Where the Emergency Oil Went

Correction, 29 August 2026. We wrote that the first Strategic Petroleum Reserve exchange return window opens on 1 September 2026, and told readers to watch September. It should be November 2026: the first solicitation's borrowed barrels are due back between November 2026 and September 2028.

Start at the gas pump. This year, when the war made the Strait of Hormuz too dangerous for most oil tankers, the price of Brent crude, the international benchmark, jumped from about 71 dollars a barrel at the end of February to above 100 by the middle of March, and you probably heard some version of the same line on the news: the government is releasing oil from the Strategic Petroleum Reserve. It sounds simple, like draining a tank when the fuel light comes on. What actually happened is stranger than that, and more interesting, and you can understand every part of it with nothing but a little arithmetic. This is not an argument about the war. It is an explanation of a machine.

The tank

After the oil embargo of 1973, when Middle Eastern producers cut off shipments and Americans sat in gas lines for hours, Congress decided the country needed an emergency stockpile. It created one in 1975, and the result is the Strategic Petroleum Reserve. It is not a tank farm. It is a set of enormous caverns hollowed out of underground salt formations along the coasts of Texas and Louisiana, at four sites with names worth knowing because they show up later: Bryan Mound and Big Hill in Texas, West Hackberry and Bayou Choctaw in Louisiana. Salt is useful here because it does not leak and it heals itself under pressure, so oil can sit in a salt cavern for decades.

A barrel is 42 gallons. The reserve is measured in millions of barrels; the whole world burns about 103 million of them every day, which means the entire American reserve, at its fullest this year, held roughly four days of global consumption.

When the war began at the end of February, the reserve held about 415 million barrels. As of the end of July it holds about 305 million, a level not seen since 1983, when the reserve was still being filled for the first time. That is about 110 million barrels gone in five months. In May, the busiest month, oil left the caverns at about 1.3 million barrels a day. That is faster than any comparable stretch we can find in the reserve's record; by our own arithmetic on the government's weekly figures it beats even the best single week of the big 2022 release after Russia invaded Ukraine, which ran at about 1.2 million a day.

The part that gets skipped

The government did not sell that oil. It lent it.

The program is called an exchange, and the mechanics are simple. A company takes barrels out of the caverns now. In return it signs a contract promising to put the same number of barrels back later, plus extra barrels on top. The extra barrels are the interest, and the interest is paid in oil rather than money. The Energy Department set a floor for that interest in each solicitation, generally 18 to 22 percent, and told bidders to bid higher. We can see what one whole batch came to: the 53.33 million barrels awarded in May carry 15.1 million barrels of premium, which is about 28 percent for that batch taken together. What any individual company agreed to pay was never published. The repayments are due between September 2026 and September 2028. The contracts also require that the oil coming back be American-produced crude, so the companies cannot simply buy foreign oil to settle up.

The Energy Department was direct about why it liked this structure. In its announcement it said the reserve would get back roughly 200 million barrels for 172 million lent, within the next year, at no cost to taxpayers. The government spends nothing, and if every promise is kept, the reserve ends up bigger than it started. Hold on to that phrase, within the next year. The contracts the department actually signed run to September 2028.

Who borrowed it

The Energy Department announced how much oil went out but never announced who took it. The names are in the award documents, which the department posts with its procurement paperwork rather than in any announcement, covering every award we could find from March through June:

† Gunvor, along with Vitol and Trafigura, has acquired refining capacity since 2022; Gunvor's is the largest, roughly 200,000 barrels a day across two wholly-owned plants. Independent coverage still describes all three primarily as trading houses. ‡ Energy Transfer is a midstream pipeline company; its own securities filings describe active crude buying/selling at pipeline hubs, but it owns no refineries and isn't a standalone trading house either. Excluded from the trader/refiner split below. § Alon USA Energy Inc. was acquired by Delek US Holdings in 2017; the refining entity today operates as Alon USA Partners LP, majority-owned by Delek.

Some of these are refiners: companies like Marathon Petroleum, Phillips 66 and Alon own the American plants that turn crude oil into gasoline, diesel and jet fuel, so when they borrow crude they generally run it through their own equipment. Others are commodity trading houses: Trafigura, Vitol, Gunvor, Mercuria, Macquarie and Atlantic Trading and Marketing, TotalEnergies' physical trading arm, buy and move physical cargoes around the world for profit rather than refining them, so when they borrow crude they generally sell it wherever it fetches the most money. Shell, ExxonMobil and BP do both. Energy Transfer fits neither: it's a pipeline company whose crude-marketing arm buys and sells at pipeline hubs but owns no refineries and runs no standalone trading book, so it sits outside the split entirely. By our tally, the trading-house side accounts for 46 percent of the volume borrowed. In the comparable 2022 program, when the government sold rather than lent, refiners took the large majority of the barrels: about 84 percent under the same standard applied above, where a company's trading arm doesn't count as its parent's refining business; EPRINC's own published tally, which counts those arms as refiners, puts it at 93.3 percent.

