal year. By 6 October, however, the shares had given back two thirds of the gain. The larger part of the 41% monthly gain came afterwards, without new company news, and the largest single daily gain, 7.7%, came on 22 October, the day of Meta’s fall and of the sharpest drop in yields.

IT services, indexed 30 September 2026 = 100

The sector moved in lockstep, even on bad numbers. On 23 October Infosys cut its growth guidance to 1% to 2% and, according to media reports, pointed to pricing pressure, and on 29 October Cognizant lowered the top end of its range. Both stocks ended the month with double-digit gains. Accenture’s chief executive Julie Sweet had described the situation herself on 1 October: “In Q4, we saw lower pricing in many areas of our business.” This is the behaviour of a basket that investors buy as a counterweight to the AI builders, even as the companies themselves report weaker pricing. In October the consultancies reported weaker pricing while AI capital became more expensive, and that was enough to trigger a re-rating whose durability only the next set of results will show.

In October banks and the car industry followed their own drivers

In mid-October the large US banks reported strong revenue from equity trading. JPMorgan increased it by 49%, Goldman Sachs by 72% and Morgan Stanley by 41%, while trading in rates, currencies and commodities was weaker. The executives were reticent about AI financing. JPMorgan called its own exposure “still quite modest”, and Goldman chief executive David Solomon, asked when the investment cycle would reach its limit, replied: “I don’t know and nobody knows.” The US financials ETF (XLF) gained 5.8% in October.

Banks, US financials and automobiles, indexed 30 September 2026 = 100

European banks moved the other way and lost 10.9%, UniCredit by 11.9%, even though Deutsche Bank reported its third record profit in a row on 28 October. My reading is that the explanation lies in the revenue mix: European banks earn the larger part of their profits from net interest income, while the large US houses have a second engine in trading and investment banking, and in October it was firing. When ten-year Bund yields fall by 22 basis points, as they did in October, the market first marks down future interest income, regardless of how good the past quarter was. That is how Deutsche Bank’s record profit and the sector’s share price loss fit together.

European carmakers’ shares were almost unchanged over the month. The pressure showed up in company results instead. On 27 October Mercedes-Benz cut its forecast for the adjusted return on sales of its car division in 2026 to 2% to 3%, citing China and higher costs in the fourth quarter. On 29 October Volkswagen reported an operating loss of EUR 5.9 billion for the quarter, mostly due to write-downs on its stake in Porsche. According to estimates by the industry association CPCA, Chinese passenger car retail sales in October were around 29% below the previous year. In the car industry AI so far shows up in partnerships, for example between Mercedes-Benz and NVIDIA for automated driving, but not yet in the numbers.

At the end of October only oil stood at its threshold, and only on expiry day

On 31 October the quarterly report named signals for each scenario that show which path is prevailing, and derived their thresholds from the extreme values of October. The table therefore shows where the month closed within these ranges, and the thresholds only serve as a yardstick from November onwards. The table records the position as at 30 October 2026.

The trends are the author’s own assessment and not investment advice.

The oil threshold was crossed on the last trading day, but in the December contract on its expiry date. According to market reports the January contract, the front month from November, closed at USD 113.43. The scenario will only be confirmed once that contract also rises above the threshold and yields follow. In October yields did not follow. As long as the same conflict is read both as an inflation risk and as a growth risk, a higher oil price no longer feeds through reliably to interest rates.

Several dates in November could help decide between the scenarios. The US midterm elections take place on 3 November, and the oil market links them to the fear of new strikes on Iran. Later in the month CoreWeave and NVIDIA report, and the ECB usually publishes its Financial Stability Review in November. According to market reports, a further rate rise is largely priced in for the Fed meeting in December.

Terms and abbreviations