EuroOilWatch Analysis — a bounded estimate of the maintenance Europe borrowed from its own future to get through the shock, and when the bill falls due.
Brussels asked refiners to defer maintenance. Nobody has published what that cost — so we estimated it.
On 30 March 2026, EU Energy Commissioner Dan Jørgensen wrote to member state energy ministers ahead of an emergency council session. Among the measures he set out — demand reduction in transport, wider use of biofuels, closer coordination through the Energy Union security task force — was a single line about the refining system. The letter's own language was an encouragement — "Member States are encouraged to defer any non-emergency refinery maintenance" — which the Commission's public announcement the next day rendered more firmly as "any non-emergency refinery maintenance should be deferred," to safeguard product availability during the Middle East disruption. It was a coordinated policy request, not a binding order.
The request was defensible. The Strait of Hormuz was severely disrupted. Gulf export refineries were down. Europe was drawing on emergency stocks and competing for replacement cargoes in a tightening market. Removing European conversion capacity for planned work would have cut product output at the worst possible moment.
Nearly four months later the position holds: on 24 July the EU Oil Coordination Group said there was "no supply problem at this stage," while warning that a prolonged conflict could significantly tighten markets in the coming weeks and months.
Both statements are compatible. Europe may have avoided an immediate shortage partly because refineries ran through windows that would ordinarily have taken units offline.
The question is what that borrowing costs, and when it falls due.
This article attempts an answer. Not a precise one — the underlying data is not public — but a bounded one, with the assumptions stated so they can be attacked.
The measured intervention and the unmeasured one
Emergency stock releases are counted in public. The Commission noted that EU countries contributed roughly 20% of the more than 400 million barrels released under IEA coordination. Markets can estimate what entered circulation and what remains.
Deferred maintenance is invisible by comparison. There is no European register showing which refineries postponed work, which units were affected, how much capacity stayed online as a result, how long the work slipped, or whether the revised windows now overlap.
Turnaround timing is treated as competitive intelligence across the industry — refiners do not publish schedules, because timing reveals regional supply and affects contractor pricing. That commercial norm is understandable. It also means a continent-wide policy request on maintenance scheduling happened without a public number attached.
There is one detail that makes this more than an information gap. Jørgensen's letter asked ministers to report back on current market capacity and to propose practical actions. Member states were asked for data. Some of the answer to "how much maintenance moved" therefore plausibly exists inside the Commission and the national administrations already.
It has not been published in aggregate.
What could actually have been deferred
The instruction's effect is bounded by when it arrived.
European maintenance runs in two seasons. Recent years put the spring peak in April at roughly 1.2 mb/d of capacity offline, with the autumn peak comparable — Kpler and IIR put European downtime at about 200 kb/d at the August 2025 seasonal trough, rising to roughly 850 kb/d in September and 1.2 mb/d in October (a 2025 historical pattern, not a 2026 forecast). Total European capacity is around 12.5 mb/d nominal, with runs peaking above 12 mb/d.
The 2026 spring season was already running when the letter went out. IEA forecasts cited in mid-March had European throughput dropping from 11.3 mb/d in February to 11 mb/d in March on seasonal maintenance. Work in execution on 30 March could not be recalled. Contractors were mobilised, scaffolding was up, units were open.
Turnaround procurement compounds this. Industry planning sequences put scope definition 18 to 24 months ahead of shutdown, long-lead equipment orders 9 to 15 months out, contractor prequalification and bid packages 6 to 9 months out, and mobilisation in the final three months. A programme scheduled for April 2026 had passed every one of those gates by the time the letter was written.
So the deferrable tranche was not the spring season. It was the residue: late-season work not yet mobilised, plus programmes scheduled from roughly June onward whose scope had not frozen.
That is a materially smaller target than "European maintenance."
An estimate, with its assumptions exposed
Take a normal European spring season across March to May at roughly 2.5–2.7 mb/d-months of cumulative downtime, using the 2024–25 seasonal shape as the baseline.
Assumption 1. Work already in execution on 30 March — essentially all of March and much of April — was not deferrable. This is a physical constraint, not a judgement.
Assumption 2. The deferrable fraction was therefore 25–50% of April–May downtime, or roughly 0.5–1.05 mb/d-months of work displaced. This is the load-bearing assumption and the weakest one. It is where criticism should be aimed.
Assumption 3. Displaced work does not all land in the same place. Procurement and contractor constraints mean some slips to autumn 2026 and some to 2027 or later. Assume 50–70% lands in autumn 2026.
That yields roughly 0.25–0.75 mb/d-months added to a normal autumn season of about 2.65 mb/d-months.
Peak-month European downtime would rise from roughly 1.2 mb/d to roughly 1.35–1.6 mb/d. European runs in the peak autumn month would fall to approximately 10.6–10.9 mb/d, against a baseline near 11.2.
