DJI Wants Developers to Build the Future of Aerial AI. The Prize Is Real.

DJI just opened a competition that could change what drones are capable of. Not the hardware. The brain.

The DJI Enterprise Drone Onboard AI Challenge 2026 is not a drone racing event. It is not a photography contest. It is DJI saying, out loud, to every developer and researcher on the planet: we have built the most capable commercial drone platform in the world, we have strapped a supercomputer to it, and we want to see what you can make it do.

A Supercomputer the Size of a Deck of Cards

To understand why this competition matters, you have to understand the Manifold 3. It is a small box. Embarrassingly small. It measures 3.86 inches by 2.24 inches by 1.42 inches and weighs 4.2 ounces. You could lose it in a jacket pocket.

That pocket-sized box delivers 100 TOPS of onboard computing power. TOPS stands for Trillions of Operations Per Second. The Manifold 3 processes 100 trillion operations every second, using a GPU and a Deep Learning Accelerator running in parallel, while mounted to a Matrice 400 or Matrice 4 Series drone flying an active mission.

It does all of that without meaningfully affecting flight time, because it draws only 25 watts and weighs almost nothing against the aircraft’s payload capacity.

It operates in temperatures from minus 4 to 122 degrees Fahrenheit. It carries an IP55 weather resistance rating.

It connects to the drone’s full sensor ecosystem through PSDK V3, meaning it has access to camera feeds, thermal data, LiDAR point clouds, flight telemetry, and gimbal control simultaneously. Apps install through DJI Pilot 2 like a smartphone, with no command line required.

In plain language: DJI built a flying AI inference engine, and now they want developers to tell them what to do with it.

What It Can Already Do

The competition is not starting from zero. DJI has already demonstrated what onboard AI looks like in practice, and the examples are genuinely impressive.

In South Africa’s Sabi Sand Nature Reserve, AI algorithms deployed on a Matrice 4T are helping rangers fight rhino poaching in real time. The drone combines thermal imaging with long-range zoom and an onboard detection model that identifies threats and alerts the ground reaction team before a poacher can disappear into the bush. At night. Autonomously. Without a human watching every frame.

In river inspection operations, a Matrice 400 paired with a Manifold 3 flies waterways and detects floating debris, invasive aquatic vegetation, and silt accumulation in real time, feeding that data directly into ecological assessment workflows. No post-processing. No waiting for a human analyst to review the footage back at the office. The drone flags the problem while it is still in the air above it.

These are not demos. They are deployed operational systems. The competition is asking what comes next.

What Developers Are Actually Competing For

The DJI Enterprise Drone Onboard AI Challenge opened in February 2026, with submissions running through March. Finalists will be announced in May, public testing and voting run through June, and winners are announced at the end of June.

Fifteen top solutions take home prizes. The five Best Onboard AI Model Award winners receive a Manifold 3 combo and an Osmo Pocket Series combo alongside the trophy. The ten Industry Application Excellence Award winners receive a Manifold 3 combo and a trophy. But the hardware is honestly the smaller part of the prize.

The real value is the platform access that comes with winning. Solutions are featured in DJI Enterprise’s official Onboard AI Solutions Catalog, promoted across DJI’s global enterprise channels, and showcased to DJI’s entire customer base.

Winning teams get fast-track access to the DJI Enterprise Ecosystem partner audit, with priority access to new product beta testing and direct technical support.

For a developer or startup trying to build a commercial AI drone product, that distribution pipeline is worth significantly more than any hardware prize. DJI’s enterprise customer base spans utilities, public safety, agriculture, infrastructure inspection, and environmental monitoring in over 100 countries.

Getting your algorithm featured in their official catalog is the kind of exposure that does not have a price tag.

Judges include directors from Tsinghua University, Wuhan University, and Shenzhen Polytechnic University, alongside specialists in electric power, maritime safety, natural hazards, and forestry big data. This is not a student hackathon. It is an industry evaluation panel reviewing solutions intended for real operational deployment.

DroneXL’s Take

Here’s what I actually think: this competition is DJI making a strategic bet, and it is a smart one.

The hardware race in commercial drones is largely over. DJI won it. The Matrice 4 series, paired with a Manifold 3, is a better aerial intelligence platform than anything else commercially available at any comparable price point. The next frontier is software. Specifically, what kinds of AI can run natively on that hardware while the aircraft is in flight, making decisions without a data connection, without a human in the loop, without waiting for a ground station to process anything.

No sugarcoating this: most of the best ideas for what to do with that computing power are not going to come from inside DJI. They are going to come from a wildlife conservationist in South Africa who needs to track rhinos at night. From a utility inspector in Texas who needs to identify hairline cracks in transmission towers from 200 feet. From a river ecologist in Southeast Asia who needs to map invasive species across 500 miles of waterway in a week.

DJI knows this. The competition is how they find those people.

The drone industry talks constantly about what the technology can do. The Onboard AI Challenge is a rare moment where a major manufacturer is actively asking the world to show them. The answer might protect a rhino. It might save a flood victim. It might do something nobody has thought of yet.

That is worth paying attention to.

Photo credit: DJI