Orbital Power for Compute Is Closer to Terrestrial Parity Than Most Expect ⚡🛰️📉 In Part 2 of our orbital compute series, we projected the $/W of powering compute in high Earth orbit (HEO). At ~$2,000/kg to HEO, orbital power & cooling costs ~18-26 $/W; about 2× the ~12 $/W terrestrial datacenter benchmark. Under a reusable Starship with orbital refuelling, HEO delivery cost falls fast. We modelled three different satellite architectures and where they reach parity with terrestrial benchmarks, in terms of launch costs... Starlink-satellite HEO parity: ~500 $/kg to HEO Compute-Optimized Starlink (standard PV) HEO parity: ~1,000 $/kg to HEO Thin-PV 'Frontier' Tech Satellite: ~500 $/kg to HEO At 100 $/kg to HEO: orbital power hits 6-9 $/W, beating Earth by 25-50%, depending on architecture. The drivers and assumptions: 1️⃣ W/kg of the power + cooling subsystem (Starlink: 107 → Compute-Optimized: 160 → Thin-PV 'Frontier': 250) 2️⃣ Power + cooling hardware $/W at scale (Current Starlink: 6.1 → Compute-Optimized Starlink: ~5.0 → Thin-PV: ~9.0) 3️⃣ HEO sunlight advantage (~95% vs ~65% in LEO), and higher PV efficiency (~30% in space vs ~20% on Earth). The three satellite architectures behave differently: 🔴Thin-PV Frontier (only wins when launch is expensive) Thin-PV is cheapest at high launch cost because its high W/kg minimizes the launch penalty, but once launch drops below ~500 $/kg, its high hardware $/W makes it the most expensive option. ⚫️Starlink-Class (steady baseline) Starlink-class hardware becomes roughly cost-equal with terrestrial power at ~500–600 $/kg to HEO, with no redesign required. 🟢Compute-Optimized Starlink (cost leader long-term) Compute-Optimized Starlink becomes the cheapest overall once launch falls below 1,000 $/kg, showing that low-cost hardware beats chasing extreme W/kg at scale. Despite economic parity being around the corner, Elon’s push isn’t primarily about cost. Orbit offers optimal solar flux and unconstrained physical volume, resources Earth cannot scale. Full analysis here🧐

Our TPUs are headed to space!  Inspired by our history of moonshots, from quantum computing to autonomous driving, Project Suncatcher is exploring how we could one day build scalable ML compute systems in space, harnessing more of the sun’s power (which emits more power than 100 trillion times humanity’s total electricity production). Like any moonshot, it’s going to require us to solve a lot of complex engineering challenges. Early research shows our Trillium-generation TPUs (our tensor processing units, purpose-built for AI) survived without damage when tested in a particle accelerator to simulate low-earth orbit levels of radiation. However, significant challenges still remain like thermal management and on-orbit system reliability.  More testing and breakthroughs will be needed as we count down to launch two prototype satellites with @planet by early 2027, our next milestone of many. Excited for us to be a part of all the innovation happening in (this) space!