Ukraine's military intelligence says Russia is using Nvidia's Jetson Orin modules, consumer-grade AI chips designed for students and developers, inside its latest S-71 'Monochrome' cruise missiles for autonomous targeting. The discovery highlights a growing tension in AI hardware policy: while the U.S. restricts powerful chips used to train large AI models, it does not control the smaller, cheaper chips that run those models in the field.

How Are Consumer AI Chips Ending Up in Military Weapons?

According to Ukraine's Main Directorate of Intelligence, the S-71 'Monochrome' cruise missile contains an Nvidia Jetson Orin NX module marked as SNVUP6.MOP TE980M-A1, packaged in October 2025. The Jetson Orin NX is a compact system-on-module built around Nvidia's Ampere GPU architecture, featuring up to eight processing cores and 1024 CUDA cores (Compute Unified Device Architecture, Nvidia's software platform that lets developers write code for Nvidia hardware). The chip delivers up to 157 sparse INT8 TOPS (trillion operations per second) of AI performance, making it suitable for real-time image recognition and autonomous decision-making in space and power-constrained environments like drones and missiles.

In the S-71 'Monochrome', the Jetson Orin NX likely functions as the perception engine for the missile's electro-optical guidance system, working alongside a Chinese-made Honpho TS130C-01 camera module to recognize targets in real time and assist terminal guidance. The missile can be launched from Su-57 fighter jets or S-70 Okhotnik drones, carries a 250-kilogram high-explosive payload, and has a range of up to 300 kilometers.

This is not the first time Ukrainian intelligence has reported Russian use of Jetson Orin NX modules. Previously, they documented the chips in Russia's Shahed MS001 autonomous drones, where they performed local decision-making and terminal guidance functions.

Why Doesn't the U.S. Block These Chips from Russia?

Nvidia positions the Jetson Orin NX as a consumer-grade product sold openly to students, developers, and startups for beneficial applications. The company does not export-control the Jetson line, even though it restricts far more powerful chips like the H100 and B200 accelerators used to train large language models. The reasoning is straightforward: the U.S. government controls the hardware that trains AI models, not the hardware that runs them. Since AI models are trained anyway and inference hardware is widely available through retail channels, controlling consumer-grade chips would have limited practical effect.

Jetson Orin NX modules have been shipping since 2023 and are sold globally through numerous reseller channels. A German retailer, for example, sells them for approximately 685 euros with value-added tax included. Nvidia manufactures hundreds of thousands, if not millions, of these modules annually. While the modules themselves are available for purchase, most Jetson Orin NX products in retail are small-form-factor development kits rather than bare modules, though the modules can be obtained separately.

"Our Jetson Orin modules are consumer-grade products sold to students, developers, and startups for a wide range of beneficial applications. They are not available in Russia and are not designed for military purposes," said an Nvidia spokesperson.

Nvidia Spokesperson, Nvidia

Nvidia acknowledged that it cannot track products after they leave its supply chain. The company stated that if it determines any customer is violating U.S. export controls, it will take appropriate action. However, the widespread availability of Jetson modules through reseller networks makes enforcement difficult.

What Does This Reveal About AI Hardware Longevity?

The use of Jetson Orin NX in Russian weapons also underscores a broader trend in AI infrastructure: older hardware remains valuable and mission-capable far longer than many expect. Nvidia's Chief Executive Officer Jensen Huang recently emphasized that the company's A100 GPU fleet, which launched in 2020, remains mission-capable through 2029. This extended lifespan is enabled by CUDA, Nvidia's software platform that allows developers and engineers to continually upgrade and optimize Ampere, Hopper, and Blackwell GPU generations throughout their useful lives.

"The mighty A100 fleet is mission-capable from 2020 through 2029. NVIDIA computing is more than chips. CUDA gives developers and NVIDIA engineers a common platform to continually upgrade Ampere, Hopper and Blackwell throughout their useful lives," stated Jensen Huang.

Jensen Huang, Chief Executive Officer, Nvidia

CoreWeave, a GPU rental company, recently signed an agreement to continue renting Nvidia A100 GPUs through 2029, demonstrating that legacy hardware can support long-term workloads and remain financially viable. The arrangement shows that even as newer Blackwell and Rubin chips arrive, older accelerators retain real value rather than becoming obsolete. This longevity principle applies equally to consumer-grade chips like the Jetson Orin NX, which can serve in applications for years after their initial release.

Key Factors in the Export Control Debate

• Training vs. Inference Hardware: The U.S. restricts high-end accelerators like the H100 and B200 that train AI models, but does not control consumer-grade chips that run pre-trained models locally, creating a regulatory gap.
• Global Supply Chains: Jetson Orin NX modules are sold through numerous reseller channels worldwide, making it difficult for Nvidia or regulators to prevent unauthorized acquisition or re-export to sanctioned countries.
• Dual-Use Technology: The same AI chips designed for robotics, drones, and autonomous vehicles can be repurposed for military guidance systems, blurring the line between civilian and military applications.
• Hardware Longevity: Older AI chips remain mission-capable and valuable for years, extending their operational lifespan in both commercial and military contexts.

The discovery of Nvidia chips in Russian missiles raises uncomfortable questions about the limits of export controls in an era of widely available AI hardware. While the U.S. can restrict the most powerful training accelerators, the proliferation of consumer-grade inference chips means that adversaries can acquire capable AI hardware through legitimate commercial channels, resellers, or secondary markets. As AI becomes embedded in more weapons systems, the gap between what regulators control and what militaries can actually obtain may continue to widen.