Refining Capacity, Utilisation and the Crack Spreads

United States refining capacity, measured twice, five months apart

This is the Two-Clock rule in its purest form: one quantity, two primary sources, two dates, and a difference with a named mechanism.

These are the same quantity measured five months apart and they differ by about 133,000 barrels per calendar day. EIA supplies the mechanism itself: Valero's Benicia refinery, 145,000 barrels per day, was still counted in the January 1, 2026 annual survey, and its capacity was removed from the monthly capacity estimates as of March 2026. The annual figure is not wrong — it is correct as of its own cover date, and a refinery that shut in the first quarter was operating on January 1. A reader who quotes the annual number in September is quoting a plant that no longer runs.

The gap is 133.5 thousand barrels per calendar day, which is 0.74% of the annual figure. That sounds like a rounding difference and it is not: it is one refinery. The mechanism EIA supplies is Valero's Benicia plant, whose capacity was still counted in the January 1 annual survey and had been removed from the monthly estimates by March 2026.

The operating instruction that follows is narrow and worth stating. Use the weekly figure for anything about the present — utilisation, spare processing capacity, whether the system can absorb a demand increase. Use the annual figure only for the things it uniquely carries, which are the plant-by-plant and state-by-state detail no weekly report contains. Do not mix a numerator from one and a denominator from the other, which is the error that produces utilisation rates above 100 per cent.

The count of operable refineries carries the same warning. There were 131 as of early 2026, down from 132 at the start of both 2024 and 2025. A count is a cruder instrument than a capacity figure — refineries differ by more than an order of magnitude in size — but the direction has been one way for a decade and the direction is the point.

The global picture, and the one figure on this reference without a date

Global refining capacity is the one figure on this reference that does not carry a measurement date. It is an industry compilation, not a figure the Institute has traced to a dated primary publication, and global refinery utilisation is computed against it. The numerator — August 2026 throughput — is dated and primary. The denominator is not. A fresher numerator over the same undated denominator does not improve this figure; it makes it look newer than it is, which is worse. Every utilisation figure quoted on this reference therefore carries a dated top and an undated bottom, and that is why it is quoted to the nearest whole percentage point and never to a decimal.

Global refinery throughput of 81.4 million barrels a day in August 2026 against capacity of about 103.3 million gives a utilisation rate of roughly 79%. It is quoted to the whole per cent and never to a decimal, for the reason in the note above: the numerator is dated and primary, and the denominator is an industry compilation with no measurement date. A decimal place on that quotient would be a claim the underlying figures cannot support.

Read the throughput figure against its own history rather than against the capacity. August 2026 throughput was up 960 thousand barrels a day on the month and still 4.2 million below August 2025. That is the shape of this episode: refineries running hard into extraordinary margins, and the system as a whole still processing far less crude than a year earlier because the crude is not there to process. The barrels missing from the top of the system are on the production ledger, and the passages they would have moved through are on the chokepoints ledger.

The crack spreads, computed from five price legs taken in one week

All five legs are the week ended October 16, 2026. That single condition is the only reason the arithmetic below means anything.

