On 13 July 2026, northwest of El Obeid in central Sudan, a Turkish-built Bayraktar Akinci operated by the Sudanese Armed Forces intercepted and destroyed a Chinese-built CH-95 operated by the Rapid Support Forces. The Chinese aircraft had reportedly been supplied by the United Arab Emirates.
Two export platforms, from two manufacturers, on two continents, shot at each other over a third, flown by proxies for a war neither manufacturer is party to. It was described as the first air-to-air interception of its kind, and whatever the eventual verdict on that specific claim, the shape of the event is the finding: two export products, competing commercially, meeting in combat as somebody else’s equipment.
That engagement is the end point of a specific commercial development, and it is worth stating what makes it different from everything preceding it. For eighty years this industry existed because somebody had to physically fly crates to a place, and the crates contained things that other people would then use. The drone collapses that relationship. The cargo is the air force. It arrives in a container, gets assembled by a technician, and then flies itself.
The drone bazaar is what happens when the payload acquires an engine.
The word bazaar is doing precise work. This is not a single market with a single structure. It is a range of stalls selling wildly different things at wildly different prices, from twenty-million-dollar platforms with technology transfer agreements down to five-hundred-dollar quadcopters bought online, and a customer walks along it buying whatever the budget and the threat environment justify.
What the drone market changed about supply
The manned-aircraft market this subject has described was a liquidation. A superpower collapsed, several hundred freighters became available at distress prices from states that needed currency, and the crews came with them. It was a one-time event with a finite stock and no replenishment.
The drone market is the opposite in every respect. It is an active manufacturing sector, with competing vendors, product roadmaps, marketing budgets, trade show appearances, and export finance. Baykar recorded 2.2 billion dollars in exports across thirty-seven buyer countries in 2025, with eighty-eight percent of revenue from foreign sales, and reports export agreements with thirty-nine countries, thirty-six for the TB2 and sixteen for the Akinci. Turkey’s defence exports overall reached 7.1 billion dollars in 2024.
China delivered more than two hundred combat drones to seventeen countries between 2013 and 2023 according to SIPRI, with the UAE, Saudi Arabia, Egypt and Pakistan accounting for roughly seventy percent of the total.
SIPRI’s assessment as of March 2026 is that the United States, Israel, China and Turkey collectively hold over ninety percent of the global military UAV export market.
Notice what that means for supply. An Il-76 cannot be replaced because the production line closed and the fleet ages. A combat drone can be reordered, and the manufacturer would like you to. That single difference changes everything about how the market behaves under pressure. A finite stock of ageing freighters shrinks when one is lost. A production line responds to attrition by building more, which means combat losses are a demand signal rather than a constraint.
The control regime that built the market
The most consequential fact about armed drone proliferation is that it was substantially produced by the attempt to prevent it.
The Missile Technology Control Regime, established in 1987, treats systems capable of delivering a 500 kilogram payload to 300 kilometres as Category I, subject to a strong presumption of denial. Large armed drones fall within that definition, and the United States applied the guideline to itself strictly for two decades, declining to sell Predator and Reaper class aircraft to most customers outside a narrow circle of allies.
The demand did not disappear. It relocated.
Saudi Arabia and the UAE wanted armed drones in the 2010s, could not buy American, and bought Chinese. China is not an MTCR member, having applied in 2004 without being admitted, and its manufacturers sold capable platforms at prices well below Western equivalents with fewer conditions attached. Turkey, which joined the regime in 1997, developed the TB2 around the thresholds and marketed aggressively to customers the Americans would not serve.
By 2021 Turkey had overtaken China as the largest supplier. In 2022, six new countries acquired military drones, and all six bought Bayraktar TB2s.
So a control regime designed to restrict a capability instead determined which countries would supply it, and the answer was the countries least bound by the regime. That is the Rhodesian finding restated in a different technology: prohibition does not eliminate a trade, it selects for whoever is willing to conduct it, and the selection favours parties with fewer scruples and less to lose. The American position eventually shifted, with successive administrations reinterpreting the guidelines to permit more export, but by then the customers had suppliers and the suppliers had reference customers. A market entered late is a market already held.
