AMD Starts Sampling a Space-Grade Versal AI Core SoC Built to Survive 15 Years in Orbit

AMD has started shipping engineering samples of its Versal AI Core XQRVC1902 adaptive SoC in a new space-grade package designed for missions lasting up to 15 years. The company announced the sampling program with early-access customers on October 1, marking the point where its adaptive computing line formally aims at the most demanding end of the space market rather than just commercial "new space" missions.

The new package is being qualified against MIL-PRF-38535 Class Y, the highest level of quality and reliability assurance under the US military specification for spaceflight components. AMD says the design combines conservative design rules with an enhanced organic substrate material to absorb the thermal and mechanical stresses of launch and orbit, paired with space-grade chip capacitors that already carry flight heritage. Flight-qualified Class Y units are expected to become available in the second half of 2027.

Packaging is the hard part of putting compute in orbit. AMD chose a lidless organic design, which exposes the die directly and improves heat dissipation - a meaningful advantage for onboard processing workloads that run continuously for years without maintenance. The new package is also pin-compatible with existing Versal AI Core VC1902 commercial, defense-grade and space-grade devices in the 2197-ball grid array, so teams can move an existing design to the flight-qualified part without a board redesign.

The XQR Versal AI Core line is built for onboard payload applications: AMD cites generational increases in compute density for vector-based algorithms, system logic cells, on-board SRAM and multi-gigabit transceivers. Ken O'Neill, AMD's aerospace and defense lead systems architect, said the samples "mark an important milestone toward Class Y qualification" and let customers build engineering models for missions "demanding absolute reliability" - from geostationary and cislunar orbits to deep space and human-rated flights.

The announcement lands in a week when in-orbit AI compute has become a live topic rather than a thought experiment. Google sent its first TPU-carrying satellite up on a SpaceX Falcon 9 Transporter-18 rideshare as part of Project Suncatcher, and satellite software startup Satlyt raised $8 million to run models like Gemma directly on orbiting hardware. Those efforts focus on GPUs and TPUs in low Earth orbit; AMD is attacking the same problem from the reliability and packaging side, where qualification cycles are measured in years.

That timing gap defines the opportunity. A 2027 flight-qualified date means the customers sampling the XQRVC1902 now are building engineering models, not launching hardware - the classic cadence of space-qualified silicon. If AI processing genuinely migrates onto spacecraft, the winners will be the vendors whose parts clear military qualification before the market consolidates, and AMD has just moved its adaptive SoC line to the front of that queue.