ISM Hit 55.6 in July. Here’s What That Means for Your Cost Model.

The ISM Manufacturing PMI printed 55.6 for July, the highest reading since May 2022. If you run a plant, that is good news. If you buy for one, it is a scheduling warning with a price tag attached, and the headline number is the least useful thing in the report.

The useful part is underneath. Production came in at 58.5, the strongest since November 2021. Employment reached 52.8, expansion for the first time in thirty-three months. New Orders held at 56.7 and Backlog jumped 4.5 points to 55.0. Read together, those four say demand is real and factories are responding to it.

Then look at Supplier Deliveries: 58.9. On this index, above 50 means deliveries are getting slower. Your suppliers are not keeping up, and they were already not keeping up in June.

Expansion without a capacity constraint

Here is the part that gets misread. Federal Reserve data for July puts manufacturing capacity utilisation at 76.0%, which is 2.2 points below its 1972–2025 average of 78.2%. Total industry sits at 76.3%, more than three points below its own long-run average.

So American manufacturing in aggregate is not running hot. There is slack. Which means a 55.6 PMI does not, on its own, justify a broad supplier price increase, and you should not accept one framed that way.

The constraint is real but narrow. High-technology production (semiconductors, computers, communications equipment) ran an index of 199.6 in July, up 11.8% year over year. That is where the queue is. Everywhere else, a supplier claiming capacity pressure is claiming something the national data does not support for their sector, and it is fair to ask them to show it at their own plant.

The three lines that actually moved

Producer prices tell a more precise story than the PMI. Headline final demand PPI was flat month over month in July and up 4.7% over twelve months, which is unremarkable. But processed goods for intermediate demand, which is the line manufacturers actually buy, rose 9.9% over the same twelve months. Roughly double the headline. That gap is your margin.

Three components carried most of it.

Memory: decelerating, not cooling

TrendForce forecasts server DRAM contract prices rising 13–18% quarter over quarter in the third quarter, with RDIMM bit supply growing only 15–20% year over year against faster server CPU shipment growth.

Be careful how you read that, because a lot of people are reading it wrong. Thirteen to eighteen per cent is a deceleration. Conventional DRAM rose roughly 90–95% quarter over quarter in the first quarter of this year; PC DRAM more than doubled. Against that base, the third quarter is the market calming down. It is still a double-digit quarterly increase on a component that has already roughly doubled inside a year.

The detail that matters more than the percentage: US cloud providers hold multi-year agreements that cap what suppliers can charge them. From the third quarter onward, the increase lands disproportionately on buyers without long-term agreements. If you are a mid-size manufacturer buying memory on the spot market, you are not paying the market price. You are paying the market price plus everything the hyperscalers negotiated away.

Steel: the line most models fixed in January

Steel mill product prices rose in every single month of 2026 through July, up 18.7% year to date and 22.5% year over year. Unlike memory, this one is still accelerating.

It compounds with a change most cost models have not absorbed. Since 6 April, Section 232 duties on steel, aluminium and copper apply to the full customs value of an imported article, derivatives included. Previously that value was split between metal and non-metal content. If your landed-cost model still carries the split, it is understating duty on every affected part, and it has been for five months.

Freight: rates up, volume down

The Cass Truckload Linehaul Index rose 8.6% year over year in July while shipment volumes fell 4.8%. Rates rising into falling demand is not a demand story. It is carriers exiting faster than freight is disappearing.

That asymmetry is the condition under which contract renewals go badly, because your carrier knows the replacement capacity is not there. Ocean is worse: the Drewry composite roughly doubled between late May and late August, with Shanghai–Los Angeles at $6,818 per FEU by 27 August.

What I would do this quarter

1. Re-baseline the model on July, not January. An annual cost model fixed at the start of 2026 is understating steel by roughly nineteen per cent and electronics by nearly thirty. Every quote you have validated against it since spring has been validated against the wrong number.

2. Separate the two claims your suppliers are making. Input cost inflation is documented and largely legitimate. Capacity pressure mostly is not, outside electronics. Concede the first, contest the second, and make them show plant-level utilisation rather than a national headline.

3. Get an LTA on memory or accept the marginal-buyer premium. There is no third option. If the volume does not justify one alone, aggregate across business units before the fourth quarter closes.

4. Recheck the Section 232 basis on every affected part. This is a mechanical fix with a real number behind it, and it is the single most likely place your landed cost is wrong today.

5. Read Supplier Deliveries before New Orders. Backlog at 55.0 and deliveries at 58.9 mean lead times lengthen from here. Whatever you were going to buy in the fourth quarter, the order date moved earlier than your planning calendar thinks it did.

The wider read

One number sits awkwardly against all of this and is worth carrying into your capital conversations: private manufacturing construction spending is down 21.7% year over year. The factory-building boom of 2022–2024 is unwinding sharply even while the PMI expands.

That combination of strong current demand, tightening lead times, slack aggregate capacity, and collapsing investment in new capacity is not a boom. It is a market working its existing assets harder without adding new ones. For a buyer, that is the condition in which lead times lengthen structurally rather than cyclically, and it does not resolve on its own.

The 55.6 was the good news. The report around it was the brief.