At Data Center World Europe this October in Vienna, I sat down with David Warwick, Executive Vice President of Enterprise at Overhaul and former supply chain executive at Microsoft, to cover a topic that doesn’t get enough airtime at data center conferences: data center equipment theft.

With so much value in the equipment supply chain, how well do we understand the theft risk, especially from organised crime?

Warwick brings more than three decades of supply chain experience to the table, including years developing and optimizing Microsoft’s supply chain infrastructure. Recently he’s been spending his time embedded in AI infrastructure risk intelligence.

Understanding Risks in the AI Gold Rush

Warwick started our conversation by explaining that the economics of a gold rush like today’s AI infrastructure buildout are defined by one thing: supply can’t keep pace with demand, so whoever moves fastest and secures the most resources wins.

“In AI, the gold is compute. GPUs above all, with memory, switch gear, and networking components” as smaller nuggets, he said. “Hyperscalers and neocloud providers are the miners, and the data centers themselves are the gold banks.” The data center facilities themselves are extremely well defended, but what about the period before those golden components arrive in the building?

Warwick described the supply chain connecting manufacturers to data centers as “the Wild West, an unprotected dirt road leading up to the data center with no protection around it at all.” This dirt road transports one of the most valuable supply chains in modern history.

To give a sense of scale, Warwick pointed out that the U.S. interstate highway system, one of the largest infrastructure projects ever undertaken, took 37 years and $600 billion to build. The current AI infrastructure buildout surpassed that spending level in roughly six years. Much of this value is in physical form, meaning computing equipment moving around the world in cargo.

Understanding Organised Crime’s Role

I wanted to understand what a modern theft looks like. I pictured a scene from a crime film, where masked robbers are chased down by police with sacks of GPUs slung over their shoulders. Warwick laughed and said that the reality isn’t too far from it. “This is organised crime and organised at a global level,” he said.

These groups have their own supply chain experts and digital specialists, and they’ve identified high-value, highly mobile targets such as GPUs and switch gear as lucrative new revenue streams.

He shared an example of a theft at a distribution center where a truck pulled up to the facility, "a worker came out with a high visibility branded vest on and a clipboard” and “within 7 minutes of being parked, the back doors of the vehicle were opened and there was a $4 million theft.”

AI infrastructure theft is becoming more commonplace in part because it offers a better risk-reward than other crimes. “The risks associated with trying to get a truckload of fentanyl across the US-Mexico border versus stealing one pallet in a Sprinter van of GPUs are stark,” Warwick said. Fencing stolen hardware is relatively easy, he said, and the prices soaring: “A pallet of smartphones runs about $1 million. Right now, a pallet of Blackwell chips will run about $17 million.”

Overhaul has formed a partnership with the international criminal police organisation INTERPOL, as “even they are struggling to understand just how dynamic these criminal organisations are.” He said Interpol came to his team for insight because tracking these groups requires specific expertise in how the industry’s supply chains work, in addition to traditional law enforcement methods.

Endless Subcontracting Exposes a Major Vulnerability in Transport

To understand where the vulnerabilities are in a typical data center supply chain, I asked Warwick whether it’s technology, human behaviour, or trust gaps between parties that create the greatest risk. His answer was all of the above, including using legacy systems to track deliveries, poor training, and a lack of proper verification and vetting processes.

One weakness he emphasised is the layered subcontracting model common in the supply chain and transportation industry. “The majority of over-the-road transport in the US is delivered by owner operator truckers, around 700,000 independent companies, so the entire network is built on contractors subcontracting,” he said.

This layering makes it difficult to verify who is picking up a high-value shipment, and criminal organisations have learned to exploit this gap. “Criminals have figured out how they can run a legitimate operator who will go to a load board and pick up a load that has been subcontracted” where there is little to no vetting, he said.

I’ve seen subcontracting problems in other sectors, notably from my experience covering telecom infrastructure in Africa, where major players moved to a policy of single-tier subcontracting to reduce such risks.

Warwick agreed that tightening subcontracting rules is a necessity but pointed to a legal shift that is already underway. A recent U.S. Supreme Court ruling involving CH Robinson, one of the largest freight brokers in the world, is changing how negligence liability works right now.

