Musk maps out SpaceX AI compute roadmap: 13GW by 2027, $500B ARR target

On SpaceX's Q3 earnings call, Musk laid out an extremely aggressive AI compute expansion roadmap, while also revealing key targets for Starship and Starlink, setting a vision of reaching $500 billion ARR and 20GW of compute by the end of next year.

Confirmed

Musk said that for the rest of 2026, roughly 200,000 NVIDIA GB300 GPUs will come online each month, bringing online compute capacity to 2.5–3GW by the end of 2026
In 2027, the plan is to add another 10GW of compute, bringing cumulative online capacity to 13GW; GPUs deployed in 2027 will be predominantly Vera Rubin
Plans call for a single-cluster training facility made up of 440,000 Vera Rubin GPUs to be completed by Q2 2027
On Starship: Flight 14 will reach orbit, and Flight 15 is planned to achieve full two-stage recovery; a new launch site in Louisiana was announced, targeting weekly launches first, then ramping to twice a week to achieve aircraft-level full reusability
Musk's overall goal for SpaceX is to hit $500 billion ARR and 20GW of compute by the end of next year

Why it matters

If delivered, 13GW of compute and a single 440,000-GPU cluster would place SpaceX's compute infrastructure among the largest in the world, directly rivaling the training capabilities of top AI labs
A deployment pace of 200,000 GPUs per month places extreme demands on supply chains, power, and data center construction; its feasibility will be a key industry watch point
Progress on Starship orbital flights, two-stage recovery, and the new launch site, combined with Starlink's commercialization, forms the core pillars underpinning the $500 billion ARR narrative

All of the above are verbal plans stated by Musk on the earnings call and should be treated as aspirational targets; actual execution remains to be verified.

2026-10-21 ~ 2026-10-21 · 6 related posts

Primary sources

Musk targets 13GW AI compute by 2027 with 440,000-GPU Vera Rubin cluster — scaling01 · 2026-10-21 05:00 (UTC+08:00)
SpaceX is deploying 220K Nvidia GB300s per month, and Musk says the pace will accelerate — firstadopter · 2026-10-21 05:24 (UTC+08:00)
Musk: 440,000-GPU Vera Rubin training cluster targeted online by Q2, 10GW added in 2027 — firstadopter · 2026-10-21 04:53 (UTC+08:00)

Musk claims SpaceX will add 200k GPUs monthly, targeting 13GW by 2027 — analysts skeptical — JOBhakdi · 2026-10-21
[source] Musk: 440,000-GPU Vera Rubin training cluster targeted online by Q2, 10GW added in 2027 — firstadopter · 2026-10-21
[source] SpaceX is deploying 220K Nvidia GB300s per month, and Musk says the pace will accelerate — firstadopter · 2026-10-21
Musk lays out SpaceX roadmap: 20GW compute next year, $500B–$1T ARR by end of 2027 — elonmusk · 2026-10-21

2 near-duplicate retellings: scaling01 · rickasaurus