In product terms: incremental crude not run versus a normal autumn of roughly 7.5–22 million barrels, of which — at European middle distillate yields — perhaps 3–10 million barrels of diesel and jet equivalent, with the balance in gasoline and other streams.
For scale: the IEA reported 44 million barrels of OECD government reserves drawn in June alone.
The honest conclusion is that this is a real but mid-sized exposure, not a capacity cliff. Under central assumptions the deferred-maintenance effect is roughly a tenth to a quarter of one month's government stock release. Anyone claiming Europe has engineered a fuel crisis for autumn is over-reading the arithmetic.
That conclusion is contingent on Assumption 2. If the deferral fraction was at the top of the range, if it clusters into a single window, and if it coincides with unplanned outages, the picture worsens quickly. If the fraction was near zero — because most spring work was safety-critical, contractually fixed or already underway — the intervention was symbolic.
The entire spread turns on one number nobody has published.
The counter-case, which is strong
The most serious objection is that the Commission's request changed nothing.
Refined product cracks and refining margins reached four-year highs in early July. A full-site turnaround forgoes something on the order of $3 million a day in margin, by one industry estimate. Refiners defer flexible work on margin signals every cycle, without any government involvement. On this reading the letter endorsed a decision the economics had already made, and the resulting backlog is the ordinary backlog refiners manage every year.
There is direct evidence for it. Argus reported on 12 March 2026 — two weeks before the letter — that European refiners were not shortening or delaying their spring turnarounds despite the crack surge, with throughput falling as expected. The commercial pressure to defer was already present, and the observable behaviour was that operators were largely sticking to plan.
This objection should be conceded, not deflected. It probably means the marginal effect of the instruction is smaller than the headline suggests.
Two things survive it.
First, the aggregation problem is unchanged. Whether deferral was driven by margins, by Brussels, or by both, the work still accumulates in later windows, and no one is measuring the accumulation across borders.
Second, the request changes the decision at the margin in a specific way: it removes the reputational and regulatory downside of deferring. A plant manager weighing a slip against inspection-interval risk faces a different calculus when the deferral is not merely commercially attractive but officially encouraged. That shift is real even if it is small, and it is exactly the kind of shift that leaves no trace in any public dataset.
By late June the aggregate picture had bent. Kpler assessed that refiners in the United States, Europe and India had "deferred or minimised" spring maintenance to maximise utilisation, and warned of heavier-than-normal autumn work. So the honest chronology is neither nothing moved nor Brussels shut the continent down: operators were largely holding to plan in mid-March, the Commission asked for deferral on 30–31 March, and by late June the behaviour had tilted toward postponement — with no public plant-by-plant account of what moved, by how long, or when it must return.
Why the backlog does not simply reschedule
The instinctive assumption is that deferred work returns to the calendar at the next available slot. The procurement chain says otherwise.
Turnarounds are not staffed from permanent site headcount. Shell's 2024 Pernis programme mobilised roughly 2,500 additional workers drawn from more than 80 service companies, running close to 700,000 working hours. Neste's 2026 Porvoo turnaround, scheduled across August–October, is planned around 7,500 participants from more than 100 contractor companies. The EIA puts major turnarounds on three-to-five-year cycles requiring one to two years of preparation and 1,500 to 2,000 contractors on a single site. Rompetrol's spring 2026 programme at Petromidia and Vega covered more than 700 jobs, catalyst replacement in six units, checks on around 900 pieces of equipment and inspections across nearly 1,200 pipelines.
Major turnarounds and their associated upgrade programmes can cost several hundred million euros — Neste has put Porvoo near €390–400 million and TotalEnergies' Antwerp turnaround and projects at about €400 million. These are not schedulable at short notice.
The labour pool is the binding constraint. Welders sit among the most widespread shortage occupations in Europe — reported in shortage across at least three-quarters of countries in the European Labour Authority's 2024 EURES assessment — and a UK fuels-sector survey by UKPIA (now Fuels Industry UK) and Cogent Skills found 62% of respondents expected it to become increasingly difficult to recruit people with the right skills. The same specialist firms and travelling crews serve multiple countries. Two refiners can independently postpone into the same autumn; the workforce cannot service both.
The precedents are documented. Germany's Leuna refinery postponed a planned general maintenance and upgrade programme in 2020 on supply-chain and travel grounds. And Bloomberg reported that TotalEnergies' 2025 Antwerp turnaround — the largest in that refinery's history — included work on a major FCC unit that had been postponed from the previous year: a deferral that did not vanish but rolled forward into a later, larger shutdown.
That is the mechanism. A deferred turnaround does not slot neatly into the next window; it competes for a window, and frequently becomes a larger, longer programme when it eventually happens.
Reliability debt
The right frame is a balance sheet entry.
Keeping a unit online produces fuel now. The inspection still has to occur, the catalyst still has to be replaced, the exchanger still has to be opened, the statutory certification still expires on its original date. The obligation has moved from the present period to a future one, and it accrues interest in the form of procurement friction and elevated unplanned-outage hazard in the interim.
