June: V3 Starship operational for high-volume satellite deployment (post-April debut + quick iterations). Deploy 500–1,200 V3 comms satellites (Starlink Gen3 upgrades) in 2026. June IPO closes → $50–100B cash. Immediate ramp of new solar/GPU satellite factories + 2–3 additional launch pads. Late 2026 AI V1 prototypes (70 kW, your power/weight Version 1) in orbit for testing. Part of one launch. Initial 10–50 AI sats as proof-of-concept. Q2 2027 (AI V1 Ramp Begins — First Operational Orbital Data Centers) Full switch to Starthink AI V1 satellites (70-140 kW each, 5–10 MW power per launch). Could deploy 2,000–10,000 AI satellites but what is the mix of communication to AI in 2027 (only 14–140 launches needed — easily within 100+/yr cadence across Texas + Florida + new pads). Maybe 5000 high speed internet, 5000 Direct to cellphone and 5000 AI satellites in 2027. Could be double or one third. First clusters of orbital AI data centers go live. Constant sunlight + vacuum cooling = MW-scale compute running 24/7. This is the “first Dyson swarm phase” milestone — recursive self-improvement hardware in space. 2028 (V4 Scale-Up + GW-Level Power)V4 satellites (140 kW next-gen, double power density) enter production/launch. Annual deployment jumps to 10,000–50,000+ sats/year. Total constellation: 20,000–100,000+ AI sats. Power output in tens of GW. Full-scale AI training clusters operational. Lowest-cost compute on Earth (or in orbit) achieved. 2029–2030 (Full 1M Constellation)