The SPR Is Falling Because Washington Lent Its Oil Out on Credit, and Nobody Has Recorded a Barrel Coming Back

The Strategic Petroleum Reserve held 415.4 million barrels in the week the Energy Department announced a 172 million barrel emergency release. It held 286.6 million in the week ending August 28, the lowest weekly reading since November 1982 in a series that begins in August of that year. The 128.8 million barrels in between did not leave as a sale. They left as an exchange: thirteen companies were awarded 133.56 million barrels under five award sheets, take title at the meter, and owe the reserve those barrels plus a premium at dates of their own choosing that run as late as July 2029. Not one returned barrel is on the public record.

Six numbers that are not the same number

Announcement, solicitation, award, delivery, return obligation, actual return. Six stages, six figures, and only the first of them is 172.

• Announced: 172 million barrels. DOE, March 11, 2026: a release of 172 million barrels from the SPR, the U.S. share of a 400 million barrel action by 32 IEA members, with "approximately 200 million barrels within the next year" to come back and "at no cost to the taxpayer". (Our page on how the 2026 release worked country by country covers the international half.)
• Offered: 258.5 million barrels, across five requests for proposals between March 13 and June 10: 86, 10, 30, 92.5 and 40 million. DOE offered half again what it announced, because an exchange only moves what someone bids for.
• Awarded: 133.56 million barrels, on five award-information sheets dated March 20, April 10, April 17, May 11 and June 22. The recipient lines sum to each sheet's own total in every round. That is 77.7 percent of the announcement and 51.7 percent of what was offered.
• Delivered: about 128.8 million barrels, by inference. The award sheets do not report deliveries. DOE gave two cumulative figures in prose, "more than 10 million barrels" by April 17 and "approximately 35 million" by May 11. The EIA weekly series fell 10.4 million through the week ending April 17 and stood 31.3 million lower on May 8 and 41.3 million lower on May 15 (31,347,000 and 41,267,000 barrels on the series, a 9,920,000 barrel week), so the prose and the series agree where they overlap. From there we read the series as the delivery meter: 128.8 million barrels out between the week ending March 13 and the week ending August 28 (415,442,000 to 286,604,000, a fall of 128,838,000), 96.5 percent of the 133,560,000 awarded. This is an inference. The weekly figure nets every barrel that moves in either direction, and DOE has published no delivery total since May.
• Owed back: the barrels plus a premium, each contract on its own schedule. The RFPs set return windows, not a date: as early as September 1, 2026 for round 1 barrels returning to Big Hill, and as late as July 2029 for round 2 barrels returning to Bryan Mound and West Hackberry. "Within the next year" is what the announcement said. The contracts do not require it.
• Returned: none reported. The earliest permissible return date passed on September 1. The EIA series has not risen in any week since March 20, and DOE has announced no receipt.

Who took it

Thirteen firms, from 34.35 million barrels down to one million. An award sheet is a name and a barrel count and nothing else, so this list is the whole of what the public record says about who has the oil.

• Trafigura Trading LLC, 34.35 million barrels, in four of the five rounds, 25.7 percent of everything awarded
• Marathon Petroleum Company, 22.10 million (16.5 percent)
• Shell Trading (US) Company, 18.10 million (13.6 percent)
• ExxonMobil Oil Corporation, 14.40 million (10.8 percent)
• Macquarie Commodities Trading US, 11.05 million
• BP Products North America, 8.10 million
• Vitol, 6.00 million, including the only award of the June round
• Phillips 66, 5.55 million
• Mercuria Energy America, 4.50 million
• Gunvor USA LLC, 4.185 million
• Energy Transfer Crude Marketing, 2.525 million
• Atlantic Trading & Marketing, 1.70 million
• Alon USA, 1.00 million

The top four hold 88.95 million barrels, 66.6 percent. Six of the thirteen are trading desks rather than refiners (Trafigura, Vitol, Mercuria, Gunvor, Macquarie and Atlantic Trading), and between them they hold 61.8 million barrels, 46.3 percent. The sheets vary a legal suffix between rounds (Marathon Petroleum Company LP in March, Marathon Petroleum Company in April) and we treat each pair as one recipient; whether any pair is two contracting entities is not settled by the record. Which site, which delivery month and which return period each firm took is not on any sheet, except that the April 10 and April 17 rounds were single-site, so those barrels came from Bryan Mound and West Hackberry respectively.

What the contract does: title moves at the meter

Here is what happens when one of these awards is executed, from the RFP's own terms. A trader bids on, say, 5 million barrels of Bryan Mound sour crude, crude with enough sulfur in it that only a refinery built to strip it out can run it, for June delivery, offering a premium above the round's minimum of 18 percent. DOE ranks the offers by premium, awards, and the trader's barrels flow to the custody transfer meter at the site. At that meter, under section H.1(a) of every one of the five RFPs, "title to the Exchange Oil delivered from the SPR will be transferred to the Contractor". The oil is now the trader's. It can be refined, sold to a refiner, loaded on a ship or held in a tank; the contract restricts none of that, and the only marine clause is the Jones Act requirement on the vessels that carry it. What the trader owes is "Return Oil": the 5 million barrels plus the premium, so 5.9 million or more, delivered back to the site inside the return period the trader picked from the RFP's menu, with title passing back at receipt under H.1(b). Each later return period carries a further adder, half a point to 4.5 percentage points, so waiting costs barrels.