Previously, if a broker subcontracted a load and that subcontractor subcontracted again, the original broker was usually not liable if something went wrong. The new ruling makes the original carrier liable regardless of how many layers of subcontracting occurred. “It doesn't matter who you subcontract to, if you were the bespoke carrier or broker to begin with, you are now liable," he said. This ruling is already tightening subcontracting practices across the U.S.

The Pain of Stolen Hardware Goes Far Beyond Cost

Let’s say a pallet of equipment gets stolen. It’s a lot of money, but for multi billion dollar projects, surely you just buy a replacement? It’s true, Warwick said, that one reason supply chain risk isn’t a bigger consideration already is because the blistering pace of implementation pushes this risk down the priority list.

But given today’s equipment scarcity, stolen equipment from the AI infrastructure supply chain can’t be treated like retail shoplifting, where some level of stolen goods is just a financial loss you absorb and replace. “Scarcity is such a big problem that the theft isn’t just a financial loss, it’s an industry disruption as there’s no spare replacement equipment,” Warwick said.

Warwick said he’s aware of three separate thefts over the past six weeks targeting hyperscalers that totalled $222 million in losses.

GPU, semiconductor, and other equipment vendors have order books full for years, meaning there’s little inventory available to replace stolen components. Hyperscalers can use their size and leverage to negotiate priority access, but once equipment is purchased and in transit, there's no backup supply waiting in the wings.

Theft is a particular risk for neocloud providers that often pre-sell compute capacity. “There’s a lot of venture capital going into new providers, building bespoke data centers,” Warwick said. “They sell it in advance, and suddenly a theft of compute means they can’t deliver on time, and there’s no plan B.”

Another major theft incentive is black market sales into embargoed countries. Warwick explained that the United Nations and European Union restrict 27 countries from receiving certain technology goods for security concerns. “These markets have become some of the most profitable places in the world for selling racks, servers and other related infrastructure,” he said.

This black-market trade also creates new risk as regulators increasingly hold OEMs and technology firms accountable if their branded hardware ends up in restricted markets such as Iran, Russia, or China, regardless of how the diversion happened. That has turned supply chain risk into a regulatory and reputational issue for hardware makers, as well as an operational one.

Walking Through a Real Case

Warwick walked me through a real, anonymised case of a thwarted potential theft of $40 million worth of switch gear and GPUs.

The load had been subcontracted multiple times, allowing for it to be “picked up with fictitious paperwork. So, the driver had a false license with a fake DOT (Department of Transportation) number. ... It was essentially a shell company and a fake carrier."

Using IoT sensors and live telemetry that his team deploy on high-value loads, they detected the truck deviating from its planned route. Coordinating with the carrier, local law enforcement, and the destination transload center, they intercepted the shipment before the stolen goods could be offloaded.

"Who do you blame?” Warwick asked. “Was it the operator for hiring that carrier? Was it the carrier for subcontracting? Was it the subcontractor for subcontracting again? And the answer is all of the above."

The operator, the carrier, and multiple layers of subcontractors all bore some responsibility, but the fragmented nature of the chain meant no single party owned all the risk.

Industry Steps to Cut Supply Chain Theft Risk

Warwick points out signs of maturing practice among hyperscalers that include tighter vetting, more rigorous planning cycles, and stronger contractual accountability for transport providers.

He also talked about an emerging trend of inventory financing, where specialised finance companies now own infrastructure assets such as generators, racks, and servers on behalf of hyperscalers, who lease them. This model is popular because it means operators no longer need to worry about owning the underlying assets.

However, Warwick feels that these finance companies don’t yet fully grasp the theft risk they’re taking on, and that these companies need to think more about security as a preventative investment rather than a reactive cost.

I ended by asking Warwick for one immediate, low-cost action operators could take. His answer wasn't a tool or a technical fix: “Stop thinking about all of these pieces as commodities or as inventory and start thinking about them as strategic assets.” That reframing provides a foundation everything else needs to be built on—including security strategy.