The numbers

MeasureAugust, monthlyForecast12 months
Headline CPI, all items+0.4%+0.4%+3.4%
Core CPI, ex food and energy+0.3%+0.2%+2.4%
Energy+2.1%+16.3%
Gasoline+3.9%
Food+0.1%+2.7%
Shelter+0.3%+3.0%

Two rows carry the whole report. The headline matched to the decimal. The core row is one tenth hot on the month while the annual figure in the same row went down, from 2.5% in July to 2.4%. Everything that follows is about why the first fact moved markets and the second did not.

Headline versus core

Headline CPI is the whole basket, everything an urban household buys, weighted by how much of the budget it takes.

Core CPI is the same basket with food and energy removed. Not because they do not matter, they obviously do, but because they swing with things a central bank cannot touch: harvests, weather, and in August 2026, a war. Energy is up 16.3% over the year, and gasoline rose 3.9% in a single month as the conflict with Iran intensified. If the Fed raised rates every time oil spiked it would be chasing a number it has no lever on.

So core is the piece of inflation that policy can actually influence, which makes it the piece the Fed acts on. That is the entire reason a report with a perfectly matched headline still moved twenty points of hike probability. The tenth that missed was in the number that counts.

Why annual core fell while the odds rose

This is the part that confuses people every month, so it is worth slowing down.

The 12 month rate is a rolling window. When August 2026 enters it, August 2025 leaves it. If the month leaving was hotter than the month arriving, the annual rate falls even when the new month is bad. That is what happened: a hot August 2025 dropped out, so 2.5% became 2.4% despite a 0.3% monthly print.

The Fed knows this, which is why it weights the recent monthly pace far more than the annual average. And the monthly pace is the problem: a 0.3% monthly core rate, if it simply continued, compounds to roughly 3.6% a year, well above the 2% goal. With three officials already on record for a hike and a meeting five days away, one monthly overshoot was enough.

Put simply: the annual number tells you where you have been, the monthly number tells you where you are going, and the Fed steers by the second.

The shape of 2026 inflation

One month is noise. Seven months is a shape.

MonthHeadlineCoreEnergy
February+0.3%+0.2%+0.6%
March+0.9%+0.4%+10.9%
April+0.6%+0.4%+3.8%
May+0.5%+0.2%+3.9%
June0.4% lower+0.0%5.7% lower
July+0.1%+0.2%1.5% lower
August+0.4%+0.3%+2.1%

Read the headline column against the energy column and the pattern is obvious: the headline goes wherever energy goes. March's 0.9% was a 10.9% energy spike. June's negative print was energy falling 5.7%. The headline has been an oil chart with a lag.

Now read the core column on its own. It has run between 0.0% and 0.4% all year, with no trend up and no trend down, and August's 0.3% sits inside that band. That is the honest version of the story: core is not accelerating, it is refusing to decelerate, and at a moment when three voters already want a hike, refusing to decelerate is enough.

What the market did with it

Fed funds futures moved the priced chance of a hike on September 16 from under 70% before the release to roughly 90% within minutes, and by the close were pricing two hikes before the end of the year. Treasury yields rose across the curve.

For context on how fast that number has been moving, here is the same probability across our coverage of the last three weeks.

MomentPriced chance of a September hike
Before Jackson Holeabout 35%
After Warsh's remarks66%
After Waller leaned toward holdingabout 50%
After August payrolls58%
After Thursday's PPI63% to 70%
After Friday's CPIabout 90%

A hike on Wednesday would be the Fed's first since July 26, 2023, roughly three years and two months, and the first within a year of a cut since 1999. We explained the meeting itself, the vote, the dot plot and what a quarter point does to your accounts in Friday's article, and none of that has changed. What changed is that the question moved from whether to how many.

What Bitcoin did in the release hour, as promised

On Thursday we published a test in advance. The last nine CPI release hours had averaged 1.9 times a normal hour's volume with a buy share near 52%. The PPI hour the day before had traded 4.8 times normal at 29.8%. We said the CPI hour would show which of those was the better guide.

It was both at once, which is not the answer we offered.

MeasureFriday's CPI hourThe nine CPI hoursThursday's PPI hour
Volume vs an average hour4.3x1.9x, max 3.6x4.8x
Share of the day's volume17.8%7.8%20.1%
Taker buy share51.5%52.4%29.8%
Hour move+1.29%+0.57%1.21% lower

On volume it looked like Thursday: $272 million, 4.3 times a normal hour, more than any of the nine CPI hours we measured. That is the second consecutive inflation print to break the calm pattern, and we have to say plainly what that does to the finding. The nine release average of 1.9 times now reads as a description of a quieter stretch rather than a rule, and we presented it with more confidence than two counterexamples in two days can support.

On the tape it looked like the nine: a buy share of 51.5%, balanced, nothing like the 29.8% of the PPI hour. And on direction it did the thing nobody would have guessed. Bitcoin rose 1.29% in the release hour and another 1.55% in the hour after, at a 57.1% buy share, on a report that pushed the chance of a rate hike to 90%. Hot inflation, higher rates, and the risk asset went up.

Then it gave most of it back. The 14:00 hour fell 0.56%, the 15:00 hour fell 1.35%, and after touching $79,890 the session closed at $77,226, up 0.86%. Whatever the first two hours were, they were not the day.

The streak ended at eight

Friday's full session buy share was 49.98%, below 50 by two hundredths of a point, which stretched the run of net selling sessions to eight. Saturday read 53.50% and ended it. Sunday is partial and trades near $76,750, slightly lower. Consistent with the no bounce finding, and, as always, one observation is not evidence.

One correction to Thursday's figures: we published Thursday's session with two hours left, at $77,129 and 1.50% lower. It closed at $76,569, down 2.22%, at a buy share of 41.95%. The partial label was there, and the final numbers are here.

What to watch on Wednesday

At 18:00 UTC the Fed announces. We measured Bitcoin on the last eight decision days in Friday's explainer: six of eight closed lower, all three cut days closed lower, and the decision hour traded about three times a normal hour. A hike would be a fourth event type for that table. We will report what the hour and the day did, in the same format, whichever way it goes.