Africa, and the drone proliferation rate
The proliferation numbers for one continent make the diffusion legible.
Between 1995 and 2019, Africa recorded an average of roughly two military drone transfers a year. Between January 2020 and September 2023, that average rose to nearly thirteen a year. Of eighty-four recorded transfers to the continent in the dataset, fifty-one occurred after 2020, a substantial number of them armed TB2s.
A sixfold increase in transfer rate inside four years is not a trend. It is a phase change, and it corresponds to the moment when armed drones became purchasable by states with modest budgets from suppliers who did not ask what they were for.
What that buys a government is specific. A TB2 gives a state with no air force and no pilot training pipeline a persistent armed reconnaissance capability for a fraction of the cost of a manned combat aircraft, operable by personnel trained in months, with no risk of a captured pilot. For a government facing an insurgency across terrain it cannot patrol, that is transformative, and it is transformative in the way the pickup truck was rather than in the way a fighter squadron would be. Cheap, available, operable by people you already have, and effective against an opponent who has no answer to it until they buy the same thing.
The razor and the blades
The commercial model deserves attention because it is completely unlike the used-freighter trade and it explains the supplier behaviour.
Selling a drone is the beginning of the relationship rather than the end. Turkey has paired sales with electronic warfare packages, counter-drone systems, and training, and has built interoperability requirements into the platforms that keep the buyer returning. Saudi Arabia’s Akinci order came with technology transfer and joint production, and Saudi crews completed training in Turkey in October 2025. The UAE’s Edge Group integrated its own Desert Sting guided bomb onto the TB2 airframe.
That is a manufacturer cultivating a customer over a decade. Munitions, sensors, spares, software updates, and training are recurring revenue, and the airframe is the entry point.
It also creates a dependency that the seller can use. A government whose air capability runs on one supplier’s platform, munitions and support has bought something more entangling than an aircraft. The access relationships described earlier in this subject work in the same direction: what looks like a purchase is the beginning of a leverage arrangement.
The used-Ilyushin market never produced anything like this. You bought an airframe from a liquidating state and never spoke to them again, which is why that market produced no leverage for anybody and this one produces a great deal.
The other bazaar, which has no vendor
Underneath the export market sits a second layer with completely different economics, and it is the one that has made this technology universal.
Commercial quadcopters cost a few hundred dollars, are sold everywhere, and can carry a mortar round. First-person-view racing drones cost less and fly faster. In Sudan, both sides use them: the RSF has modified off-the-shelf drones to drop mortar shells, and FPV and loitering munitions have been used heavily around Khartoum, El Fasher and elsewhere.
Export control does not function at this layer. The airframe is a consumer product. The flight controller is a hobbyist component. The motors, batteries, cameras and radio links are mass-produced electronics with a hundred legitimate uses, sold on open marketplaces, shipped as parcels. There is no chokepoint because there is no specialised item.
That is genuinely new. Every previous capability in this subject required something scarce: an airframe, a crew, a registry, a compliant ramp. A weaponised commercial drone requires a credit card and somebody who can watch a video. The component-level dual-use problem is the hardest export control question of the decade precisely because the answer would require restricting products with overwhelmingly civilian markets.
The consequence, as one analyst put it about the RSF, is that a decentralised force with external supply options has shorter approval chains and a higher appetite for improvisation than a conventional military. The technology suits irregular forces better than it suits the armies trying to contain them, which inverts the historical relationship between air power and insurgency. Air superiority was the one advantage a state could rely on against an irregular opponent, and the cheapest layer of this market has taken it away.
Sudan, where both layers meet
The war that this investigation ends at is the clearest available demonstration of the whole market operating simultaneously.
