The 34 Per Cent Is Not Restocking. It Is a Bet on the Oil Leg

A 34 per cent rise in imports is the kind of figure that gets filed under restocking and forgotten. Colombia imported 34 per cent more soybean grain in 2025 than in 2024 — big enough that most readings stop at the explanation which costs nothing to check: buyers had run inventories down, beans were cheap, and a country that had been buying hand to mouth decided to hold more cover.

Restocking is what a 34 per cent year usually is. Here it is the wrong answer, and the difference decides whether Colombia needs crush plants.

The tell is price. A restocking buyer adds tonnes because supply is long, so volume rises while realisation falls. Colombia’s pool did the opposite. Pheonix puts 2025 at 1,420 thousand tonnes and USD 632 million — USD 445 a tonne — and models 2033 at 2,760 thousand tonnes and USD 1.31 billion, or USD 475 a tonne. Split the resulting 9.5 per cent value CAGR and it is 8.7 points of volume and 0.8 of price. Realisation rose 6.6 per cent across eight years while the tonnage nearly doubled.

Volume that grows into a firming price is not a restock. It is a different purchase. The full model, the segmentation cuts and the scenario band are in the new report, Colombia Soybean Grain Market Size, Share & Forecast 2026–2033.

What changed is which commodity arrives

Colombia is moving from imported soybean meal to imported whole beans and taking the oil leg at home. The logic is not exotic. Whole beans carry the oil; meal does not; and the oil leg is where the crush margin has been made since biofuel demand reset the economics of the soybean complex. A crusher in the Cauca valley and a crusher in Illinois watch the same ratio, but only the Colombian one is choosing between two import forms rather than one.

That is what makes this forecast falsifiable instead of merely directional, because the leg can be priced. At a crush yield of 78 per cent meal, 18 per cent crude oil and 4 per cent hulls and loss, oil’s share of crush revenue is 18.8 per cent when soybean oil and meal trade at parity, 31.6 per cent at two to one, 36.6 per cent at two and a half, and 40.9 per cent at three to one.

Read that ladder backwards and the thesis becomes modest in a useful way. The shift from meal to beans does not need oil to be unusually expensive. It needs the oil leg to keep covering the extra freight, working capital and processing cost of importing a bean rather than a meal. Above roughly two to one, the arithmetic leaves room. Below parity, imported meal wins again on working capital and the 34 per cent reverses — with no change whatsoever in Colombian demand for chicken, pork or cooking oil. The tonnage is a function of a ratio set in Chicago, Rotterdam and Jakarta, not in Bogotá.

A fifth of the pool owns the asset that makes the shift pay

Four cuts describe the pool, and all shares are Pheonix estimates against the 1,420 thousand tonnes.

By end use, poultry feed takes 46 per cent — 653.2 thousand tonnes — with swine at 21 per cent (298.2), edible oils and food-industrial refining at 17 per cent (241.4), dairy and beef at 9 per cent (127.8) and aquafeed at 7 per cent (99.4). Protein demand dominates, which is exactly why the shift is surprising: the biggest buyers are not the ones who benefit from it.

By buyer, integrated protein producers take 38 per cent (539.6 thousand tonnes), independent mills 31 per cent (440.2), crushers and refiners 21 per cent (298.2) and traders the remaining 10 per cent (142.0). Only the crushers and refiners are structurally long the oil leg. Everyone else is buying protein and would rather not carry a bean they cannot extract from. That asymmetry — a fifth of the pool owning the asset that makes the shift pay — explains why toll-crush arrangements and least-cost formulation decide more in this market than any supplier’s brand. It also explains why the import mix can change quickly: the decision sits with a small number of balance sheets.

Origin and port then set the logistics frame. The United States holds 34 per cent of volume, Brazil 29, Argentina 17, Paraguay 13, Bolivia and others 7. Buenaventura handles 41 per cent, Cartagena 23, Barranquilla 19, Santa Marta 9, and river and inland routes 8. Origin share is a policy variable before it is a commercial one. Deforestation-free sourcing requirements on South American soy are the live test: if compliance raises the delivered cost of Brazilian and Paraguayan beans relative to US beans by more than the oil leg is worth, the substitution reverses without a single tonne of demand changing.

Capacity arrives in lumps, and that is the whole risk

The forecast’s fault line is not demand. It is crush capacity, and capacity does not arrive smoothly.

One two-thousand-tonne-a-day plant is 660 thousand tonnes a year: 46 per cent of the entire 2025 pool and 24 per cent of the 2033 one. The whole 2033 pool is only 4.2 plants at that scale. So three or four separate commitments — each taken by a different board, none of which sees the others’ business case, all of them reading the same 8.7 per cent headline — would be enough to strand the lot. Utilisation, not technology, is the risk in this market.

The report prices both ends. At the parent feed industry’s own growth rate, with no further shift from meal to beans, the pool still reaches USD 864.9 million by 2033 — a real business, but not one that justifies extraction capacity at the pace the largest processors are currently contemplating. At an accelerated build-out it reaches USD 1,621.6 million. The USD 757 million between those two numbers is decided by refinery and crush utilisation in the Cauca valley and around Cartagena, and by whether port and silo capacity at Buenaventura clears fast enough to keep imported beans cheaper than imported meal. That is a logistics question wearing a market-sizing label.

Test any capacity decision against the increment rather than the headline. Value created across the window is USD 678 million for 1,340 thousand tonnes added — USD 506 per tonne of new volume. That is not a margin. It is close to the landed value of a tonne of beans at USD 445 plus the price gain, which means an added tonne largely pays for itself in pool value. What a processor captures is the crush spread on that tonne, not the tonne. Confusing the two is how a business case gets signed for a plant the pool cannot fill.

Which numbers to interrogate

Ranked by leverage, one percentage point off the growth rate is USD 96.2 million of 2033 value: 632 times 1.085 to the eighth is 1,214 against the modelled 1,310. A 2025 pool 15 per cent lower than modelled — 1,200 rather than 1,420 thousand tonnes — is USD 202.9 million. One point of tariff is USD 6.3 million, and so is one point of storage shrink. Plus or minus 1 per cent on either volume or price at 2033 is USD 13.1 million.

Price is the least consequential variable in the model. Volume and the base year are the ones to interrogate, and the base year is the weaker of the two: nothing in this forecast rests on a published Colombian customs tonnage, because none was reachable. The 34 per cent is the anchor. The absolute level is a Pheonix estimate built bottom-up, labelled as such on the page, and re-basing it against a DANE or DIAN figure would move the level without touching the mechanism.

Two things would kill the thesis outright. If imported bean volumes fall back toward 1,060 thousand tonnes while meal imports hold, the share shift is dead and this is a feed-growth story at 4 per cent, not 8.7. And if origin regulation breaks the oil leg’s cover, the mix reverts on its own. Neither shows up in an annual tonnage figure, which is why the report argues for watching them quarterly.

The wider read-across sits in the adjacent Pheonix work: the regional feed pool in Latin America Animal Feed 2026–2033, the global frame in Global Animal Feed 2026–2033, the competing oilseed in Global Sunflower Seeds 2026–2033, and the inclusion-rate side of the formulation decision in Global Feed Phosphates 2026–2033.

Colombia’s 34 per cent is a good number to have on a chart. It is a better one to argue with.