The 2026 International Utility Locate Rodeo runs December 7th and 8th in Springfield, Missouri, and registration is open now. It is free to compete. The prize pool tops out at $2,500 for the “Locate From Hades” winner, with $1,000 per division winner across electric, gas, telecom, and water. If you have not heard of it, that is the point. The International Utility Locate Rodeo is the most honest benchmark in applied perception work right now, and almost nobody in the AI industry is paying attention.

Here is the actual mechanism. Competitors get a set of buried utility lines, a locator rig, and a clock. They mark where the pipes and cables run. Scores in the 2025 results sit in the 990-to-1000 range at the top and the 1980-to-1996 range for division winners, with time used as a tiebreaker. That is a precision-and-latency benchmark. It is the same shape of problem that autonomous vehicle teams, warehouse robotics companies, and infrastructure-inspection startups keep trying to solve with vision models, and the rodeo has been running it in the dirt for years.

The data problem nobody wants to name

Utility locating is a perception task with adversarial ground truth. The ground moves. Frost heaves shift conduits. A contractor backfills a trench and forgets to update the as-built. A gas line installed in 1974 has no digital record. The locator walks out with an electromagnetic receiver and has to reconcile what the signal says against what the drawings claim against what the soil is doing.

Read the 2025 division winners and you see the tell. Jacob Blackwell of PG&E won the electric division. Sean Eckman of Northern Lights and Mark Willers of City Utilities Springfield tied for runner-up on time. Beau Myers won gas with a 1995.5 and Krystal Ramler of UtiliSource matched him at 1995.5, losing only on the tiebreaker. Paul Bess of SCI Engineering and Stefan Diezyn of GTEL tied at 1985.5 in water, again decided by time. These are hundredth-of-a-point separations. Human locators are operating at the edge of what the measurement can resolve.

Now ask the obvious question. Where is the model that beats them?

What the rodeo reveals about AI’s blind spots

The AI industry has spent 2024 through 2026 shipping agents, coding assistants, and multimodal models that can read a PDF and write a contract. It has spent far less effort on the physical-world perception tasks that underpin the built environment. Utility locating sits exactly in that gap. It requires fusing electromagnetic sensor data, historical records, and spatial reasoning under time pressure, in weather, with liability attached.

The 2025 competition ran in 30-to-40-degree temperatures. Competitors came from 18 states and Canada. Seventy-five registered. The event is hosted by the Missouri Common Ground Alliance with the National Utility Locating Contractors Association, and the same week includes the 2026 MOCGA Summit on December 9th and 10th, with free safety and damage prevention classes for competitors. That is a conference and a benchmark in the same room. AI labs run evaluations in clean harnesses. This one runs in December in Missouri.

The prize structure is also a signal. Two wheel winners per division get $100 each. Division runners-up get $500. Division winners get $1,000. The “Locate From Hades” winner gets $2,500, more than double any division. The organizers are explicitly rewarding the hardest case, the one where the records are wrong and the signal is noisy. That is the opposite of how most ML benchmarks are designed, where the tail is trimmed and the median is reported.

The compute angle is real, and it is not GPUs

Every locate ticket in the United States is a data point about where infrastructure actually is versus where the drawings say it is. The 811 system generates millions of these annually. Almost none of it is labeled in a form a model can train on. The rodeo produces a small, high-quality, human-verified slice of that distribution every year, and it is sitting on a website with a countdown timer.

If you are building perception systems for field robotics, inspection drones, or construction, the rodeo is a free source of hard negatives. The “Locate From Hades” category is essentially an adversarial test set with a cash bounty. The 2026 event is in Springfield, and the 2025 program is still posted, including a Christmas toy drive for CASA run by MCGA. The whole thing is volunteer-organized and free to enter.

The hardest perception benchmark in America is decided by hundredths of a point in 35-degree weather, and no AI lab has entered it.

What this means for builders

The lesson is not that AI should replace locators. It is that the data moat in physical infrastructure is not a model problem, it is a labeling and access problem, and the people who own the access are not AI companies. They are utilities, contractors, and volunteer associations like MOCGA and NULCA.

If you want to build in this space, the move is to show up. Register a team. Sponsor a division. Put a sensor rig on the course and see how it scores against Beau Myers and Krystal Ramler. The entry fee is zero and the ground truth is right there.

The 2026 rodeo runs December 7th and 8th. The MOCGA Summit follows on the 9th and 10th. Registration is open on the event site, and the countdown clock is already running. The AI industry talks constantly about grounding models in the real world. Here is a real world with a scoreboard, and it has been waiting.