More than $100 million of data center equipment has been stolen in the US since June, according to NBC News reporting summarized by Gigazine, and the two largest heists share a template.
The first: a trailer carrying about $13 million of server equipment, with a security vehicle following it. Another vehicle rear-ended the escort, stopping it. The trailer was taken and driven away.
The second, about three weeks later: a trailer with roughly $10 million of server rack components. Again an escort, again rear-ended and disabled. This time the trailer was being tracked by GPS — and the coordinates were falsified. Its location became unknown.
The cyber part and the physical part are not separate attacks. They're one operation.
The recon is a phishing problem
Thieves aren't cruising industrial parks looking for unattended trailers. They're compromising freight broker and shipping company accounts through phishing, reading shipping schedules, and impersonating legitimate carriers to win the transport contracts for high-value cargo.
By the time a truck is loaded, the attackers may already know the route, the escort arrangement, and the tracking setup. Theft becomes an appointment.
That's a supply-chain attack in the original sense: compromise the coordination layer, not the target. Freight brokers are small businesses with email, spreadsheets and a portal — exactly the profile that phishing works on — and they sit upstream of every shipment.
The economics are the story
Verisk CargoNet's numbers are the cleanest way to see what's happening. Total cargo theft incidents are projected to fall 26% in Q2 2026 against the same quarter last year. The value of stolen goods is projected to rise by more than $100 million — more than double.
Fewer incidents, much more value. That's not a crime wave; it's a reallocation. Thieves are optimizing for value density, which is exactly what the AI buildout has created: GPUs, servers, RAM, fiber, copper. Keith Lewis of CargoNet put it plainly: "It's like stealing a bar of pure gold. It's far easier to steal this than to rob a bank."
Value density changes the security calculus. A trailer of consumer goods needs volume to be worth the risk; a pallet of accelerators doesn't. And unlike cash, the goods are identifiable but not registered — there's no serial-number database that a buyer checks before wiring funds.
The demand side is the uncomfortable part
Experts quoted in the coverage say AI hardware rarely shows up on consumer marketplaces. In some cases buyers are secured before the theft; in many cases the equipment leaves the country, with China named as a destination.
That inverts the usual theft story. Stolen consumer goods are hard to move because the market is visible. Stolen infrastructure has a B2B market where provenance is weakly checked, demand is urgent, and the buyers may not care. The thieves aren't creating demand; they're servicing it.
The defenses are boring
Scott Cornell of the Transportation Assets Protection Association says the thefts are preventable if transport companies thoroughly check drivers' backgrounds. That sounds anticlimactic next to GPS spoofing, and it's probably correct: the attack needs an insider or an impersonator to get the load, and identity verification is the cheap control.
The other controls are equally unglamorous:
- Authenticated tracking. GPS coordinates are an unauthenticated claim from a device that can be spoofed. A signed position report, or a second independent channel, raises the cost.
- Chain of custody. Who held the trailer, when, and where it was handed over. Paperwork is a security control.
- Verified carrier identity. Not just a phone call and an MC number, but something that resists impersonation.
- Treat logistics data as sensitive. The schedule is the reconnaissance. A leaked route is a plan.
My take
This is what the physical buildout of AI looks like from the logistics side: a gold rush with gold-rush security. The interesting failure isn't that thieves are clever. It's that the coordination layer — brokers, carriers, tracking — was built for a world where the cargo was worth less than the fraud required.
The cyber-physical convergence is the part to internalize. Security teams tend to model digital threats and physical threats separately, with different owners. A stolen trailer located by a spoofed GPS is a single incident that crosses both, and it will be handled badly by any org that has never rehearsed it.
The value-density shift also has a longer tail. As long as AI hardware is scarce and expensive, the arbitrage exists. The defense that scales isn't guarding every truck; it's making the goods harder to resell — provenance, serialization, and buyers who check.
Sources: The AI boom has led to a surge in thefts of data center equipment — GIGAZINE, Highway thieves are spoofing GPS to steal millions in data center-bound cargo — Ars Technica.