Data Center Cargo Theft: What AI Hardware Operators Can Verify Between Factory and Rack

What the reports say

The NBC News investigation by Lisa Cavazuti, published on 26 October 2026, opens with a trailer carrying about $13 million of server equipment toward Fayetteville, Georgia, in early June. According to the Quartz summary of the report, NBC News built its total of more than $100 million from accounts by trade executives, advisers and investigators who help police on the largest cases. The stolen loads include servers, memory chips, bitcoin mining rigs, copper wire and fiber optic cabling.

In the NBC News account, thieves take over freight broker accounts on load boards, post fake shipments and send a malware link to every trucking firm that bids, then use the hijacked carrier accounts to bid on real loads. Sometimes a genuine driver is booked without knowing about the scam; in other cases fake driver's licences get accomplices hired. In one of the largest heists the tracking coordinates were falsified. In two of the biggest, off-duty police officers hired as escorts were hit in hit-and-run incidents, and none of the equipment has been found.

DCD reported the same day on a report from Overhaul. It says more than $150 million of AI equipment has been stolen in 2026, and that single losses regularly exceed $30 million. Reports of stolen data center equipment, compute components and semiconductors are up 34% since 2024, and deception such as false carrier identities, altered paperwork or spoofed messages featured in 92% of the high value technology thefts Overhaul analysed. DCD notes that the report did not disclose the incident count or the shipment volume behind the increase. In one case, $60 million of hardware was redirected to an unauthorised location and recovered after the owner was alerted.

On 14 August 2026, FreightWaves quoted Scott Cornell of SPG Cargo & Logistics, who chairs TAPA Americas. He counted $110.6 million in known reported data center and high tech thefts since 3 June, with single cases between $2 million and $38 million, and he linked both escort ramming cases to carriers that were never properly vetted. All of these totals are industry estimates from different sources and periods, and they overlap.

Why the factory to rack window matters

An operator rarely books the truck. The manufacturer, distributor or integrator does, and custody changes hands several times before a rack reaches the dock. The operator's records usually start at receipt, the first point where it can see that something is missing.

The cost runs past the hardware. Overhaul told DCD that insurance will likely cover the stolen equipment but not the indirect losses from timeline disruption or delayed customer contracts. DCD also cited The Information reporting that data center late fees have reached around $5 million a day. If a shipment of GPU servers is stolen, the cluster it was meant for comes up late.

Paperwork is a weak point. Cornell told FreightWaves that paper bills of lading let thieves replace product descriptions with broader terms. A delivery note that only says servers cannot prove which servers arrived.

What operators can verify on their side

Carrier vetting belongs to the shipper and the broker, and FreightWaves recommends checking drivers and carriers before any load is released. The receiving site can still ask for three things before a shipment leaves: the expected serial numbers per pallet or rack, the name of the carrier that will collect it, and an alert from the logistics provider if the shipment tracker stops reporting or leaves its planned route.

At the dock, count the units, check the seals and scan every serial number against the shipper's list. Create the asset record at that moment, with the time and the person who signed for it. A record made weeks later during installation cannot tell you whether a unit went missing on the road or inside the building.

After that, keep three lists in step: what was shipped, what was received and what was installed. Each handoff, from dock to storage cage to rack, adds a line to the chain of custody. When the lists disagree, the gap is recorded and followed up.

What the record looks like in Sensaka

Sensaka does not track trucks or secure shipments. Its part starts once hardware is in the building. Sensaka for AI Data Centers reads each server's serial number, model, firmware and component health from the BMC over IPMI, Redfish and SNMP, without agents on the host. That gives a second check: the serial recorded at the dock against the serial the hardware itself reports when it is powered on.

The AI infrastructure CMDB holds the asset record. Its reconciliation view compares discovered server and accelerator configuration with that record and shows missing assets, configuration changes and location mismatches. Accelerators are inventoried at component level, for NVIDIA, AMD and other accelerators, so a server that arrives with fewer cards than listed shows up as a difference. Every change keeps the previous and new values, who or what made it, and when, which is the history a chain of custody review needs.

New servers reach the GPU pool through provisioning with a burn-in gate, and the handover registers each one in the CMDB with its rack location before it takes work. A mismatch opens a ticket, and through CMDB and ITSM integration the records and tickets sync with ServiceNow or Jira when those are the system of record.

Takeaway

The 2026 totals are estimates, but every source puts fraud at the centre of these thefts. An operator does not choose the carrier. It does control the serial list it expects, the check at the dock and a CMDB that matches what the hardware reports.

Frequently asked questions