By the end of May 2026, more than 1.1 billion barrels of oil had not reached the market, IMF economists Jean-Marc Natal and Azim Sadikov wrote in a blog on 15 July 2026. That is about 10 days of typical global consumption. At the same stage of the disruption (a stage their blog does not define in months), the shortfall was already larger than in 1973, the Iran-Iraq war, or the Gulf War. They describe a war in the Middle East that effectively closed the Strait of Hormuz, and a later US-Iran framework aimed at reopening it. That account does not give a start date or a full list of combatants.
By early October the official mood was still not a 2008 rescue drill, at least in Larry Elliott’s telling. Writing in the Guardian, he accepts the IMF’s view that countries have shown resilience, then calls the moment a lull before the storm: a crisis postponed, not avoided. He sets the Bangkok meetings of finance ministers and central bank governors against autumn 2008, when those gatherings sat inside a banking meltdown. The meetings themselves were due the week after 7 October 2026.
For anyone signing an AI budget, the split is the part worth keeping. Spending on AI for business does not hedge an oil shock. A tighter fiscal climate will not carry an AI automation programme that only works if energy and money stay cheap.
What actually kept prices from breaking
Natal and Sadikov called this the largest oil-market disruption in decades. The closure cut off about 20 million barrels a day of crude and products, roughly a fifth of global consumption. Saudi flows through the Yanbu pipeline, and UAE shipments through Fujairah, replaced only a fraction. Gulf diesel and jet-fuel output fell hard. The region, they note, supplies about 10 percent of those two products worldwide. That 10 percent is a share of global supply. It does not measure how far output fell. The blog does not give the missing volume.
Crude spiked, then settled at $90 to $100 a barrel, which the authors described as much lower than many had feared. That is not a return to pre-war prices. Higher prices did much of the work of cutting demand. Inventories stopped a worse spike. They did not cancel the shock.
For March through May, they counted three absorbers. None of them was a data-centre buildout.
- Demand. It fell, especially in Asia, including a shift toward coal and renewables. Transport demand was stickier, because price caps, subsidies, and tax rebates dulled the signal, at a fiscal cost.
- Extra production versus 2025. Non-Gulf output ran nearly 2 million barrels a day above 2025 levels, led by the United States, with Venezuela, Guyana, and Russia also higher. Do not add this to the other “2 million.” Before the war, supply was already about 2 million barrels a day above demand. That second figure is a pre-war surplus versus consumption. It is not extra barrels versus last year.
- Stocks. An estimated deficit of about 4.0 million barrels a day was met almost entirely by draws on commercial inventories, including in China, and on strategic reserves.
A US-Iran framework to reopen the strait then sent prices sharply lower, because oil stranded on Gulf tankers could return quickly. Tensions flared again. Industry estimates cited by the IMF authors put the lag at two to three months, even after a full reopening, before a significant share of flows would resume. Spare capacity has been used. Demand has already been compressed. Stocks are nearing operational minimums. Prolonged shutdowns, they warn, can permanently lose output where the money to restart is scarce.
The cushion bought time. It is thinner now. An operator should not conclude that an AI spending boom soaked up the shock, or that the drawdown has been refilled. Weaker demand, higher non-Gulf supply, and stock draws can explain why crude settled at $90 to $100. Folding an AI spending boom into that oil story overstates what the Fund’s oil economists have shown.
One buffer holds. Two are Elliott’s
Elliott’s account overlaps that story in one place. Past oil shocks of this scale, he writes, hit growth and inflation quickly. This time they have not, and he offers three reasons: massive investment in artificial intelligence, the use of oil inventories to offset lost Middle East crude, and China’s export of cheap goods, which he says is holding down import prices elsewhere.
On inventories he lines up with the Fund. Without those drawdowns, prices would have risen more sharply. That is what the barrel math says. China does show up in the oil analysis, as a holder of commercial stocks being run down. It does not show up as a permanent lid on everyone else’s prices.
The other two supports are his argument. They are not in the IMF blog. He says AI investment has been particularly important in the United States, which he calls comfortably the fastest-growing G7 member year on year. He cites no growth rate. He also treats cheap Chinese goods as a threat to manufacturing jobs in the West, with Germany especially exposed. That is a trade judgement. It is not an oil finding.
He is on firmer ground calling this a lull. The IMF economists’ lesson is similar, and less theatrical: the system bought time, and the padding is wearing out. A long winter unless peace arrives quickly is his forecast. Useful as a warning. Not a dated path you can drop into a model.
The squeeze on every AI budget
Growth and debt are pulling in opposite directions. That matters more for AI for business than another demo.
Speaking in Singapore on 7 October 2026, ahead of the Bangkok meetings, IMF managing director Kristalina Georgieva said global debt-to-GDP was the highest since the Second World War and on course to reach 100 percent in coming years, the Guardian reported. Governments could not rely on rapid growth to carry that debt. They would have to make “very tough political choices.” High-debt advanced economies, she said, need credible medium-term fiscal consolidation, “supported in some cases by upfront fiscal measures.” She told policymakers they had the tools and should stop delaying.
The same account said bond yields had jumped to multi-decade highs as markets priced higher inflation from the Middle East war. Elevated yields raise interest bills alongside defence and other spending. She called tightening by the European Central Bank, the Federal Reserve, and the Bank of Japan “highly appropriate, ” noted that the Bank of England had left rates at 3.75 percent, and suggested “a prudently hawkish bias.” She did not lump the Bank of England in with the central banks she said had already tightened.
