The planned addition of 660,000 Nvidia GB300 GPUs in the fourth quarter will increase training capacity for Grok in record time. New analyst projections place the combined server power of the expansion at 2.7 GW by year-end, though some researchers argue actual operational capacity is currently only 1 GW. This discrepancy stems from a SpaceX 10-Q filing claiming 0.4 GW of capacity had been deployed, which critics contend is only now becoming functional.
The rollout will bring the total GPU fleet to 1.44 million units, adding to the existing Colossus 1 and Colossus 2 clusters. Elon Musk scheduled the deployment of 220,000 chips next week, followed by similar installments in November and December at the high-density Minihard facility. The combined hardware investment exceeds $100 billion and puts SpaceX's compute potential 20% ahead of 2026 estimates.
Key sources
- SOURCE@elonmusk“Another 220k GB300 will be fully operational next week and another 220k in November. If we get lucky, yet another 220k GB300 by late December.”x.com
- SUPPORT@sawyermerritt“possible total of 660,000 GB300s coming online in Q4 alone. Upwards of $50 billion of GPUs.”x.com
- SUPPORT@sawyermerritt“over $100 billion of GPUs combined, and brought online in record time. Over 2 GW of compute.”x.com
- SUPPORT@kakashiii111“while SpaceX claimed in its latest 10-Q to have deployed an additional 0.4GW, that capacity was not operational or online.”x.com
- SUPPORT@shanumathew93“implies ~2.7 GW of server power (~3+ GW facility) by year-end if the December tranche arrives on time”x.com
- SUPPORT@wintonark“If they deliver on these installs it would put them 20% ahead of our 2026 expectations.”x.com
- SOURCEmarketbrief.now