Crude leg: WTI at Cushing at $99.93 a barrel, which is $2.37929 a gallon over 42 gallons. Product legs and crude leg source: EIA spot price series: WTI at Cushing; New York Harbor conventional gasoline regular; New York Harbor Ultra-Low Sulfur No. 2 Diesel; New York Harbor No. 2 heating oil — five-day Monday-to-Friday unweighted means for the week ended October 16, 2026, READ OFF the publisher's own weekly spot price series, release date October 21, 2026, next release October 28, 2026. The publisher states the method on the same page: weekly prices are calculated by EIA from daily data by taking an unweighted average of the daily closing spot prices over the period, which is the arithmetic the Institute had been performing by hand for two sweeps and no longer needs to. AMENDMENT, October 22, 2026 — THE PRIOR ENTRY'S CLOSING CLAIM IS WITHDRAWN AND IS RETAINED HERE RATHER THAN SWAPPED OUT. It read: these legs are 'also, from October 21, 2026, the only arithmetic available: the publisher has announced that Tables 11 and 12 are discontinued from that date, with spot prices moving to daily publication, which makes a daily-to-weekly derivation the standing method for this row rather than a one-week expedient.' The first half of that is correct and still stands: Tables 11 and 12 of the weekly status report are discontinued from October 21, 2026 and the week ended October 9 was the last full week they carried. The conclusion does not follow. The discontinued tables are one surface; the weekly spot series is a different one, it was never announced as discontinued, and on October 21, 2026 it published a column for the week ended October 16. So a published weekly average exists for this week and the derivation is not the only arithmetic available — it is a fallback, correctly built and now not needed. The discriminator that settles it, and the reason this is an amendment rather than a bare correction: the page's own Release Date line reads 10/21/2026 and its newest column header reads 10/16/26, both of which were read before any figure was taken off it, and the 10/09/26 column on that same page reads WTI 96.88, gasoline 3.177, ULSD 4.911 and heating oil 4.567 — every one of the four legs the Institute derived by hand last week, reproduced to the last printed decimal. The derivation was right about the numbers and the sentence was wrong about the publisher. The validation recorded in the superseded text is retained as evidence rather than dropped, because it is what makes the fallback usable if the weekly series ever goes dark: applied to the week ended October 2 the five-day mean reproduced every one of that week's published weekly averages exactly — WTI 98.07, gasoline 3.433, ULSD 4.871, heating oil 4.661 and Brent 120.03 — and applied to the week ended October 9 it reproduced all four legs that the published series has now independently confirmed. The method is therefore validated twice, against two different weeks, and is held in reserve. Every crack figure in the last two columns is the Institute's computation from those legs and from nothing else.

