$Rocket Lab(RKLB.US)$Amazon(AMZN.US)$Alphabet - C(GOOG.US)

Never understood how space relates to AI narratives. The following is excerpted from GPT, cost accounting is subjective.

:

II. Comparison of unit power heat dissipation capacity (core quantification)

1️⃣ Space: Pure radiation cooling (hard constraint)

Radiation power formula:

P = \varepsilon \sigma A T^4

Adopting aggressive but feasible engineering parameters:

• Emissivity ε ≈ 0.9

• Operating temperature T ≈ 350 K (≈ 77 ℃, chip reliability drops sharply above this)

Thus:

• Heat dissipation per square meter ≈ 700–800 W/m²

Which means:

• 1 MW heat → Requires ~1,300–1,500 m² radiator

• Equivalent to half a football field, just for "heat rejection panels"

⸻

2️⃣ Ground: Air/water cooling (engineering reality)

Typical ground data centers:

• Air cooling:

• 1 m² heat exchanger → 10–30 kW

• Liquid cooling (cold plate/immersion):

• Same volume → 50–100 kW

👉 Heat dissipation per unit area: Ground is 50–100× more efficient than space

⸻

IV. Key conclusions (extremely important)

① From a "heat dissipation" perspective alone

Space data centers are an economic disaster for thermal management

• 100–300× higher cost

• Technically feasible but not cost-effective

• Cannot be scaled (locked by physics)

⸻

② Why are people still researching space data centers?

Not for heat dissipation, but to bypass ground constraints:

Potential motivations Can offset heat disadvantages?

Nuclear power supply ❌ (nuclear ≠ cooling)

Military/extreme security ⚠️ (very limited cases)

Deep space edge computing ✅ (minimal computing)

High-frequency trading ❌ (speed of light limitation)

"Concept + policy funding" ✅ (real reason)

👉 Feasible scenarios:

• kW–10 kW scale

• Scientific research/military/deep space exploration

• Absolutely not hyperscale