Strip the government nouns and this is a commodity loan. Barrels now, more barrels later, the premium is the interest, paid in kind, and what the borrower does with the principal in the meantime is the borrower's business. In March 2026 that principal was worth a great deal, and the contract lets the holder monetize it at once and buy the return barrels whenever, inside the window, they are cheapest. That is legal, published, and the point of an exchange: DOE gets barrels back without an appropriation. It is also why the return schedule matters more than the headline. "Within the next year" is the announcement's reading of the arrangement. September 2028 and July 2029 are the ones in the contracts.

What DOE says it secured, and what the sheets let anyone check

The announcement promised about 200 million barrels back for 172 out, which is a 16 percent premium. Since then DOE has stated the premium four ways. March 20: 55 million barrels to be received against the 45.22 million on that round's award sheet, which is 21.6 percent. April 30: a 24 percent premium on the earlier exchanges. May 11: "approximately 28 percent return premium, representing 15.1 million barrels" for the second round, which is 28.3 percent of 53.33 million and above the top of that round's own minimum range of 18 to 24 percent. June 10: "a 26 percent premium in returned barrels" on the earlier exchanges. Four statements, four scopes, no reconciliation, and none of them can be derived from the award sheets, which carry no premium at all. Winning offers went to whoever bid furthest above the minimum, per section B.5, and the winning ratios were never published. The same June 10 announcement counted "more than 133 million barrels across three completed exchanges"; the sheets show four award rounds before June, totalling 133.06 million. Two DOE documents, two ways of counting the same barrels.

Then the June round. DOE offered 40 million barrels of sour crude with minimum premiums of 8 to 9 percent, less than half of any earlier round's, and awarded 500,000 barrels, to Vitol. That is 1.25 percent take-up after rounds that drew 53 to 87 percent. DOE has published no explanation. The lower minimums are on the RFP's face; what bidders made of the price of crude in June 2026 is their business, and the bid record that would answer it is not public.

None of this is hidden in the sense of withheld. It is unpublished. The award sheets, the RFPs and the amendments sit on a DOE server behind a page that never shows the file addresses, and DOE's own history-of-releases page listed no 2026 entry as of September 9, 2026. We fetched the documents from the posting API directly; the paths are in the sources below.

"No cost to the taxpayer" is a claim, not a finding

It might be true. It cannot be checked from the public record, and it is the wrong shape for a finding.

A present-value test would need, for each contract: the barrels delivered and the month they left; the barrels owed back and the return period selected; the market value of the outgoing barrels at delivery and of the incoming barrels at return; a discount rate for the two to three years in between; the cost of running the caverns, moving the oil and re-certifying its quality on the way back; and the value of the reserve's absence during the window, which is the whole reason a strategic reserve exists. The record publishes the first item as a total and none of the others per contract. A sale shows up in the federal budget as receipts. An exchange shows up nowhere, which is much of its appeal in Washington and most of the problem for anyone trying to audit it. A 26 percent premium in barrels over thirty months is a good deal for the reserve if crude is flat, and a poor one if the 2026 price spike gives way to a glut by 2028, and nobody outside DOE holds the schedule that would let them run the numbers. We are not asserting a loss. We are saying the claim is unaudited, and that it stays unaudited until DOE publishes the per-contract return schedule or the barrels come home.

Favoritism: what the sheets can and cannot show

Four firms hold two thirds of the barrels. Six trading desks hold nearly half. Trafigura alone holds a quarter, across four of five rounds. Those facts are consistent with an open competition among the handful of companies able to take 10 million barrels of sour crude at a Gulf Coast meter inside sixty days, and they are consistent with something worse. The award sheets cannot tell the two apart, and neither can we.

What would: the number of offerors per round; every offer, winning and losing, with its premium and return period; the ranking under section B.5 that produced each award; the evaluation memoranda; and any modification to an awarded contract, since the RFP amendments are posted (three for round 1, one for round 2) and contract modifications are not. Those records exist. They are the ordinary product of a federal solicitation, and they are what a records request or a congressional letter would ask for. Until they are public, this page reports the concentration as a fact, the missing scoring record as a limitation, and no verdict on motive in either direction. Absence of published evidence is not evidence of a clean process, and it is not evidence of a dirty one.

Method and receipts

The award figures come from DOE's five award-information sheets, fetched from the spr.doe.gov posting API on September 9, 2026 and read line by line; the recipient lines sum to each sheet's stated total in all five rounds, 45,220,000 + 8,480,000 + 26,030,000 + 53,330,000 + 500,000 = 133,560,000 barrels. The contract terms come from sections B.1, B.5 and H.1 of the five RFPs. The level figures are the EIA weekly SPR crude stocks series in our own archive, every week EIA has published since August 20, 1982, 2,292 readings, re-pulled for this piece (the calendar holds 2,298 weeks; the six EIA never published fall in September 1982 and April to May 1983); the lowest-since anchor was found by scanning the full series, not by reading a headline. Our June 19 page, The SPR is draining and did not stop at the ceasefire, called 340.3 million the lowest since 1983; at 286.6 million the anchor moves to November 1982. The delivery inference is the difference between two readings of that series and is labelled as such everywhere it appears. The premium statements are quoted from the DOE announcements they appear in, with their dates, and are not averaged. Third-party trade-press breakdowns of the rounds were read and matched the sheets; they are not cited as sources.

Next reading: the EIA weekly petroleum status report of September 10, for the week ending September 4, then every Wednesday, on /oil/spr. The grade lands December 2, on this page.