Nvidia Unveils Computing Platform for Space Data Centers at GTC 2026

Computing heads to orbit

Nvidia CEO Jensen Huang announced the development of a computing platform designed for space-based data centers at the GTC 2026 conference. Huang described space computing as "the last frontier," emphasizing that intelligence must reside where data is generated — especially as satellite constellations continue to expand.

According to Nvidia's press release, several companies will deploy the Vera Rubin Space-1 module — which incorporates IGX Thor and Jetson Orin — in upcoming space missions. The chips are engineered specifically for environments with severe constraints on size, weight, and power consumption.

Huang acknowledged significant engineering challenges. Without convection in space, heat dissipation relies solely on radiation, making cooling one of the critical problems engineers are working to solve.

Why it matters

Surging demand for AI workloads is driving up electricity costs and putting pressure on terrestrial data center capacity. Space offers a potential solution: unlimited room and constant access to solar power. However, launch costs remain a major obstacle.

Nvidia's initiative aligns with a broader industry push toward orbital computing. In February, SpaceX filed with the U.S. Federal Communications Commission for permission to deploy a constellation of 1 million satellites for orbital data processing, connected via laser links. California-based startup Aetherflux plans to launch low-orbit solar mini-farms in 2026 that beam energy to Earth using lasers, riding aboard SpaceX rockets. In November 2025, Google disclosed ambitions to build a satellite system for harvesting solar energy to power its data centers. That same month, researchers at the 33FG group estimated that AI computing in orbit will be cheaper than on Earth by 2030.

$1 trillion order forecast

Huang revealed that expected orders for Blackwell and Vera Rubin generation chips are projected to reach $1 trillion by 2027. The company had previously estimated potential revenue from these two chip families at $500 billion. CFO Colette Kress noted after last month's earnings report that 2026 growth could exceed prior projections.

According to Huang, demand from both startups and large enterprises continues to accelerate, as greater computing power enables more token generation and higher revenues.

Autonomous driving: expanding partnerships

Nvidia is deepening its footprint in self-driving technology. The company announced new agreements with Hyundai Motor, Nissan Motor, Isuzu, BYD, and Geely centered on the Drive Hyperion platform. The system supports development and integration of driver-assistance tools and Level 4 autonomous driving capabilities.

Huang declared that "the ChatGPT moment for self-driving cars has arrived," though no fully autonomous vehicles exist on the market today. Some companies, such as Waymo, already operate Level 4 robotaxi services. Most commercial autopilot systems still function at Level 2, requiring constant human supervision.

Current Drive Hyperion customers include Aurora Innovation, Nuro, Sony Group, Uber, Stellantis, and Lucid Group.

Groq 3, Kyber, and other releases

At GTC 2026, Huang introduced the Groq 3 Language Processing Unit (LPU) — the first chip from startup Groq, which Nvidia acquired in December 2025 for $20 billion. Shipments are expected in the third quarter. He also unveiled the Groq 3 LPX server rack comprising 256 LPUs, designed to operate alongside the Vera Rubin system. Huang stated the rack can boost compute efficiency in tokens per watt for Rubin by 35 times.

Huang explained that combining two fundamentally different processors — one optimized for high throughput, the other for low latency — expands available memory capacity by adding a large number of Groq chips.

Additionally, Nvidia showcased a prototype of Kyber, a next-generation server architecture featuring 144 GPUs arranged vertically to increase compute density and reduce costs. Kyber will form part of the Vera Rubin Ultra system, with shipments planned for 2027.

A new developer toolkit was also presented, including the NemoClaw stack built specifically for OpenClaw.