Why a company would want this deal

During a war, oil for delivery today is expensive, because buyers are frightened and supply is uncertain. Call it 100 dollars a barrel. But oil for delivery next year is cheaper, say 85 dollars, because the market expects the war to be over by then and supply to be normal again. Traders call that pattern backwardation. It is not a law of nature; it is a forecast, the market's collective opinion that today's problem is temporary.

Now be the borrower. You take a barrel out of the reserve and sell it today for 100 dollars. At the same moment, you arrange to buy next year's cheaper oil, at 85 dollars, to pay the government back. Even after buying the extra interest barrels you owe, you can clear a few dollars per barrel, locked in at the start, without betting on where prices go afterward. Multiply by millions of barrels and you can see why thirteen companies lined up. Some of that advantage goes back to the government, because the companies compete for the barrels by offering higher interest, and the highest offer wins. How much went back is not publicly known, for a reason we will come to.

The gap between today's frightened price and next year's calm one is what makes the trade work, not the price of oil being high alone. Both halves are necessary. If the market feared the war would grind on for years, next year's oil would be expensive too, the gap would close, and there would be nothing left to pay for the trade.

Where the oil actually went

First, American refineries genuinely ran harder. A refinery's crude runs are simply how much oil it feeds into its equipment each day, and utilization is how full it is running compared to what it could do flat out. Both rose through the spring. Refinery runs were higher in every month from March through July than in the same months last year, though in May and June the margin was slight, and by July American refineries were running at 96 to 97 percent of capacity, which is close to the physical maximum. Something had to feed that, because the oil normally arriving from the Persian Gulf had nearly vanished: American crude imports from Gulf countries fell from about 682,000 barrels a day in March to about 186,000 in May.

Second, a large share of the oil left the country. By mid-May, about 13 million barrels of released reserve crude had sailed abroad, roughly 40 percent of what had physically come out of the caverns at that point, according to the analytics firm Kpler reading United States customs documents, as reported by Bloomberg. By June, Kpler put the share at about a third of a larger total, and nine of the first eleven million exported barrels went to Europe. Those three figures all rest on the same firm's cargo tracking, so they are one measurement rather than three. The direction is independently reported: Argus described American reserve crude being offered to European refiners in April at prices that undercut the local grades. American crude exports hit records in April and May.

Both facts are true together. Some of the barrels went to American refineries. Others were sold abroad, where the market pulled them; oil is a global commodity, and nothing in the exchange contracts required the crude to stay in the country.

June, when the machine stopped

The five solicitations, in order:

The full June award, half a million barrels, went to a single company, Vitol. DOE's original target, announced when the program launched, was to lend up to 172 million barrels. The five rounds placed 133.56 million, a shortfall of nearly 40 million barrels against that number.

What changed across those months was not the price of oil so much as the shape of the curve. This is also where the benchmark changes. Brent, the international price that opened this story, isn't what a borrower here actually prices against: because repayment has to come back as American-produced crude, the number that decides whether the trade still works is the domestic one, West Texas Intermediate, priced at Cushing, Oklahoma. The government's own weekly figures show WTI averaging about 92 dollars in the week ending June 12 and about 81 dollars in the week ending June 19, a fall of eleven dollars across the days the bids were due. When the near price drops toward the far price, the gap that pays for the whole trade closes. Run the arithmetic again: borrow a barrel, sell it for 81, and you still owe a barrel plus interest bought at a price not much below 81. There is no trade. A memorandum meant to end the fighting was signed in the days after the bids closed.

That shortfall sat in the salt: the Energy Department's own inventory ledger gives the June release its own accounting code and then records nothing against it. Through the end of July, not one barrel had moved.

The program runs on the gap between today's frightened price and next year's calm one, so it works best while the market believes the emergency will continue, and it slows as confidence returns.

What people who study this for a living say

Economists have been measuring these releases for decades, and the short version is that they can move prices, but usually modestly and inconsistently. Two economists at the Dallas Fed put the effect at about 2 dollars a barrel after the 1990 Kuwait crisis, about 3 after Hurricane Katrina, and as much as 12 during the 2011 Libya crisis, a spread wide enough to be worth noticing on its own.

