SpaceX Is Building an AI Data Center Every Month—Here’s Why It Matters

Summary

The video presents a strongly bullish argument that SpaceX’s emerging AI data-center business could become more important in the near term than Starship or other long-term projects. Its central claim is that SpaceX is making data-center construction a repeatable, mass-produced process: several facilities are built at once, with a target of completing roughly one per month. The analyst says recent projects have taken about five months from construction to operation—far faster than the one-to-four-year timelines he attributes to other builders.

Main points

• Rapid capacity expansion: The discussion cites about 220,000 chips being added at a time. The analyst estimates SpaceX could reach roughly 2.6–3 gigawatts of compute capacity by the end of 2026, 10–12 gigawatts by late 2027, and potentially 20 gigawatts in 2028. These are projections, not confirmed outcomes.
• Repeatable construction and power infrastructure: The analyst describes multiple data-center buildings under development in a Tennessee–Mississippi complex, with Texas as a possible later expansion area. A reported power plan includes natural-gas turbines, grid power, solar, and batteries. The exact figures discussed are estimates.
• Leasing and internal use: The conversation suggests SpaceX may lease around 40% of its capacity and use the rest internally, though the speakers say this split is unconfirmed. Internal use could support products such as Grok and Grokbot, while leasing could provide external revenue.
• Cost and design advantages: The speakers argue that SpaceX is redesigning racks, cooling systems, and electrical infrastructure rather than relying entirely on standard third-party equipment. They cite more compact layouts, reduced cabling, liquid-cooling expertise, and an 800-volt DC approach as potential ways to lower costs and improve efficiency. The analyst estimates that a 400-megawatt facility could cost around $18–20 billion to build, while a lease could generate substantially more. These figures and the projected payback period are estimates.
• Demand, contracts, and financing: The analyst argues that strong demand for AI compute could keep rental prices high and make SpaceX’s capacity valuable, even if some customers are not publicly named. He also says revenue from SpaceX’s own services may be more useful for securing debt than relying solely on cancellable customer contracts.
• Chips and compute competition: The discussion focuses on Nvidia chips, including GB300 and anticipated Rubin systems. The analyst expects SpaceX to assemble racks in advance and deploy newer chips as facilities become ready. His broader argument is that more compute could create more opportunities to train and test models, potentially helping Grok compete with other AI systems.
• Comparison with other companies: The speakers contend that Microsoft, Meta, and other cloud providers face slower construction schedules, power constraints, or project delays. They argue that SpaceX could become a much larger AI-compute provider if its reported construction pace and expansion plans are achieved.
• Valuation forecast: The analyst makes an exceptionally bullish forecast of $600–$1,000 per share and an $8–$14 trillion valuation by 2027, based on projected data-center growth, leasing, and AI revenue. These are speculative predictions, not established valuations or guarantees.

The video is an investment and technology analysis; it is not a product review, guide, or tutorial.

Main speakers and sources

• Brian Wang, the principal analyst associated with Next Big Future, appears to be the main speaker.
• A second interviewer or co-host asks questions and responds; their identity is unclear from the subtitles.
• The speakers refer to estimates and commentary from Deutsche Bank, Epic AI, Wu Kong’s Substack, UBS, Cathie Wood, Morgan Stanley analyst Adam Jonas, Nick Gibbs, Elon Musk, and SpaceX’s CFO. These references are discussed by the speakers and are not independently verified in the subtitles.