More than $150 million of AI equipment has been stolen this year.
According to cargo-risk firm Overhaul, thieves have taken more than $150 million of AI equipment in 2026. Deception featured in 92% of thefts of high-value technology shipments, and individual losses now regularly exceed $30 million. Overhaul calls it a shift “from force to fraud”.
$150M+
AI equipment stolen in 2026
92%
High-value tech thefts involving deception
$30M+
Individual losses, now regularly
Source: Overhaul, as reported by Land Line and DCD, October 2026.
The Nature of the Threat
The methods Overhaul lists are fraudulent carrier identities, double-brokering, spoofed communications and falsified paperwork. In each case the attacker forges the surface form, a name, an email, a document, and the process accepts the form as the fact behind it. The load is handed over willingly.
This is a semantic attack: the system checks that a message looks right, not that it means what it claims.
The Defence: Check What the Form Cannot Fake
- Verify the carrier, not the paperwork. Confirm who is collecting through an independent channel you already trust, never through the contact details in the email that arranged the pickup.
- Know the cargo by serial number. A manifest that lists every GPU and server serial, checked at departure and at arrival, turns “a truck of servers” into specific, traceable assets.
- Make stolen hardware hard to sell. A GPU with a recorded serial and health history is an asset with a provenance. One that appears on the market without that history should raise questions, just as a car without a service record does.
The third point is where this connects to our work on GPU health records: the same history that tells a buyer what a card has been through also tells them where it has been.
A shipment should be trusted for who verifiably collects it and which serials are inside, not for what its paperwork claims.