A New York Harbor product crack is properly struck against a seaborne crude — Brent, or a Gulf Coast grade — not against WTI at Cushing, which is landlocked a thousand miles inland. The Institute does not hold a Brent quote for this exact week and so computes against WTI and says so. Because WTI normally trades below Brent, a WTI-based crack normally overstates a coastal refiner's margin by roughly the Brent-WTI spread. 🔴 AMENDED on the September 30, 2026 sweep, and the prior wording is retained above rather than deleted. That wording continued: 'In this episode that spread has behaved abnormally and has at times inverted, so the direction of the error is not stable.' The September 30 amendment then read that on the nearest observation the spread was neither inverted nor small — Brent at $102.58 against WTI at $88.92, $13.66 a barrel — so the direction of the error was determinable and ran one way. 🔴 AMENDED AGAIN on the October 2, 2026 sweep, and the September 30 figure above is retained and labelled rather than swapped. The DIRECTION still runs one way: the cracks below overstate a coastal refiner's margin. What has changed is that the SIZE can no longer be read off the two front-month quotes this reference carries, because they are no longer on the same contract month. The publisher's Brent series rolled from November to December between the September 30 and October 1 quotes without saying so — the price ledger sets out the evidence — while its WTI series stayed on November. Differencing the two October 1 opens gives $7.58, and that figure is not a spread; it is a November barrel subtracted from a December barrel, and the apparent $6.08 narrowing against the September 30 reading is almost entirely the month mismatch. What can honestly be quoted instead are the exchange's own same-month spreads on the September 30 settlements, which are the newest both-months figures this reference holds: $13.11 on the November contracts and $9.61 on the December contracts. Both are real, they differ by $3.50, and a reader needing the size of the overstatement should take whichever month matches the barrel the refiner actually buys rather than averaging them. The clock gaps are stated rather than hidden, because they are the reason this is an amendment and not a correction. Three clocks were in play on that sweep: the product legs were Friday September 25 closes, the crude quotes were October 1 opens, and the two same-month spreads were September 30 settlements. What could be said was the direction; what could not be said was the size of the overstatement for that exact week. 🔴 AMENDED AGAIN on the October 8, 2026 sweep, and this amendment RESOLVES the open question above rather than restating it. The two paragraphs preceding this one are retained and labelled, because the reasoning in them was correct and the constraint they described has been lifted by a change of series rather than by a change of market. The size of the overstatement could not be stated because the only Brent and WTI quotes this reference held were FUTURES front-month opens, and the two had rolled onto different contract months. The crack legs are now read off the publisher's WEEKLY SPOT series, and that series carries Brent and WTI on one definition, one week, and no contract month at all — a spot average has nothing to roll. For the week ended October 2, 2026 it reads Brent $120.03 and WTI $98.07, a spread of <strong>$21.96 a barrel</strong>, or about 52.3 cents a gallon. The prior week on the same basis reads $117.08 and $93.57, a spread of $23.51, so the spread NARROWED $1.55 while both legs rose. That figure is the size of the overstatement, and it is the first time this note has been able to give one for the week the cracks are actually struck on. The three cracks above therefore overstate a coastal refiner's margin by roughly $21.96 a barrel, and a reader wanting the Brent-based figure should subtract it: the gasoline crack becomes about $24.16, the distillate crack about $84.55, the 3-2-1 about $44.29. 🔴 THE LESSON IS ABOUT THE SERIES AND NOT ABOUT THE SPREAD, and it is worth more than the number. The problem these amendments have been describing for two sweeps was never a market problem; it was an artifact of comparing two futures series that each carry a contract month the quote does not name. The Institute spent two amendments documenting a mismatch that a differently-defined series does not have. Where a comparison keeps breaking on a definitional seam, the question to ask is whether a series exists that has no seam, before building further apparatus to reason across it. 🔴 AMENDED OCTOBER 16, 2026 — THE SPREAD THIS NOTE EXISTS TO MEASURE HAS WIDENED BY HALF AGAIN, AND ON THESE LEGS THE BRENT-BASED GASOLINE CRACK IS NOW NEARLY ZERO. On the week ended October 9, 2026, both legs computed as five-day means from the same daily table, Brent averaged $127.43 and WTI at Cushing $96.88. The spread is $30.55, from $21.96 the week before — $8.59 wider in one week, against a prior-week widening of a dollar and a half. Subtracting it: the gasoline crack becomes about $6.00, the distillate crack about $78.83, the 3-2-1 about $30.28. The gasoline figure is the one to sit with. A Brent-based New York Harbor gasoline crack of six dollars is a margin that would not cover the cost of running the unit, and this note's own standing position is that neither basis is the refiner's actual economics — a Gulf Coast or East Coast refiner buys a slate that is neither pure Brent nor pure WTI. What the pair of figures does establish, and it is the only thing it establishes, is that on this week the ENTIRE gasoline margin on the WTI basis is the crude spread. That is a statement about where the spread sits, not a verdict on anybody's refinery, and no inference about refining behaviour is drawn from it here or anywhere else on this page. 🔴 AMENDED OCTOBER 22, 2026 — THE SPREAD NARROWED FOR THE FIRST TIME IN THREE WEEKS, AND THE BRENT-BASED GASOLINE CRACK HAS CROSSED ZERO ANYWAY, BECAUSE THE GASOLINE LEG FELL FASTER THAN THE SPREAD CAME IN. On the week ended October 16, 2026, both legs now read off the publisher's own weekly spot series rather than derived from dailies, Brent averaged $125.59 and WTI at Cushing $99.93. The spread is <strong>$25.66</strong>, from $30.55 the week before — $4.89 NARROWER, which reverses two consecutive weeks of widening and is the first narrowing on this basis since the week ended October 2. Subtracting it from the WTI-basis cracks: the gasoline crack becomes about <strong>minus 81 cents</strong>, the distillate crack about $85.54, the 3-2-1 about $27.98. A NEGATIVE NUMBER HERE IS AN ARITHMETIC RESULT AND IS NOT A CLAIM THAT ANYBODY LOST MONEY, and the refusal the prior amendment stated at six dollars is repeated rather than abandoned now that the sign has gone. What the figure says, and the only thing it says, is that on this week the crude spread between two spot assessments is 81 cents larger than the entire New York Harbor gasoline margin measured against the inland grade. It does not say a coastal refiner ran gasoline at a loss. This note's standing position is that neither basis is a refiner's economics: a Gulf Coast or East Coast plant buys a slate that is neither pure Brent nor pure WTI, on term contracts at negotiated differentials, and the true crude cost sits somewhere between the two legs rather than at either of them. A margin struck against the more expensive endpoint of that range is the pessimistic bound, not the answer. AND THE DIRECTION OF THE TWO MOVES MUST NOT BE COLLAPSED INTO ONE, because they point opposite ways and a reader who sees only the crossing will get the cause backwards. The spread narrowing is a move in the refiner's favour: $4.89 a barrel of the overstatement came off. It was more than consumed by the gasoline leg, which fell 20.6 cents a gallon — $8.65 a barrel — while crude rose $3.05. So the Brent-based gasoline crack fell about $6.81 on the week despite the spread improving, and attributing the crossing to the crude spread would be wrong by sign. The figures below should still be read as the shape of the margin rather than as a refiner's realised margin: these are spot assessments, a refiner buys on term contracts at negotiated differentials, and nothing here nets out the cost of running the plant.

