US equities did not shrug off September's rate shock. They paid for it in three places the index level hides. Earnings from AI capital spending flow from the companies doing the building to the companies supplying them. That flow lets the market absorb a 111 basis point rise in the ten-year Treasury this year by spending its equity risk premium, cutting the share of profits returned to shareholders, and leaving most individual stocks to take the losses. The cap-weighted index looks resilient. The transmission of higher rates is real, and it is showing up in the payout ratio, in book debt, and in market breadth.
A Rate Shock in the Top Decile
The size of the move matters. The ten-year yield rose from 4.75% at the start of September 2026 to 5.29% at month end. That 54 basis point jump ranks in the top ten percent of the 770 monthly changes recorded since 1962. By October 7 the ten-year sat at 5.28%, the two-year at 4.77%, and the 2s10s spread at a positive 0.51%. A positive curve means the long end is carrying the repricing. That is the part of the curve that matters most for discounting long-duration equity cash flows.
Over the same month, US equities added $2.5 trillion in market capitalization. About $1.5 trillion of that came from technology alone. The textbook expects a long-duration, growth-heavy market to fall when long yields jump this much. Instead the most rate-sensitive sector led the gains.
The implied equity risk premium shows how that happened. At the start of 2026, with the ten-year at 4.18%, the expected return on US equities was 8.41%, a premium of roughly 4.23 points. By September 30 the expected return had risen to 8.99%. The ten-year had risen to 5.29%, so the premium over it fell to 3.70%, below 4% for the first time this year. Against the dollar risk-free rate of 5.07%, the premium stands at 3.92%.
Stocks did reprice, but only about half as much as bonds. The ten-year rose 111 basis points over the year while the equity expected return rose 58. The remaining 53 basis points came out of the risk premium. That is the first cost of resilience. Investors are accepting less compensation per unit of equity risk than they did in January, and the cushion against the next rate move is thinner.
How One Company's Capex Becomes Another's Earnings
The premium could compress without a collapse in prices because the earnings stream underneath it grew, and the reason for that growth is mechanical.
Aggregate capital expenditure at US public companies in the second quarter of 2026 was $133.4 billion higher than a year earlier, an increase of almost 36%. Technology, communication services and consumer discretionary each grew capex by more than 50%. The ten largest hyperscalers account for more than $2 trillion of AI capex.
A hyperscaler's capex is a cash outflow. It does not hit the income statement as a cost when spent. It enters the balance sheet as an asset and is amortized over time. For the supplier, the same dollars are revenue as soon as they are invoiced. Nvidia's chips, TSMC's fabrication, Broadcom and Micron's components, Constellation Energy's power, and Equinix and Digital Realty's data-center space all book that spending as sales, and their margins turn it into net income.
The builders also lengthened their amortization schedules. The depreciation charge that should offset the supplier's profit therefore arrives slowly. Across the market, the result is a timing gap: supplier earnings are recognized now, while the builder's matching expense is spread over years.
The aggregate numbers reflect this. Net income at US companies rose from $511 billion to $692 billion in the first quarter of 2026, a 35% gain. In the second quarter it rose from $576 billion to $904 billion, a 57% gain. Technology, financials and communication services led. Health care, utilities and real estate lagged.
Higher rates raise the discount rate on future cash flows. An earnings stream growing 35% to 57% year over year can offset that. Equities held up because the numerator moved faster than the denominator.
Where the Cash Went
The accounting gap shows up in cash returned to shareholders.
Free cash flow is measured after capex, so it cannot keep pace with earnings when the earnings themselves come from capex. Dividends over the last twelve months are up about $43.3 billion, roughly 5%, from calendar 2025. Buybacks are up about $106.7 billion, roughly 9%. Stock issuance spiked in the second quarter, so net buybacks grew only about 7%.
Those increases are reasonable. Set against earnings growth of 35% to 57%, they are small. S&P 500 companies returned 63% of earnings to shareholders over the trailing twelve months. That is the lowest ratio since 2004. For most of the past two decades the figure ran at 80% or higher, and at times above 100%. The divergence began around 2024 and has widened since.
