Strong U.S. business activity got stronger on Sept. 23, and that was enough to jolt the market’s cost-of-capital assumptions. S&P Global’s flash September PMI release showed the composite output index rising to 58.4 from 56.0 in August, the strongest expansion since July 2021. Treasury yields climbed to 2007-era highs, with Reuters reporting the 10-year at 5.054%, and the Nasdaq fell 1.05% on the day.
For technology companies and their customers, the message was not that demand is disappearing. It was that growth may now have to be financed at a much higher price.
The useful question is whether this marks a real financing reset for the AI buildout or just a noisy day in which strong growth and inflation worries temporarily overpowered hopes for easier money. The answer, for now, is narrower and more practical: Sept. 23 did not prove a permanent new regime, but it did show which projects become fragile when strong demand, higher input costs and 5%-plus government yields collide.
Stronger growth, tighter capacity
The PMI surprise was strong enough to matter on its own. Because a PMI is a diffusion index, a reading above 50 signals expansion and a higher number indicates faster reported growth. S&P Global said the September flash survey pointed to roughly 4.0% annualized growth in the third quarter and a 5.0% pace for September. Services led the acceleration, manufacturing also improved, and payroll growth was the fastest in more than four years.
That is the growth side of the story. The more uncomfortable side is what accompanied it: severe capacity and supply-chain bottlenecks, a sharp rise in backlogs, delivery delays that were the most widespread since July 2022 outside the pandemic period, and input-cost growth near a four-year high. In other words, businesses were not merely selling more; they were running hotter.
That combination matters because it cuts two ways. Stronger demand can lift revenue, utilization and near-term earnings. But if companies are struggling to source inputs, add capacity and fulfill orders, the same growth burst can keep inflation pressure alive. Markets do not need proof of permanently reaccelerating inflation to react; they only need a stronger reason to think the Federal Reserve may keep policy tight for longer.
Reuters said fed-funds futures lifted the implied probability of another October rate increase to 73%, up from 53% earlier in the day. That was a market judgment, not a policy decision. Still, it helps explain why rate-sensitive shares sold off more sharply than the broader market: the Dow slipped 0.18%, the S&P 500 fell 0.53%, and the Nasdaq dropped 1.05%.
Why 5% Treasuries change the math
The technology angle is not just that higher rates are “bad for tech.” It is that many technology investments are long-duration assets: they require heavy spending upfront, while the payoff arrives over years through subscriptions, compute demand, cost savings or future market share.
When the benchmark risk-free rate moves up, two things happen at once. The discount rate used to value future cash flows rises, which reduces the present value of long-dated returns. And the cost of funding those projects rises too, whether the money comes from debt, equity or a mix of both. At the same time, government bonds yielding around 5% become a more credible alternative to expensive growth assets.
That is why the same macro data can be good news for revenue and bad news for valuation. A profitable software platform with sticky customers may benefit from a stronger economy. A speculative AI training cluster or a marginal data-center expansion may see its economics deteriorate even if demand still looks healthy.
It is also why the market move should not be reduced to a single cause. The PMI surprise mattered, but so did other forces Reuters cited, including Brent oil near $101.62, along with Federal Reserve commentary, Treasury supply and auction dynamics, fiscal expectations and the term premium investors demand for holding long-term debt. The official Treasury daily yield table later listed the 10-year constant-maturity rate at 5.11%, the 20-year at 5.45% and the 30-year at 5.40%. Those figures are interpolated from market bids and will not exactly match every intraday quote, but they confirm the broader point: long-term U.S. rates were elevated enough to change boardroom assumptions.
A financing scorecard for AI and cloud projects
For executives, investors and enterprise buyers, Sept. 23 is best read as a project-selection test. The key question is not whether “tech” wins or loses. It is which specific investments still clear the hurdle rate.
A practical scorecard looks like this:
- Cash-flow timing: How soon does the project begin generating revenue or savings?
- Funding mix: Is it being financed with internal cash flow, debt, new equity or customer prepayments?
- Rate sensitivity: Does the return still work if the hurdle rate rises another 100 basis points?
- Power and equipment commitments: Are electricity, networking gear and hardware already contracted, or still exposed to cost pressure and delays?
- Utilization assumptions: Does the model require very high occupancy or compute demand to pay back?
- Flexibility: Can the project be phased, delayed or canceled without stranding capital?
That framework quickly separates the likely winners from the vulnerable. Cash-rich incumbents with recurring revenue can often keep building through higher rates, especially if they are selling into the AI buildout rather than funding it from scratch. A cloud provider, profitable software company or infrastructure vendor with strong demand may absorb a higher cost of capital and even gain relative advantage if weaker competitors pull back.
The pressure is greater on the businesses that must finance the buildout themselves. Startups heading into a new funding round, highly leveraged operators, and data-center projects that depend on optimistic utilization assumptions look more exposed. So do enterprise customers considering discretionary software migrations or ambitious AI deployments if their own borrowing costs have risen and their finance teams are recalculating payback periods.
That does not mean a 5% 10-year yield automatically stops technology investment, let alone ends the AI cycle. Periods of higher rates have not historically prevented companies from funding strategically important technology. They have changed who can fund it, which projects get approved first, and how much slack investors will tolerate in the plan.
What Sept. 23 showed, more clearly than many generic market selloffs do, is that the AI-and-cloud expansion has entered a more selective phase. Demand is still there. Hiring is still growing. New orders are still coming in. But if backlogs, supply constraints and input costs keep yields elevated, the buildout will favor shorter paybacks, stronger balance sheets, contracted demand and projects that can be staged. If those pressures ease, the financing squeeze could prove less durable than the one-day market reaction suggested.
For now, the safest takeaway is not that strong growth is bad for technology. It is that strong growth no longer guarantees cheap money for building it.
By
By
By
By
By