The single most consequential line in that table is the distillate crack at $111.20 a barrel against gasoline at $24.85. Distillate is out-earning gasoline by 205.6¢ a gallon, or $86.35 a barrel, in the same week from the same crude. A refiner facing that spread will push every unit of yield flexibility it has toward distillate, and the limit it runs into is physical rather than commercial. That constraint, and why the crude removed from this market was the crude best suited to making distillate, is the subject of the diesel and distillates ledger.

The 3-2-1 crack of $53.64 a barrel is the conventional composite: two barrels of gasoline and one of distillate out of every three barrels of crude, which approximates a simple American refinery's yield. It is a useful single number and it hides the thing that matters in this cycle, because averaging a strong distillate crack with a weaker gasoline crack conceals precisely the divergence a refiner is responding to. Read the composite for the level and the two components for the behaviour.

Note also what a crack is not. It is a gross margin between two market prices, not a profit. It carries none of the energy the plant burns to do the conversion, no maintenance, no capital charge, no hedging result. A refiner's realised margin is lower and is not published weekly by anyone.

The unit trap, and how the same crack is quoted two ways

EIA quotes crack spreads in cents per gallon and the trade press quotes them in dollars per barrel. Forty-two gallons separate the two conventions, so a 74-cent distillate advantage is about $31 a barrel, and a figure quoted in the wrong unit is wrong by a factor of forty-two. Every crack on this page is computed in both units from the same underlying legs.

These two are EIA's own published statements and they are period averages rather than dated measurements, which is why they sit in a separate table from the computed cracks above. A period average and a single-week computation are different instruments and they are not reconciled here.

What the official forecast says, and where it is already behind

Forecast made forecast completed October 1, 2026. Printed as a forecast throughout.

Source: EIA Short-Term Energy Outlook, released October 6, 2026 (forecast completed October 1, 2026; next release November 10, 2026). The 2025 column is an actual and the other two are projections; the columns are not the same kind of number and the header says so.