At the top layer, the RSF operates Chinese platforms. Yale’s Humanitarian Research Lab identified three aircraft assessed as FH-95s at Nyala airport from satellite imagery between December 2024 and January 2025, and noted that the RSF built three drone hangars there over roughly five weeks in early 2025. Reuters separately reported CH-95s at the same airport in May 2025.
In the same period, imagery showed thirteen delta-wing loitering munitions with launching gear north of Nyala airport on 6 May 2025, of a type with a range around 2,000 kilometres, assessed as one of two possible Chinese models, with comparable systems produced in Russia and Iran. China’s foreign ministry denied knowledge.
That range matters because it converts a regional conflict into a national one. On 4 May 2025 the RSF struck Port Sudan on the Red Sea, roughly 1,600 kilometres from its Darfur strongholds, with eleven drones against the Osman Digna air base. The Sudanese army had been storing its Bayraktars at Port Sudan and moved them underground.
Yale’s director has described at least eighty-six drone launchers identified across two locations around Nyala.
At the lower layer, both sides fly modified commercial quadcopters, and the casualty figures are substantial. Reported totals put RSF drone attacks at over 780 killed and army strikes at more than 1,800, with individual incidents including a December 2025 attack on a kindergarten and hospital in Kalogi that reportedly killed over a hundred people, mostly children.
And the supply arrives the old way. Nyala airport reopened under RSF control in September 2024 and has since handled near-daily flights from bases in Chad, Libya and Somaliland, which local sources describe as operating under Emirati oversight. Strikes on Nyala have repeatedly followed cargo arrivals within hours, which is the clearest available demonstration that both sides are watching the same flight activity and drawing the same conclusions from it. Yale identified forty-three shipping containers at the site between mid-December 2024 and mid-January 2025.
Containers, delivered by freighters on the corridor this subject has been tracing, containing aircraft that assemble on site and then fly themselves.
What the drone does to the logistics problem
The structural change is worth isolating from the weaponry, because it alters the economics of the entire industry.
A conventional air capability requires a runway, a maintenance organisation, a fuel supply, trained pilots, and an aircraft too expensive to lose. That is why deniable air power historically meant chartering somebody else’s freighter rather than owning an air force.
A drone capability requires a container, a technician, a generator and a clear patch of ground. Loitering munitions in particular need a launch rail rather than a runway. The aircraft is expendable by design, which removes the entire problem of pilot recovery, hull insurance, and attribution through a downed crew.
That last point is the one that matters most for this subject. The single most reliable way that deniable air operations have been exposed for eighty years is a crash: a location, a date, a hull number, a body. Pope over Indonesia. McGovern at Dien Bien Phu. Every panel of experts report that begins with wreckage.
A crashed drone produces a debris field and a component analysis. It produces no prisoner, no testimony, no nationality, and no funeral. Attribution becomes a forensic exercise on manufacturing marks rather than an interrogation, conducted by whoever reaches the debris first, and manufacturing marks can be removed. The tracking apparatus that grew up around this trade adapted by moving up a level, from identifying aircraft to identifying components, which works and is slower and requires physical access to wreckage.
The industry spent eight decades building corporate and jurisdictional structures to solve the attribution problem. The drone solves a large part of it in hardware.
What has not changed
The continuities are as instructive as the break, and there are three.
The platforms still arrive by air freight, on the same corridor, through the same airfields, in the same containers, flown by the same kind of operator. A drone is cargo before it is an aircraft, and cargo has to be delivered.
The paperwork is unchanged. The company that signs for the shipment, the registry on the freighter, the end-user documentation and the transit arrangements are all exactly as described in earlier lectures. A container of disassembled aircraft moves under the same architecture as a container of ammunition, and the free zone entity that signs for it is as disposable as it ever was.
And the denial is unchanged in form. China’s foreign ministry stated it had no knowledge of the drones and that it adopts a prudent and responsible attitude in military exports. The UAE denies supplying the RSF. Those are the same category of statement as the Emirati and Libyan denials elsewhere in this trade, and they are made in the same knowledge that establishing otherwise requires cooperation across jurisdictions that will not be forthcoming. Nothing about the drone changed the enforcement arithmetic, which was never about evidence.