Exchanges are more complicated still, because today's added supply creates tomorrow's replacement demand: the borrower has to come back and buy. The one large exchange in the record before this year, the heating oil loan of 2000, first pushed the real price of oil down by about 14 percent, then finished about 2 dollars a barrel higher across the whole life of the intervention. That is a single case, so treat it as a direction rather than a law.

And the 2026 release itself? Oil did not fall. The announcement came on 11 March, and by nine the next morning Brent had risen about 8.6 percent to 98.76 dollars. The Energy Information Administration records it passing 100 dollars later that day and continuing to climb through the month. Prices stayed high through the spring while the reserve emptied at its fastest rate: on the agency's weekly series Brent peaked at 124.61 dollars in the week ending 10 April, and was still at 109.62 in the week ending 24 April. What eventually brought prices down is not something anyone has established; no formal study of this release exists yet, and the Government Accountability Office has said its implications are not yet realized. What is on the record is that demand fell: world oil consumption in 2026 is now expected to run 1.1 million barrels a day below 2025, a figure the International Energy Agency and the Energy Information Administration reach independently; the IEA calls it the first annual decline since 2020.

What everyone else did

America did not act alone. In March, all 32 member countries of the International Energy Agency agreed to a joint release totalling about 426 million barrels, of which America's 172 million was the largest single share.

The other countries mostly used different tools, and one of them is worth understanding. Several contributed simply by lowering the amount of oil private companies are legally required to keep in storage, which does not move a single barrel by itself. It only permits companies to sell down what they were already holding. Whether they do is up to them.

A loan auctioned against the futures curve can fail visibly, on the record, with a number attached. A country that lowers a storage requirement and watches companies quietly decline to sell has the same problem with no number and no headline.

How it gets refilled

This is the part that has not happened yet, and it starts next month.

The thirteen companies owe back the 133.56 million barrels they borrowed plus interest, which comes to somewhere around 160 to 170 million barrels of American-produced crude. That figure is our own arithmetic, and it has to be: the per-company interest rates were never published. It is also well short of the 200 million the department advertised coming back, since only 133.56 million was ever lent against DOE's original 172-million target. The earliest return window opens on 1 September 2026 and the last closes on 30 September 2028.

Whether that is easy or painful depends mostly on something nobody knows: how and when the war ends. If the war winds down and prices fall, the companies buy cheap oil, hand it over, keep their profit, and more oil comes back than went out. The Energy Department's design works as advertised, though the reserve would still sit far below where it stood before the war. If the war grinds on and oil is still expensive in 2027, those same companies have to buy scarce, costly crude and pump it back underground at precisely the moment the country would rather have oil above ground.

History is reassuring only up to a point. Small loans have come back on time; the 5 million barrels lent after Hurricane Harvey in 2017 returned with their premium. Larger ones have been renegotiated: the 30 million barrels of heating oil lent in 2000 were stretched into 2003, and the borrowers paid an extra 3.3 million barrels for the delay. But nothing in the record is anywhere near this size. Every SPR exchange through 2019 comes to about 75 million barrels in total, less than half of what is outstanding now.

The Government Accountability Office measured the system's fill capability at 440,000 barrels a day against a design rate of 785,000, and put drawdown capability at 61 percent of design. That sounds like a bottleneck. For this program it is not one: returning 170 million barrels across a window that runs to September 2028 works out to about 224,000 barrels a day, roughly half of what the system can already manage in its impaired state. Unless the returns bunch at the very end, the limit on getting this oil back is commercial rather than mechanical. It depends on whether thirteen companies buy the crude, not on whether the pipes can take it. One of the four sites, Big Hill, has been recorded at zero drawdown capability during a construction outage not expected to close out before 2028.

The one number nobody outside the government has

Companies compete for these barrels by offering to pay back higher interest, and the highest bid wins; some of that advantage returns to the government this way. How much did the government actually capture, and how much did the borrowers keep?

Nobody outside the Energy Department can say. The department's press releases named no counterparties at all; the award documents, once you find them, name every winner and every volume. But neither carries the exchange ratio, which is the per-company interest rate, and that is the number that decides the split. The omission was not an oversight. The solicitation itself says the department may publish offer data "excluding exchange ratio," the one field carved out in advance, before a single bid arrived.