Elliott hears the push for tighter budgets and predicts a backlash. Voters, in his view, want politicians to do more, not less, and there is little appetite for a new era of austerity. Georgieva is giving institutional advice: growth will not shrink the debt, so consolidate. Elliott is making a political forecast: painful measures will not land cleanly. A finance chief should plan for the instruction to tighten, and for the chance that tightening arrives late, messy, or both.
Britain is a small version of the same bind. The Guardian reported that Chancellor John Healey said he would keep Rachel Reeves’s rules: balance day-to-day spending with tax revenue, borrow only to invest, and lower debt-to-GDP over time. Investment can still be the permitted exception. Day-to-day spend cannot. AI programmes are often a mix of the two. Licences, seats, and change programmes are operating spend. Infrastructure can be capital. The label on the budget line matters more than the word “AI.” A programme booked as operating spend, rather than as investment with a payback you can inspect, sits in the part of the budget finance ministries are being told to shrink.
What officials have actually said about AI
Georgieva’s remarks on AI, as reported from that Singapore speech, are narrower than the boom narrative. AI has buoyed the US stock market. Separate IMF research, as she cited it, says effective adoption could add half a percentage point to global growth. The account does not say whether that half point is a faster growth rate or a one-off lift in the level of output, or over what horizon. It is a possibility attached to effective adoption. It is not evidence that current spending has already delivered the gain, and it is not evidence that AI investment has starved other sectors of capital.
Elliott does make that crowding-out claim. He argues that massive US AI investment has raised borrowing costs and caused shortages of capital outside AI. Government bond yields are clearly higher. The cause he assigns is not what Georgieva documented. Yields moved as markets repriced war-driven inflation risk, on top of debt loads not seen since the Second World War. Attributing the whole yield jump to AI capex gives one sector more power over the macro picture than these figures can prove.
The risk that cuts the other way is easier to audit. If a board approved a multi-year AI build because equity markets were celebrating it, it underwrote a valuation mood. The half point she pointed to arrives only if adoption is effective. In a profit and loss account, effective means shorter cycle times, a lower cost to serve, and revenue work that used to need a larger team. Unused licences are not adoption. A pilot that cannot survive today’s cost of money is not a strategy.
There is a stability warning attached, and it does not belong only to the IT team. On 31 August 2026, Bank of England governor Andrew Bailey warned that frontier AI poses “real and significant” risks and argued for a right to intervene. As documented, that warning was about financial stability. It was not about tax and spending. Firms wiring AI agents into payments, credit, or market operations should assume supervisors want the power to step in.
How to spend while the cushion is thin
Oil in a $90 to $100 range is a tax on moving goods and people, and on inputs that still track fuel. Subsidies hid some of that tax and shifted it onto the same budgets Georgieva wants tightened. The IMF authors’ own lessons (rebuild stocks, diversify routes and energy, and keep consumer support temporary and targeted) are a reminder that the padding was borrowed, not earned.
AI does not insure a company against that. The firms that come through this lull will be the ones that already know which workflows a system has made cheaper, and which only looked cheaper because capital was in a storytelling mood. If the payback cannot be inspected, it is not a strategy. If an AI agent touches payments, credit, or a customer, keep a human who can stop it.
Questions worth answering before the next planning round
Has this oil disruption already exceeded the famous historical shocks?
At a comparable stage, yes, according to IMF economists Jean-Marc Natal and Azim Sadikov. By the end of May 2026 more than 1.1 billion barrels had not reached the market, a larger shortfall than in 1973, the Iran-Iraq war, or the Gulf War. Their 15 July blog does not define that stage in months. They put the cutoff at about 20 million barrels a day of crude and products, roughly a fifth of global consumption, after a Middle East war effectively closed the Strait of Hormuz.
Why did crude settle at $90 to $100 instead of going through the roof?
Not because AI spending absorbed the shock. For March to May, Natal and Sadikov point to weaker demand, non-Gulf output nearly 2 million barrels a day above 2025 levels, and inventory draws that covered a deficit of about 4.0 million barrels a day. Less extreme than feared. Not harmless.
Is the AI boom why the wider economy has not cracked?
Not as an oil absorber. As a growth story, Elliott asserts it, especially for the United States, and does not show a rate. The public official line is narrower. AI has buoyed the US stock market, and effective adoption could add half a percentage point to global growth, with neither the horizon nor a level-versus-rate distinction stated.
Will tighter budgets raise the bar for AI spend?
They are being pushed that way. Georgieva wants consolidation, in some cases upfront, because debt-to-GDP is on course to reach 100 percent in coming years and growth will not carry it. Higher yields already raise the hurdle rate. Elliott’s claim that AI itself caused a capital shortage remains a hypothesis.
What should an operator do while the cushion is this thin?
Fund AI automation that still pays for itself if energy stays expensive and money stays tight, and split operating spend from capital so the payback can be inspected. Keep a human override where an AI agent touches payments, credit, or customers.
Elliott called this a crisis postponed rather than a crisis avoided. On the barrels, that is a fair reading of the Fund’s own oil economists. On AI, the delay runs the other way. The productivity is still mostly a promise. The capital is already being spent. If you are waiting for the macro story to justify the budget, you have the sequence backwards.
The speech that set the frame
Georgieva’s Singapore remarks are the public source for the energy, debt, and AI warning that the rest of this argument turns on.