The technicians, who are the new crew
One element of the old model survives in modified form, and it is the one nobody photographs.
A container of disassembled aircraft is not a capability. Somebody has to unpack it, assemble the airframe, install and calibrate the payload, configure the ground control station, establish the datalink, plan the mission, and operate the system. For the larger platforms that is a team with specific training on specific equipment, and it is not training a militia acquires by watching videos.
Reporting on the May 2025 Port Sudan strike indicated that foreign technicians assisted in assembling the drones at a base near the Libyan border. That is a claim worth flagging as reported rather than established, and it fits a pattern visible across this whole trade.
Because it is the same arrangement as the crew market that staffed the freighters. A capability transfer to a non-state client is incomplete without people, the people are specialists, they work on contract, they are nationals of somewhere other than where they are working, and their presence is denied by everybody including them.
What has changed is the risk profile. A pilot who crashes is captured. A technician who assembles a drone at a base four hundred kilometres from the target is not in the aircraft when it arrives, which means the personnel exposure that has defined this industry for eighty years has been relocated away from the point of delivery.
The operator can be somewhere else entirely, on a laptop, and increasingly is.
The claims that do not hold up
An audit, since drone proliferation generates more confident futurism than almost any other military subject.
Drones have made conventional air power obsolete is contradicted by the July 2026 interception, in which a large crewed-by-nobody platform was destroyed by another one, and by the fact that both Sudanese parties are still trying to acquire manned aircraft where they can.
Anybody can build a drone army conflates the two layers. Weaponised quadcopters are trivially available; a 2,000 kilometre loitering munition delivered in containers with foreign technicians to assemble it is a state-to-state transfer.
Export controls have failed entirely overstates it. Controls determined who supplies the market, which is a real effect, and they continue to shape what the most capable platforms cost and who can obtain them.
Chinese and Turkish suppliers are indiscriminate understates commercial and political calculation on both sides, which includes choosing clients with an eye to reputational and strategic consequences.
Drone warfare has no logistics tail is the version of this that matters most and it is wrong. Drone warfare is cheap is true per unit and false per campaign, once launchers, ground stations, munitions, communications, training and attrition replacement are counted.
Attribution is impossible is too strong. Component analysis, serial recovery and procurement tracing all work, and the imagery and tracking apparatus has repeatedly identified platforms at specific airfields on specific dates.
The RSF built this capability itself is contradicted by the type of platform, the assembly requirements, and the container deliveries.
What the bazaar is actually telling us
The drone completes an arc this subject has been describing since 1950, and the shape of the completion is not what anybody intended.
The original problem was that a state wanting deniable air logistics had to find someone to fly the aircraft. The Agency solved it by buying an airline. Rhodesia solved it by renting a flag. The post-Soviet market solved it by hiring men whose state had vanished. Each solution reduced the sponsor’s exposure while leaving a human being in the cockpit, and that human being was always the weak point: capturable, identifiable, and eventually a witness.
The drone removes the cockpit. What remains is a supply chain, a container, a technician and a launch rail, and the only attributable object is a serial number on a component that may have been manufactured by a company with no knowledge of where it ended up.
Which means the industry this subject has been describing does not end. It becomes less visible, because the most exposed element has been engineered out, and more widely distributed, because the barrier to entry has collapsed from a freighter to a shipping container, and at the lower layer from a shipping container to a parcel.
The investigation terminates at Nyala for good reason. The airport is where the two halves meet: an old freighter on a rented registry lands with containers, and what comes out of the containers takes off by itself and flies a thousand kilometres to somewhere the freighter could never have gone. The old industry delivers the new one, and then the new one does work the old one was never capable of.
A Turkish drone shot down a Chinese drone over Sudan in July 2026. Neither manufacturer is at war. Neither operator built what it was flying. The aircraft were both cargo before they were weapons, and both of them arrived on the same kind of aeroplane that has been landing in that desert since 1987.