Capacity Utilization Is a Ratio: The Denominator Is an Estimate Revised Every Year

Total industry capacity utilization was 76.3 percent in July 2026. The Federal Reserve's G.17 release of August 18, 2026 put that rate 3.1 percentage points below its 1972–2025 average of 79.4 percent, and that gap is the figure most often carried forward as a statement about industrial slack.

The gap is real and it is published. What it is not is a measurement of output. A utilization rate is a quotient, and the quantity underneath the line is an estimate that the Federal Reserve builds, updates three times a year, and periodically rewrites back to 1972. A reader who treats the rate as a production series will sooner or later attribute to factories something that happened to the estimate of what those factories could produce.

A Rate With Two Moving Parts

The G.17 documentation states the construction plainly: "The rate of capacity utilization equals the seasonally adjusted output index expressed as a percentage of the related capacity index." Both sides are index numbers on the same base. "The production index measures real output and is expressed as a percentage of real output in a base year, currently 2017," and "The capacity index, which is an estimate of sustainable potential output, is also expressed as a percentage of actual output in 2017."

That second sentence carries the weight. The numerator is a measurement of what was produced. The denominator is an estimate of what could have been produced. They are not the same kind of object, and they do not move for the same reasons.

The August 18, 2026 release shows the arithmetic working. Total industrial production rose 0.2 percent on the month while total industry capacity rose 0.1 percent, and the utilization rate went from 76.2 to 76.3 percent. Utilities output rose 0.5 percent against capacity growth of 0.2 percent, and the utilities rate went from 69.8 to 70.0 percent. Because all three figures are published rounded to one decimal, the identity reproduces to within a rounding step rather than exactly.

Figure 1 sets out the four cases. In two of the four cells the production index and the capacity index move in opposite directions, so they push the quotient the same way and the direction of the rate is determined. In the other two both move in the same direction, and the sign of the change depends on which side moved faster. In those two cells the rate on its own does not identify which side did the work. A capacity index that is unchanged is the boundary case, and there the rate moves with production alone.

The Denominator Has Its Own Growth Rate

Table 8 of the release prints capacity as percent changes rather than as a level, and the spread across rows is wide. The 2026 column reads 1.1 percent for total industry, 1.0 percent for manufacturing, 0.1 percent for mining, 2.8 percent for utilities, and 13.2 percent for selected high-technology industries. The column headed 1995–2026 reads 15.8 percent for that same high-technology group against 0.3 percent for manufacturing excluding it. 2026 is not a completed year, so those columns are not realized annual outcomes; they are what the August 18, 2026 release printed.

A denominator growing at 15.8 percent a year and a denominator growing at 0.3 percent a year do very different things to a quotient. For the same path of output, the first holds the rate down and the second lets the rate follow production almost step for step. The monthly column makes the same point at higher frequency: for July 2026 it reads 1.0 percent for selected high-technology industries and 0.0 percent for manufacturing excluding that group.

The high-technology rows are where the effect is visible in Table 7. July 2026 utilization for selected high-technology industries was 72.4 percent against a 1972–2025 average of 77.3, a gap of 4.9 points. Communications equipment read 59.1 percent against an average of 75.3, and 18.6 points below the 77.7 printed for that row in Table 7's 2009 Low column. Semiconductors and related electronic components read 75.4 against an average of 79.1, with a 1994–95 high of 93.0 and a 2009 low of 62.9.

What Tables 7 and 8 together do not supply is the production index for those individual industries, so they do not settle why communications equipment sits where it does. A denominator compounding at double digits will hold a rate down on its own. So will a numerator that has fallen. Both paths stay open on this evidence, and neither table closes either one.

Scale matters before the point is carried to the headline. Table 7's 2025 proportions put selected high-technology industries at 2.30 percent of total industry, against 75.90 percent for manufacturing as a whole, 13.03 percent for utilities and 11.07 percent for mining. Fast capacity growth inside a small slice is a large effect on that slice's own rate and a small one on the aggregate.

Reading a Gap Against a Long-Run Average

Figure 2 plots twelve of Table 7's rows, chosen to span the range rather than to reproduce the table, as the July 2026 rate minus that row's own 1972–2025 average. The range runs from plus 8.6 points for nonmetallic mineral products, at 82.6 against an average of 74.0, to minus 14.0 for utilities, at 70.0 against 84.0. That is a spread of 22.6 points across rows in a single month.

The headline gap of 3.1 points below average is an aggregate, and the detail does not line up behind it. Counting the nineteen NAICS-coded manufacturing industry rows in Table 7, seven were above their own 1972–2025 average in July 2026: nonmetallic mineral products (82.6 against 74.0), machinery (82.9 against 78.2), computer and electronic products (77.2 against 77.1), electrical equipment, appliances and components (86.8 against 81.7), aerospace and miscellaneous transportation equipment (74.5 against 73.4), miscellaneous (78.8 against 77.1), and petroleum and coal products (87.8 against 85.5). Mining was above as well, at 86.1 against 85.2. Twelve of the nineteen manufacturing rows were below.

Each row is measured against its own average, and those averages are far apart. Paper carries a 1972–2025 average of 86.5; nonmetallic mineral products carries 74.0. A reading of 78.0 is therefore 8.5 points of shortfall in the first row and would be 4.0 points of excess in the second. A row called below average is being measured against its own history, not placed on a common scale of tightness.

Utilities is the largest single gap on the chart and it also carries weight in the aggregate: 13.03 percent of total industry, sitting 14.0 points below its 1972–2025 average of 84.0. Table 8's utilities capacity figures read 4.4 percent for 2025 and 2.8 percent for 2026, the largest of the rows Table 8 prints outside the high-technology group. A capacity index rising at that pace lowers a utilization rate at any given level of generation. Whether generation is also soft is a separate question, and Table 7 does not answer it.

