The coffee market is running two sets of books. On one, the world is about to produce more coffee than it ever has and more than it will drink: the USDA’s July estimate puts 2026/27 world output at 189.7 million bags against 179.7 million of consumption, a surplus of ten million bags. On the other, the inventory that actually backs the world’s benchmark coffee contract has fallen to roughly 242,700 bags — at current consumption, about twelve hours of coffee. Both numbers are true. The gap between them is the story.
The two numbers that do not fit
Brazil is harvesting a record. CONAB’s second survey put the 2026/27 crop at 66.7 million bags, with arabica alone at 45.8 million, up 28% on the year; the USDA is more bullish still at 71.9 million bags in total. Vietnam is forecast at 32.5 million. World green bean exports are projected at a record 131.4 million bags. By every measure of supply, this is an abundant year.
Set against that, the ICE-certified arabica inventory — coffee that has been graded, approved, and placed in exchange-licensed warehouses where it can be delivered against an expiring futures contract — ended July at 274,168 bags against 791,842 at the same stage a year earlier, a fall of 65%. By mid-August it was down to roughly 242,700, the lowest since late 2023. The certified pile has emptied into the largest crop on record.
What certified stock actually measures
It is worth being precise about what this number is and is not. Certified stock is not the world’s coffee. World ending stocks for 2026/27 are estimated at 26.3 million bags; the certified inventory is under one percent of that. It is not a measure of scarcity. It is a measure of how much coffee has chosen to sit in one specific set of warehouses under one specific set of rules.
That distinction usually does not matter. It matters now because the certified pile is the physical settlement mechanism for the contract that prices coffee globally — the reference for roaster hedges, producer forward sales, and every headline about what coffee costs. When the buffer behind that contract thins to half a day of world consumption, the contract stops tracking the world balance and starts tracking the warehouse.
Futures are pricing the emptiest warehouse in a market that is in surplus. Those are two different questions, and in 2026 they have come apart.
Why the record crop has not reached the warehouse
Four mechanisms are keeping the crop away from the exchange. None of them is scarcity.
Timing. A forecast is not deliverable inventory. Coffee has to be picked, dried, hulled, graded, shipped, and approved before it can be tendered. Brazil’s harvest has run late: Safras & Mercado had it 64% complete in mid-July against 77% a year earlier, and Cooxupé’s members at 58.3% by 24 July against 67.1%. Rain did it — Minas Gerais took 31.3mm in the week to 28 June, roughly 1,956% of the historical average for that week. The crop is real. It is simply behind.
Producer retention. Commercialisation has run slower than the volume implies. Growers face operating costs up an estimated 12–16% and a deteriorated coffee-to-fertiliser exchange ratio, while a real near 5.08 to the dollar trims the local-currency value of a dollar-priced sale. Trade commentary in mid-August described Brazilian growers as still holding a great deal of unsold coffee and in no hurry to move it short of new highs. A farmer who can afford to carry the crop has reason to. This is the mechanism our fifth note in this series described from the other direction: monetary conditions in Brasília setting the pace of physical supply to an exchange in New York.
Carry economics. The curve is steeply inverted. On 19 August the September–December spread traded around 31 cents in favour of the nearby contract — spot far above deferred. In backwardation of that scale nobody is paid to store, because storing means selling the cheaper forward month. Certified coffee also accrues age-related penalties, and physical differentials have generally paid better than exchange delivery. Certification is what you do with coffee you cannot otherwise sell; when the cash market is bidding, the exchange is the last stop, not the first.
Geography. The certified pile is now largely Honduran and Ugandan rather than Brazilian. Trade flows were also rerouted by the 2025 tariff episode: the additional 40% US levy imposed on Brazilian goods on 6 August 2025 covered coffee until the food and agriculture exemption signed on 20 November 2025, retroactive to 13 November, and the Section 301 action effective 22 July 2026 again exempts green and instant coffee. The tariff is gone. The stock never came back.
The price of a thin buffer
The consequence is not a higher price. It is a less meaningful one.
In June the ICO composite indicator averaged 248.90 cents per pound, but travelled from 231.96 cents on 9 June — close to a two-year low — to 272.39 cents by month-end, a 17.4% move inside four weeks. Arabica reached the mid-350s in early July, 324.55 cents on 27 July, and 314.30 cents on the December contract in the week to 17 August. That is not the signature of a market discovering a shortage. It is the signature of a market in which a modest delivery demand meets a buffer that cannot absorb it.
Colombia made the point. A magnitude 7.4 earthquake closed Buenaventura, which handles 60–70% of Colombian coffee exports, with operators putting recovery at roughly fifteen days. In a market carrying normal visible cover, that is a differential story confined to one origin. In this one, it moved the world price.
Osiria’s read: enough coffee, in the wrong place
Ask the two questions separately. Is there enough coffee? Yes — a ten million bag surplus and record exports say so, and if Brazil’s crop arrives as forecast the physical market should loosen through the 2026/27 season. Is there enough coffee where the contract requires it to be? No, and the mechanisms keeping it away are commercial and monetary rather than agronomic.
For hedgers that is the operative risk. When futures price warehouse availability and physicals price the world balance, basis widens and the hedge does less work. Roasters should expect the C contract to over-react in both directions and should not read a squeeze as a shortage. Producers should recognise that the tax here is volatility, not level. And anyone using commodity prices as a macro signal should treat this as a general warning: in any market where a large paper contract settles against a small certified inventory — cocoa is the nearest cousin — price can decouple from the balance sheet long before anything physical is actually wrong.
The resolution is mechanical rather than dramatic. As the front month rolls and the late Brazilian crop finishes moving through drying, grading, and shipping, the nearby premium should decay and certified stocks should rebuild; parts of the trade were already looking for arabica below $3.10 on that basis. The squeeze ends when the coffee arrives, not when the world finds more of it.
One caveat runs the other way. Both the Japan Meteorological Agency and NOAA carried roughly 67% confidence in a strong El Niño, which is a risk to the 2027 crop rather than this one. With visible cover this thin, the market has very little room to be wrong twice.