8 CPU cores, 4 Xe cores, and 75 TOPS—radiation-hardened for orbit.
It turns out consumer laptops and server racks aren’t enough for Santa Clara as Intel wants to conquer outer space. Unlike Elon they don’t plan to launch data center in space, but they made a radiation hardened version of Panther Lake. Intel has officially detailed Starfire, its new space-bound SoC aimed squarely at U.S. defense satellites and orbital compute rigs. My head keeps telling me FireStarter would have been a cooler name, but that’s probably bad karma for spacecraft or a satellite.
If you were hoping to drop one of these bad boys into a Mini-ITX rig or run Doom in zero gravity, adjust your expectations. This is strictly space-grade silicon built for environments where radiation and thermal limits matter infinitely more than your 3DMark score.
Panther Lake Goes to Space
Under the hood, Starfire is a radiation-hardened relative of Intel’s upcoming Panther Lake architecture, stitched together using Foveros 3D stacking. Intel is using a hybrid process strategy across its tiles:
CPU Tile: 4 Performance cores and 4 Low-Power Efficiency (LPE) cores on the flagship Intel 18A node.
NPU Tile: 3 NPU tiles, also on 18A.
GPU Tile: 4 Xe cores (64 EUs), fabricated on the mature Intel 3 process.
Why mix 18A and Intel 3 on one package? Beyond yield economics, space silicon has to survive Total Ionizing Dose (TID) and Single Event Latch-ups (SEL). Smaller, denser transistors carry less charge and flip far more easily when hit by cosmic rays. Combining 18A with design-level hardening and mature process nodes keeps the chip from bricking mid-orbit.
Two Flavors: 10W Trickle vs. 35W Beast
Depending on a satellite’s solar panel budget, Intel will offer two configurations:
Low Power SKU (10W TDP): Clocks P-cores to a conservative 1.0 GHz and LPE-cores to 850 MHz, with the GPU scaling between 800 MHz and 1.0 GHz. Delivers up to 45 TOPS of AI compute.
Performance SKU (35W TDP): Cranked up for real-time high-res imagery, pushing P-cores to 3.1 GHz, LPE-cores to 2.1 GHz, and the GPU to 2.0 GHz. Bumps total AI output to 75 TOPS.
Both SKUs support LPDDR5/DDR5 memory and 12 lanes of PCIe 4.0. They are rated for extreme temperatures ranging from -55°C to 125°C, with a target operational lifespan of over 10 years.
Why Space Needs 75 TOPS
Historically, spacecraft relied on ancient, heavily shielded chips like BAE’s RAD750—processors running under 200 MHz that couldn’t handle serious calculations. Putting 75 TOPS of AI power directly onto a satellite allows agencies to run local inference and filter high-resolution imagery in orbit, rather than wasting precious bandwidth downlinking raw data to Earth.
Intel plans to ship engineering samples to U.S. government customers in Q3 2026, with production handled entirely on American soil. Whether commercial players like SpaceX or Blue Origin get access remains to be seen.
Say what you will about Intel, but porting an 18A Panther Lake derivative to space hardware this close to the node’s debut is genuinely impressive.