This is the cross-asset channel the rates headline misses. In the past, steady buybacks provided a large source of demand for US equities that did not depend on price, and that demand often dampened the effect of rising yields. That source has shrunk relative to the earnings base. The market is now relying on earnings to absorb rate moves, with less support from corporate repurchases. A market held up by earnings momentum is more exposed to an earnings disappointment than one held up by cash returns.
The balance sheet shows the same pattern. Between the second quarter of 2025 and the second quarter of 2026, book equity across US companies rose almost 13%, or $1.8 trillion. Total book debt rose almost 8%, or $1.9 trillion. In technology, book equity rose almost 30% and debt about 18.8%. Some of the AI build is financed with borrowed money at long-end rates now near 5.3%.
One link is unverified and should be treated as an inference. Corporate borrowing for AI infrastructure may be one of the forces pushing long yields higher, since $1.9 trillion of new book debt has to be absorbed somewhere in fixed income. The available evidence does not show how much of September's move came from supply pressure versus inflation or policy expectations. CPI stood at 334.1 as of August, which does not settle the question. If the supply channel is real, the AI boom is helping to raise the rates it is currently able to withstand.
The Median Stock Took the Hit
The rate shock that the index avoided landed on most individual stocks.
Three sectors, technology, energy and materials, have carried the market's gains this year. Only about half of listed companies are up year to date. In the third quarter, about 65% of all listed stocks fell. More than half of all companies reported declines in net income, even as aggregate earnings rose 57% in the second quarter.
These results fit together once the mechanism is clear. A utility, a real estate firm or a mid-cap industrial outside the AI supply chain faces higher discount rates and higher refinancing costs without the earnings boost that offsets them. For those companies, a 54 basis point move in the ten-year works as theory predicts. In aggregate, the market's resilience is the resilience of a concentrated group of AI suppliers and builders whose earnings are large enough to dominate a cap-weighted total.
For anyone holding the index passively, this is an AI bet whether they intended one or not. An S&P 500 fund offers partial exposure, and a total-market fund offers more. The diversification an index is meant to provide has been weakened by concentration in a single capex cycle.
The Case Against the Read
The strongest objection is that none of this is fragile. The earnings are reported net income, not projections. A 3.70% premium over the ten-year is lower than in January but not extreme. Expected equity returns did rise this year, which suggests the market adjusted to rates instead of ignoring them.
The objection also points out that the public figures understate the boom. OpenAI and Anthropic, both expected to command trillion-dollar-plus valuations, are private, so their revenues and invested capital are missing from the aggregates. Joint ventures and off-balance-sheet vehicles carry some of the debt. If the AI business works, builders' cash flows turn positive, lenders are repaid, and the low payout ratio will look like disciplined reinvestment in hindsight.
That case is coherent. It rests on one condition: returns on AI capex must exceed the cost of capital for the builders, and that cost of capital rose this year. The supplier earnings driving the aggregate are borrowed from the builders' future amortization. They are only real profits if the builders eventually earn enough to cover the depreciation. If AI succeeds only moderately, with some earnings but not enough to cover a cost of capital now set against a 5.3% ten-year, the write-downs will come later.
What Would Change the View
The market will likely reprice AI-centered businesses before accountants record any write-down. Waiting for an impairment charge before selling would mean acting too late. The useful indicators therefore come from markets and cash flows, not from accounting.
Three conditions would confirm or break the read. The first is third-quarter capex. Supplier earnings depend on the builders continuing to increase spending, and any flattening at the hyperscalers would reduce supplier revenue directly. The second is the payout ratio. If it stays near 63% or falls further while earnings growth slows, the market will have neither its earnings support nor its buyback support. The third is the equity premium against the ten-year. Another September-sized move, with earnings growth no longer accelerating, would push the premium toward 3%. At that level the market has very little room left to absorb further rate increases.
The thesis breaks in the bullish direction if free cash flow at the major builders starts to catch up with earnings before the next rate shock. That would show the timing gap closing through actual returns on investment. Until then, the index's resilience to rates depends on AI capex continuing, and the costs are falling on the premium, the payout ratio, balance-sheet debt and the median stock.