The public record contains no estimate of what the companies earned on these trades. Congress asked essentially nothing about the terms: Roll Call reported in March that the drawdown had "received little criticism on Capitol Hill," and the objections that did surface questioned whether the reserve could matter at all at this scale rather than how the barrels were priced. The Government Accountability Office reviewed the reserve this year without assessing the exchange terms, writing that the department "plans to complete this release through emergency exchanges, but, as of May 2026, the full timing and implications of this large-scale emergency release are not yet realized."

What to watch

Watch whether the Energy Department offers those stranded barrels again, and whether anyone bids. As of 8 August the department's own list of current solicitations still held only the June offer, last touched on 22 June, so nothing has been put back on the table. A successful auction means fear has returned to the price of oil. A second failure means the market still believes this ends soon.

Watch September, when the first return window opens. Barrels are not due on any particular day; the windows run to 2028. What matters is whether oil starts moving back in, and whether the word deferral appears, which would mean the schedule is being renegotiated. In 2000 a deferral cost the borrowers extra barrels, so it is a priced event, not a free one.

Watch marine war-risk insurance, which is the premium shipowners pay to sail into a dangerous place. Before the war, a Strait of Hormuz transit cost about a quarter of one percent of the vessel's hull value. It has been quoted many times higher since. If it falls back toward pre-war levels, the risk is being priced as over. If it stays high after a deal is signed, the cost of this war has outlived its shooting phase.

The memorandum

Our own July 26 Brief called this instrument closed, listing it among a set of repeating wartime patterns as "a memorandum signed 17 June and voided 17 July." That reading did not hold. Neither government ever formally voided the text: Iran said it suspended it, the United States reversed one of its provisions, and through August both sides were still treating it as the reference document in negotiations. What looked finished in July is what both sides are still negotiating variations of in August.

The memorandum was signed in June by American and Iranian negotiators, mediated by Pakistan and Qatar, meant to end the fighting; it was signed in the same days the June auction, described earlier, was failing. Its fifth article says Iran will make arrangements, using its best efforts, for the safe passage of commercial vessels, with no charge, for 60 days only. The only public copy is an unsigned draft published by an American archive, with the date and place left blank, so its exact terms are worth reading carefully rather than taking as settled.

Whether the market expects this war to end soon is the question the rest of this piece keeps returning to: it decides whether the exchange trade works, and it is what that 60-day clock will test directly. If it runs from the June signing, it expires around the middle of August, and the language is clear that free passage was temporary by design. Whether ships start paying to pass, and who collects, is a fact rather than an opinion, and it is about to become observable.

The short version

The reserve was not drained; it was lent out. Thirteen companies borrowed 133.56 million barrels, and by volume the trading houses took about half of it, where in 2022 refiners and majors took the overwhelming share. American refineries ran harder through the spring while those barrels were flowing, and of the oil that had physically come out by June, roughly a third was sold abroad, mostly to Europe. The lending worked while the market was afraid and faltered as confidence returned; the June auction placed barely one percent of what it offered, leaving nearly 40 million barrels underground. Around 160 to 170 million barrels are owed back, by our arithmetic, on a schedule that runs to September 2028.

The Strategic Petroleum Reserve is insurance, and insurance is a promise about a bad day. The tank now sits at a level last seen in 1983, when the reserve was still being filled for the first time, and the plan to refill it is a stack of promises held by private companies, promises whose cost depends on how and when this war ends, which is the one thing nobody knows. The next time you hear a news anchor say the letters S P R, you will know what is actually down in those caverns: oil, and IOUs.

How we know this

The counterparty list and the auction results come from five Energy Department award documents on the department's own website, listed with its business-opportunity postings. Two of the five cover superseded solicitations and we could recover them only through the Internet Archive. Company classifications in that roster were checked against two independent sources per company rather than asserted from general knowledge. The inventory figures come from a chart the program office publishes as an image, cross-checked against the Energy Information Administration's weekly series, where the two agree to three decimal places. The comparison to 1983 is not ours: we hold no inventory series reaching back that far, and we repeat it only because the level we measured this July sits below the one the published comparison described. The refinery, import and price series are all the EIA's: refinery runs, utilization, Persian Gulf imports, and weekly spot prices for WTI and Brent. The export shares all come from one analytics firm, Kpler, reaching us three separate ways, so they are one measurement rather than three. The 2022 comparison figure is EPRINC's own published tally of the same DOE data. The repayment total, the drawdown-rate comparison and the refill-rate arithmetic are our own, and we have said so where they appear. The one number that would settle who profited, the per-company exchange ratio, does not exist in public at all.