SpaceX Says Power and Cooling, Not Chips, Now Limit AI Expansion
SpaceX says its AI deployment is constrained more by power, cooling and electrical infrastructure than semiconductors. Ambitious capacity targets support suppliers, but investors still need verified spending, contracts and returns.
SpaceX told investors that power, cooling and electrical infrastructure have become the main constraints on its artificial-intelligence expansion, displacing chip availability as the immediate bottleneck. If the assessment is correct, the next phase of AI spending could favor grid equipment, turbines, cooling systems and data-center construction alongside semiconductor suppliers.
Elon Musk said SpaceX expects to bring 2.5 to 3 gigawatts of computing power online by the end of 2026 and is fairly confident it can add 10 gigawatts during 2027. He also described more than 20 gigawatts next year as a small possibility. Those are exceptionally aggressive targets and should be treated as management ambitions until supported by site-level capacity, power agreements and equipment orders.
The company's engineers are working on power and cooling infrastructure, but SpaceX will still require outside suppliers. GE Vernova makes heavy-duty gas turbines, while Eaton sells electrical distribution and cooling equipment used in grids and data centers. The remarks also support continued demand for Nvidia accelerators, Broadcom networking and custom chips, Micron memory, Intel processors and cloud capacity from Microsoft, Alphabet and Amazon.
SpaceX also reported rapid adoption of Grok Bot, an agentic system that performs multi-step tasks. President Gwynne Shotwell said the product had 800,000 paid weekly users earlier in October, more than double the prior month. Musk said usage per user was increasing and that the service could eventually consume most of the company's AI compute.
Management described strong economics for new compute deployments, with current pricing slightly above a range of $30 to $50 per watt and payback periods below one year. Those statements were not accompanied by enough detail to reproduce the calculation. The denominator, utilization assumption, contract duration and treatment of construction and financing costs all matter.
The source is also an investor commentary piece based on an earnings call, not an audited filing. SpaceX is privately controlled, and the report does not provide public financial statements or project-level contracts that independently validate the capacity or return claims.
For public-equity investors, the read-through is directional. A real shift from chip scarcity to power scarcity would broaden the AI trade toward utilities, industrial equipment and construction. It would not guarantee revenue for any named supplier. The strongest evidence will be disclosed orders, customer contracts and completed energized capacity, not a general statement that demand is high.
The AI buildout can support several layers of the supply chain at once, but the risk is synchronized overinvestment. Investors should compare each supplier's backlog, cancellation terms and cash conversion rather than assuming SpaceX's targets automatically translate into earnings.