Intel’s New Space-Grade SoC for Space Computing and AI Takes Aim at Rad-Hard Silicon and the COTS Market

Intel recently unveiled Starfire, a new space-grade System-on-Chip (SoC) built on the company’s x86 architecture designed to allow spacecraft to perform more sophisticated Artificial Intelligence (AI) workloads while reducing Size, Weight, and Power (SWaP) onboard. Two power classes are planned, one under 10 Watts (W) for spacecraft for tighter energy budgets, and a high-performance variant that will consume more power in exchange for greater processing capability. While the processor’s radiation tolerance hasn’t been characterized, the processors are intended to mitigate radiation effects for missions lasting more than a decade. Discussions are already underway to incorporate Starfire into upcoming missions, with an experimental mission scheduled for later in 2026.

Starfire is looking to solve a problem in space computing that the space industry has been gradually evolving to overcome, a mismatch of clocks. Terrestrial silicon improves every 2 years, while satellites live 5 to 15 years, which has historically forced a choice between older radiation-hardened designs with deep heritage or Commercial Off-the-Shelf (COTS) solutions. Alongside Starfire, AMD’s Versal XQR adaptive SoCs and Microchip’s HPSC, are the industry’s modern commercial nodes that fill that gap with the silicon vendor standing behind space-grade screening, documentation, and support. In essence, a third lane for space-computing silicon is emerging, offering vendor-backed modern silicon and familiar software environments that lower the porting barrier for developers.

While the immediate beneficiaries are flight computer and payload processor vendors, incumbent rad-hard silicon vendors such as BAE and Frontgrade will face mounting pressure to argue product roadmap alongside flight heritage. Qualified COTS solutions held by players like Teledyne, 3D PLUS, or Mercury Systems will now face competitive pressures from solutions with comparable developer familiarity and space-marketed support. The familiar software also aims to align the terrestrial stack. The x86-plus-Ubuntu stack from Starfire means terrestrial software ports in orbit, a genuine requisite for the emerging orbital cloud.

While the product is being positioned for space-based computing and AI, Intel’s Starfire is not marketed as an Orbital Data Center (ODC) platform, but more as an edge inference on space-born data accelerator. With a focus more so Neural Processing Unit (NPU)-accelerated image analytics, object detection, and sensor fusion in the 10 – 35 W power class, rather than the Kilowatt (kW) or orbital hyperscale (Megawatt (MW)) class AI systems. Indeed, AI implementation in space is not new and today the focus is on solving challenges the market has been experiencing for a few years, reducing processing overhead, overcoming limited satellite life span, and decongesting satellite downlink bottlenecks—particularly in the Synthetic Aperture Radar (SAR) and Earth Observation (EO) markets.  In essence, Starfire is betting on the demand for making satellites smarter, a real challenge with demand today across sovereign-class space communications and domain awareness, while NVIDIA’s Space-1 Rubin bets that orbit becomes a place where computing lives.  

For players entering the space industry from adjacent markets, Starfire is the template: internally funded, partner-led, and built on strengths the existing space industry cannot easily replicate, from fabs to developer ecosystems to trusted domestic supply. The differentiator will not be capability, but commitment. The question for every incumbent and new entrant is no longer on whether modern silicon can go to space, but who will still be supporting it a decade after launch.

• Silicon Vendors: One of the major takeaways for silicon vendors from this announcement is that the space-qualification moat is becoming productized across the supply chain. With Intel and AMD offering screening data, documentation, and support with the product, flight heritage on its own becomes less defensible, as competition moves to software ecosystems and fast refresh commitments.
• Telcos: For telcos and network operators, the supplier map of space is converging with a vendor they are familiar with. The same vendor-backed compute now heading to orbit is what will carry regenerative payloads and onboard network functions as Non-Terrestrial Networks (NTNs) mature, which makes space capability a line item in familiar vendor evaluations, rather than an exotic procurement process.