Elon Musk outlined Terafab's ambition at Baron Capital: global AI chip output is roughly 20-25 GW this year and 35-45 GW next year, while reaching one terawatt would require about 30X current global production. The remark offers a first quantitative benchmark for Terafab but lacks process, siting, funding or timeline details.
Gigawatt-scale Compute
吉瓦级算力
Metric expressing datacenter compute scale in electrical power, where 1GW maps to hundreds of thousands of high-end AI accelerat
Curated Intelligence
Musk said SpaceX is currently adding roughly 0.5GW of AI compute per month, equivalent to deploying about 220,000 GB300 chips plus supporting CPUs, networking and cooling, and plans to exceed 1GW per month by the end of next year. He identified power generation, not chip supply, as the real scaling bottleneck, framing that scale as approaching "superintelligence compute."
Crusoe said Google has acquired its southeast Wyoming data center site, completing Crusoe's pre-development work on site selection, government approvals and a hybrid power strategy developed with Tallgrass Energy. Crusoe says it has more than 6 GW contracted and over 40 GW in its pipeline as an 'AI factory' developer.
Broadcom is in early talks with private credit firms like Apollo and Blackstone to raise over $50 billion for OpenAI's custom 'Nexus' AI chip project, supporting a 10GW deployment through 2029. Concurrently, SpaceX seeks ~$40 billion to purchase Nvidia chips, while Oracle explores off-balance-sheet leasing structures to fund hardware acquisitions. These massive financing moves highlight the escalating capital intensity and complex supply-chain financial engineering in the AI infrastructure race.