The cascade is:
- Conflict disrupts product supply.
- Margins spike; Brussels encourages deferral; the two point the same way.
- Non-emergency maintenance slips.
- Utilisation stays higher than planned, and assets run further into their inspection intervals.
- Procurement and contractor constraints prevent clean rescheduling.
- Deferred programmes cluster, and some become larger than originally scoped.
- Output falls in a later period, against a stock cushion already partly consumed.
Step 4 deserves emphasis because it is the part that operates continuously rather than at a future date. An asset running six to eighteen months past its intended inspection point carries a higher probability of unplanned failure throughout that period. Unplanned outages are worse than planned ones not because the repair is harder but because nothing can be pre-positioned: no inventory build, no booked labour, no coordination with neighbouring plants.
Europe begins this with less redundancy than it had. FuelsEurope — an industry association with an obvious interest in stressing refining's strategic value, though the direction of travel is not disputed — counts 35 European refinery closures since 2009 and roughly 20% of capacity gone. Its statistics show mainstream capacity across the EU, UK, Norway and Switzerland falling from about 638.2 million tonnes in 2024 to 625.4 million at the end of 2025. And industry reporting puts more than 400,000 b/d closed or placed into wind-down during 2025, including Grangemouth, Wesseling, part of Gelsenkirchen and Lindsey.
The stock picture is more ambiguous than it looks
One correction to the pessimistic reading. Global observed inventories actually rose in June for the first time in four months, by 21 million barrels, because oil on water swelled by 117 million barrels — more than offsetting onshore draws of around 96 million, including the 44 million of OECD government reserves.
Floating stock in transit is not equivalent to onshore product cover in the right grade and the right region, and a barrel in the Gulf of Oman does not fuel a truck in Bavaria next week. But the headline is a build, and any argument that rests on a depleting cushion has to account for it.
The refining constraint is the more durable point. Global runs rose 1.5 mb/d in June but remained about 6 mb/d below year-earlier levels, with Gulf export refineries not restarted, Russian throughput cut by attacks and Asian plants below normal rates. The IEA expects global runs to fall 2.4 mb/d across 2026 and rebound 3.1 mb/d in 2027.
That 2027 rebound forecast is worth interrogating. It is unlikely to incorporate a deferred-turnaround backlog that nobody has quantified.
What to watch
This thesis is testable without any Commission disclosure. The observable indicators:
- Operator turnaround announcements shifting quarter. Neste, OMV, MOL, Preem, Rompetrol, TotalEnergies and Repsol all disclose programmes in earnings material. Slippage from spring to autumn, or autumn to 2027, is the direct signal.
- Turnaround services order books and guidance. Bilfinger, Altrad, Petrofac and comparable firms report forward workload. A H2 2026 or 2027 book filling faster than seasonal norm is the contractor-bottleneck hypothesis becoming visible.
- Catalyst supplier lead-time commentary. Ketjen, Topsoe and Albemarle discuss order timing on calls. Lengthening lead times indicate clustered demand.
- European CDU downtime trackers. Commercial estimates of offline capacity by month. If autumn 2026 peak downtime exceeds roughly 1.35 mb/d, the deferral was real and toward the upper end of the range. If it comes in near 1.2, the instruction was largely symbolic.
- Unplanned outage frequency. Trade press reporting on unscheduled European shutdowns through H2 2026. A rise above seasonal norm is the interim-hazard mechanism showing up.
Any of these can falsify the argument. That is the point of listing them.
Four questions for the Commission
Aggregate reporting would not expose a specific vulnerable vessel or create a security risk. Site-level detail should stay unpublished. But the Commission asked member states for capacity data in March, and it is revising the Oil Stocks Directive on the basis that the crisis exposed weaknesses in the existing framework. The directive measures barrels and days of cover — the greater of 90 days of net imports or 61 days of consumption. It does not measure the reserve of operational availability inside the refining system, which is what the March intervention actually drew down.
Four questions:
How much crude distillation and conversion capacity had maintenance deferred, in aggregate?
Into which quarters has that work been rescheduled?
Do the revised programmes overlap, and does projected contractor demand exceed expected availability?
What stock cover is assumed against the production lost when those turnarounds finally run?
The answers may well be reassuring. On the central assumptions above they probably are. But "probably reassuring" is a conclusion Europe should be able to demonstrate rather than presume, and at present nobody outside a handful of national administrations can check.
The absence of a shortage describes the result. It does not price the result.
Stocks were released. Demand was suppressed. Cargoes were bought at four-year-high margins. Refineries ran harder. Maintenance moved.
Each was a form of resilience. Each consumed something. The last of them consumed future availability, in an amount that is estimable but has not been estimated in public.
The maintenance can be delayed. The bill cannot.
This analysis applies the Compound Cascade Systems Modelling Framework. The quantitative envelope above rests on stated assumptions, principally the deferral fraction in Assumption 2; corrections and better estimates are welcome.