What the November 2025 Revision Did to Both Sides

Capacity estimates are not updated every month. The G.17 technical documentation on capacity places the updates at the February release and again at the annual revision, and a further mid-year update arrives with the July release. Figure 3 lays out that calendar. The same documentation describes capacity as "The greatest level of output an industry can maintain within the framework of a realistic work schedule, after factoring in normal downtime and assuming sufficient availability of inputs to operate the capital in place."

The most recent annual revision was released on November 24, 2025. Its documentation states that "The revision affected rates of change for IP from 1972 forward," and that it incorporated comprehensive annual production data from the 2022 Economic Census together with survey utilization rate data for 2024 from the Quarterly Survey of Plant Capacity Utilization. On the denominator: "From 2022 through 2025, capacity growth is revised down in each year, with an average revision of 0.4 percentage point." On the rate itself, capacity utilization "was 1.7 percentage points lower on average" across 2022 to 2024.

Those two statements cover different spans — capacity growth across 2022 through 2025, the utilization average across 2022 through 2024 — and the arithmetic belongs only where they overlap. Over 2022 to 2024, a downward revision to capacity growth taken by itself pushes the rate up, because a smaller denominator raises the quotient. The rate came out 1.7 points lower instead. On the overlapping years, then, the output side was revised down as well, and by more than enough to offset. How much more is not something these two published figures pin down, and nothing here identifies which industries drove it.

The delivery calendar belongs to the same picture. In late 2025 the G.17 schedule slipped twice: the October 17 release was announced as delayed, the November 18 release likewise, September data were issued on December 3, 2025, and October together with November data on December 23, 2025.

Concrete Framework

Six steps, in order, for turning a published utilization rate into a claim about output.

1. Name the two sides. Write the rate as production index over capacity index before saying anything about it. Table 7 carries the rate; Table 8 carries the capacity percent change for the same row and the same period.
2. Bracket the change. The published percent changes for production and for capacity should account for the change in the rate. Where they do not, the residual is rounding at one decimal or a revision, not a finding.
3. Use the row's own average. Table 7 prints a separate 1972–2025 average for every row, and among the rows in Figure 2 those averages run from 74.0 to 86.5. The same rate is a different verdict depending on which row it sits in.
4. Check the denominator's speed before calling a low rate slack. A row whose capacity change reads 13.2 percent and a row whose capacity change reads 0.6 percent arrive at the same rate from very different output paths.
5. Date the vintage. A rate is a rate as of a release. Record the release date beside the number — here, August 18, 2026 — because the annual revision reaches back through the history rather than touching only the newest months.
6. Keep both paths open. Where a rate is equally consistent with an output move and a capacity move, name both and wait for a second series rather than picking one.

The same discipline applies whenever a published figure is a ratio rather than a count. A stock of barrels covers 65 days or 106 depending on which rate sits in the denominator, a reserve currency's share can slide 1.47 points while only 0.12 points survive at constant exchange rates, and a housing unit can leave an annual rate and join a count in the same month without anything physical changing. Fix the denominator before reading the quotient.

Where This Doesn't Apply

None of this makes the utilization rate uninformative, and the boundaries are worth marking.

• Where capacity is nearly flat, the rate is close to an output signal. Table 8 reads 0.6 percent for manufacturing excluding high technology in the 2026 column and 0.0 percent in the July 2026 column. For those rows a month's change in the rate is mostly a month's change in production, and reading it that way is close to right.
• A published revision schedule is not a reliability problem. The revision dates are announced in advance, the method is documented, and the size of the last revision was quantified by the Federal Reserve itself. A number that moves on a published schedule is a different thing from a number that cannot be relied on.
• Cross-row comparison at a single date is not what these rates support. Petroleum and coal products at 87.8 and furniture at 67.3 are each measured against their own estimated capacity, built from their own source data. The difference between them is not a ranking of how hard each is being run.
• A long-run average is not a target. The utilities average of 84.0 covers 1972 through 2025, a span over which the generating fleet changed considerably. A 14.0-point gap against that average is not by itself a statement about the adequacy of today's fleet.
• The newest month is preliminary. Table 7 marks July 2026 as preliminary and February through June 2026 as revised. The figures here are what the August 18, 2026 release printed, and the recent months in it had already moved once.

What to Watch Next Week

The G.17 release calendar lists September 18, 2026 as the next release date. Each 2026 release has carried the prior month: June data appeared on July 17 and July data on August 18, so August is the month the September release should carry.

Three numbers from it rather than one:

• Total industry capacity utilization in Table 7, which read 76.3 for July.
• Total industry production percent change in Table 1, which read 0.2 for July.
• Total industry capacity percent change in Table 8's monthly column, which read 0.1 for July.

Three branches follow, each with an observable discriminator.

1. Rate up, capacity change at or above 0.0 percent. Production carried the move and on that month the rate tracks output.
2. Rate up, capacity change negative. Part of the move came from the denominator, and the output signal is smaller than the change in the rate suggests.
3. Rate down while production rises. Capacity growth outran output that month. In Table 8's 2026 column, utilities at 2.8 percent and selected high-technology industries at 13.2 percent are the rows where that arithmetic is most readily available.

Then the autumn. A May 15, 2026 G.17 notice announced an annual revision planned for release in the autumn of 2026. The 2025 edition arrived on November 24 and moved the 2022 to 2024 utilization average by 1.7 points. Any 76.3 written down this week is a figure against the August 18, 2026 vintage, and whether it survives the autumn is a separate question. The two delays of late 2025 are a reminder that the revision date itself can move.