SANTA CLARA, July 7 — The figure that matters: $1.5 million. That is the value of a single heist in January 2025, when thieves walked off a Santa Clara data center construction site with 50 NVIDIA H100 graphics processing units. At roughly $30,000 per GPU, the loss hit like a line-item shock.
But the thefts go far beyond one incident. By the numbers, the problem is escalating across the AI infrastructure boom.
Thieves are targeting not just copper wiring but the high-end computing gear that powers the artificial intelligence race. Servers, GPUs, and cooling units are vanishing from active building sites, often before the facilities are even operational. Investors will note the math: a single NVIDIA H100 GPU, the workhorse of AI training, carries a street price near $30,000 as of 2025.
Cooling units, essential for keeping dense server racks from overheating, run roughly $100,000 apiece. Multiply those figures by the dozens of units sitting in half-finished data centers across the country, and the exposure becomes staggering.
The bottom line: construction security has not kept pace with the value of what is being installed. A March 2025 report from Axis Communications, a security firm, found that 40% of data center construction sites lack 24/7 surveillance. That gap leaves multimillion-dollar inventories vulnerable during the critical window between delivery and final installation.
The report recommended GPS trackers on high-value items and deploying drones for perimeter monitoring — steps that remain optional for many builders. What gets stolen does not stay stolen. The equipment typically resurfaces on online marketplaces such as eBay or moves through underground resale networks.
A GPU that powers cutting-edge AI models can be wiped clean and resold to unsuspecting buyers, or to buyers who do not ask questions. There is also a data-security angle, though it is less common.
Servers and GPUs can contain sensitive information if they are stolen before being wiped. Most stolen items are cleaned before resale, the source material notes, but the risk of a breach remains for any equipment taken mid-deployment. The pattern is clear: as AI infrastructure spending accelerates, so does the incentive for organized theft.
Data center construction sites are sprawling, often located in industrial zones with limited natural surveillance. A single GPU is small enough to be carried out in a backpack.
A cooling unit requires more muscle but still fits on a flatbed truck. For now, the industry is playing catch-up. Axis Communications’ recommendation of GPS trackers and drone patrols points toward a security model that treats each GPU like a piece of jewelry rather than a commodity server component.
Whether builders will absorb that cost — or wait for the next $1.5 million loss to force the issue — remains the open question.
{“@context”: “https://schema.org”, “@type”: “NewsArticle”, “headline”: “Copper to servers: Equipment stolen from AI data centers”, “description”: “SANTA CLARA, July 7 — The figure that matters: $1.5 million. That is the value of a single heist in January 2025, when thieves walked…”, “datePublished”: “2026-07-07T09:30:00+00:00”, “author”: {“@type”: “Person”, “name”: “Ava Anderson”}, “publisher”: {“@type”: “Organization”, “name”: “InfoPulse Today”, “url”: “https://newsfeed.infopulsetoday.com/”}}