🔴 THE WITHDRAWN WTI ROW HAS RETURNED, ON THE CLOCK THE INSTITUTE SAID IT WOULD REQUIRE. The August 11 table quoted a WTI spot forecast of $80.88 for 2026 and $65.39 for 2027. The September 9 release did not restate WTI in the published summary tables the Institute had read, so rather than carry an August figure in a September table — where it would have looked like a September number — the row was removed and this note recorded that it would return when the Institute had read it on the current clock. The October 6, 2026 release restates WTI directly, at $88.21 for 2026 and $79.74 for 2027, so the row is back above. The superseded August pair is labelled in the comment block beside this table rather than deleted, and the gap of one edition is what the reader is being told about: a withdrawn row is not a missing row, and the condition for its return was written down before it was met. The forecast moved toward the observation again, which is the third consecutive edition in which it has. The August release put retail diesel at $4.85 for calendar 2026 while the weekly series was already printing above $5.90; September raised it to $5.07 and October has raised it again to $5.19, a 6.9 per cent revision across two months, with the 2027 figure moving $4.07 to $4.40 to $4.49. Gasoline has done the same thing more quietly, $3.78 to $3.84 to $3.91 for 2026 and $3.29 to $3.35 to $3.56 for 2027. 🔴 AND THE CRUDE REVISION IS THE LARGER ONE, WHICH IS A CHANGE IN WHAT THE AGENCY IS SAYING RATHER THAN A CHANGE IN DEGREE. Brent for 2026 went from $91.00 to $96.32, up 5.9 per cent in a single edition, and the fourth-quarter 2026 figure went from $90.66 to $104.69, up 15.5 per cent. For 2027 Brent moved $74 to $83.74 and WTI $69.74 to $79.74, each about ten dollars. Through August and September the revisions were concentrated in the product forecasts — the agency was conceding refined-product tightness while holding its crude view. In October it has moved the crude view as well, and it has moved the out-year by as much as the current year. A forecast that lifts 2027 by ten dollars is no longer describing a disruption that resolves inside the forecast horizon. The product forecasts still sit far below the current weekly prints, and the distance has not closed. EIA's own cited actuals for September 2026 are $4.35 a gallon for gasoline and $6.29 for diesel, against full-year 2026 forecasts of $3.91 and $5.19. Diesel averaged $5.600 in May 2026 on the monthly series and has since printed a run of weekly records — $5.967 in the week ended September 7, 2026, $6.285 a week later and $6.529 in the week ended September 21. A $5.19 annual average remains arithmetically possible, since January and February were low enough, but it still implies a fall through the fourth quarter that has not begun. These are forecasts, and the Institute prints them as forecasts. 🔴 WHAT THIS FORECAST WAS ALLOWED TO KNOW. The October edition was completed on October 1, 2026. It accounts for the 40 million barrels of Strategic Petroleum Reserve exchanges announced September 29, 2026. It does not account for the G7 announcement of October 2, 2026, which post-dates the completion date by one day. A completion date is not a formality on a forecast table; it is the boundary of the evidence the forecast was built on, and a reader comparing these numbers against this week's news is comparing across that boundary.

The retail diesel line is the one to watch. A full-year 2026 average of $5.19 against a weekly print of $6.276 in the week ended October 19, 2026 is arithmetically possible — January and February were low enough — but it requires a fall through the fourth quarter that has not started. And the 2027 figure of $4.49 is a statement that the disruption resolves. Set that against the industry's own survey expectations on the chokepoints ledger, where the modal firm expected normal flows by August 2026 and was wrong.

What this page cannot compute, stated rather than approximated

A current United States refinery utilisation rate. Utilisation is gross inputs to crude distillation units divided by operable capacity. The Institute holds the denominator for the week ended September 4, 2026 and does not hold the numerator for that same week from a primary source. A utilisation figure could be assembled by pairing the weekly capacity with a gross-inputs figure from a different week, and it would be wrong in exactly the way this reference exists to prevent. So the two capacity clocks are printed above and the quotient is not.

A dated global capacity figure. Covered in the note in the global section: the denominator is an industry compilation and every utilisation figure computed against it inherits an undated bottom. This is the weakest figure on the whole reference and it is labelled as such everywhere it appears.

A refiner's realised margin. Not published weekly by anyone, and the crack is not a substitute for it. Where the Institute reports all-time-high Atlantic Basin margins, that is the IEA's characterisation of the crack environment rather than an audited profit figure, and it is attributed as such in the global table above.

Where the margin sits, and what it is telling refiners

The rest of the reference