How SpaceX plans to get from 1.7 GW to 13 GW of nameplate compute by the end of 2027, what stands in the way and how likely it is.
What SpaceX has said it will reach, next to what analysts expected before the Q3 call. Compare the dots with the rings, but note the analyst views are older.
Musk has not published a source-by-source power breakdown. He called the plan “highly proprietary” and declined to detail it. Pieced together from filings and analysts, it is mostly on-site natural gas, deliberately overbuilt. The weak links are turbine supply and build pace. Our estimate is a median of about 7 GW at the end of 2027, with roughly an 18% chance of 10 GW or more and about 7% for 13 GW or more.
The target. SpaceX had about 1.7 GW of nameplate compute at the end of September, expects 2.5–3 GW by year-end, and says it will add roughly 10 GW in 2027. That is separate from the power-and-cooling goal. SpaceX is aiming for 20 GW of power and cooling in 2027 so capacity runs far in excess of what’s needed to turn the GPUs on. On the Q2 call Musk said he expected closer to 15 GW if about a quarter of projects slip. We read this as roughly 1.5× overbuild. It also doubles as redundancy for islanded gas, which has no grid backup.
Per SemiAnalysis, the Southaven plant grew from 27 turbines (~495 MW) in February to 69 turbines (>1.2 GW) in July. APR Energy’s fleet is about 1 GW, and FTAI Aviation is a possible aeroderivative supplier. Morgan Stanley counts about 3–4 GW secured through procurement and nearly 8 GW of disclosed commitments. A bullish blog cites Doosan orders for twelve 380 MW units (4.5 GW), which we could not verify.
SpaceX is casting turbine blades and vanes in-house, which suppliers said could pull power forward 12–18 months. This is unproven. Only three foundries worldwide make the hot-section parts, with orders booked through 2030. The Bastrop plant was still at groundwork stage in late August, with no permits found.
Megapacks smooth the load but don’t generate power. SpaceX spent $329M on them in H1 2026.
Musk cited solar, wind and factory-built modular inference clusters with power and cooling built in. The S-1 showed $2.8B for more gas turbines and no significant terrestrial solar. We would treat non-gas as close to zero for 2027.
ERCOT suspended its large-load process after requests ballooned.
| Constraint | 2027 needs | Evidence so far | Read |
|---|---|---|---|
| Build pace | ~830 MW/month, ~2.5 GW/quarter | Southaven added ~140 MW/month. The best quarter guided is +0.8–1.3 GW (Q4’26) | 2–3× the best quarter, sustained for four; ~6× the power-add rate |
| Turbines | ~12–17 GW installed if islanded (our estimate) | GE Vernova builds ~20 GW/yr worldwide and is selling 2031 slots; Mitsubishi deliveries now run to 2031 | The binding constraint. The big three can’t help, so it falls to mobile, aero and Doosan units |
| Fuel | ~2–2.5 Bcf/day (our estimate) | Memphis/Southaven only | Needs several dedicated pipeline laterals |
| Chips | 220K GB300/month is ~5.3 GW/yr at about 2 kW per GPU (see Power and cooling on the Sites tab). 10 GW needs ~330–350K Rubin GPUs/month | Nvidia exclusivity covers at least 10 GW of supply | Second constraint; Rubin ramp timing is the risk |
| Money | $350B (UBS) to $500B (SemiAnalysis at $50B/GW) | Q4 capex is guided at $35–40B. $40B of chip financing is still in talks | Needs ~$90–125B/quarter, 2–3× the Q4 guide |
| Sites | ~23 Minihard-size buildings (0.44 GW each) | Nothing confirmed outside Memphis/Southaven. Texas sites were still being scouted in July | The gate |
Our estimate, not company guidance. The ranges add to 100%.
The street was at 4–6 GW before Q3: Morgan Stanley modeled 4.9 GW, and buy-side expectations were roughly 4–6 GW. We moved up from our early-October estimate of 12% for 10 GW or more. Q4 capex guidance landed well above UBS’s $19B expectation. We held back for the 2031 turbine slots and the lack of any confirmed new site. “Nameplate” is also a loose measure, so the reported figure will likely flatter productive compute.
These probabilities are judgment calls, not a forecast from SpaceX and not investment advice.
The second 220K GB300 batch confirmed by utility filings or imagery, not posts.
At least 2.5 GW reported.
A 2027 capex guide of $300B or more supports 8+ GW. $200B or less points to 5–6 GW or below.
First groundbreaking, air permit and gas lateral outside Memphis/Southaven.
The 440K Vera Rubin training cluster Musk said would be online by the end of Q2 (~1 GW). At least 6 GW total keeps 10+ alive; 4.5 GW or less caps it near 8.