Let’s run the numbers. The published specs: 150 kW peak / 120 kW sustained, a 110 m² deployable liquid radiator assumed at ~1,400 W/m², radiating from both sides, oriented knife-edge to the sun . Note 110 × 1,400 ≈ 154 kW so the spec is exactly sized to peak load, meaning both sides combined must net 1,400 W per m² of panel, or ~700 W/m² per face.
Stefan-Boltzmann check: A surface radiates εσT⁴. With a good coating (ε ≈ 0.9), emitting 700 W/m² per face requires a radiator surface around 342 K (~69°C). Add realistic environmental loads at ~600 km (Earth IR hitting one face even when knife-edged to the sun, maybe 50–100 W/m² absorbed) and you need closer to ~350 K (~77°C) at the radiator surface.
That’s where it gets tight. The coolant must be hotter than the radiator surface (fin losses, ΔT across the loop), so call it 85–95°C fluid and the chips hotter still, putting junction temps at 100°C+. That’s at or beyond where today’s AI silicon wants to live; terrestrial liquid-cooled racks run coolant at 30–45°C precisely to keep junctions under ~90–105°C. Notably, Musk himself hedged: “Over time, we think we can do about 250W and 1,400W, respectively” i.e., 1,400 W/m² is a target, not current capability.
Verdict: It doesn’t violate physics, so it passes a strict smell test unlike comparisons to ISS radiators (~100–350 W/m²), which are misleading because ISS rejects heat at near-room-temperature for crew, and radiated power scales as T⁴. At the sustained 120 kW load, the panel only needs ~545 W/m² per face → ~57°C radiator, which is genuinely achievable with hot-running chips.
But the margin is razor-thin at peak: the radiator is sized exactly to the load, with nothing left for coating degradation over years of atomic-oxygen exposure, micrometeoroid damage, off-nominal attitudes, pump inefficiency, or non-uniform panel temperature (real radiators average cooler than their hottest point, cutting effective flux 10–20%). Skeptics making this same calculation have argued a 110 m² radiator handling 120 kW average heat is not easy once you account for solar exposure, Earth IR, orientation, and chip temperature limits , and some conclude a real system needs radiators several times larger or less compute per satellite .
So: plausible as a design point if they co-design silicon to run hot (which SpaceX, building its own chips, can do), aggressive-to-optimistic as a v1 spec, and essentially zero thermal margin at 150 kW peak.
I feel they should just create new permissions, same as location access which apps specifically have to request. They should specifically have to request access to messages etc etc.
This way you can still have Claude installed but it can’t rummage through your private messages emails photo’s etc
@johannesmkx I don’t like the orange individual either but Xi has shown these trees to other world leaders too. I don’t think there’s some hidden message in it