Elon Musk claims SpaceX's in-house AI will surpass OpenAI's capabilities within six months, raising questions about the company's undisclosed compute and training infrastructure.

Elon Musk recently stated that SpaceX's AI initiative has been underway for approximately three years, compared with roughly six years for Anthropic and ten years for OpenAI. If current acceleration continues, Musk said, SpaceX could achieve a leading position within six months. He added that once intelligence far exceeds what is required for specific tasks, additional investment becomes less meaningful. While hardware deployment remains challenging and large-scale expansion of computing power is difficult, Musk noted that SpaceX has demonstrated expertise in building infrastructure and will continue to improve.

SpaceX has integrated the former xAI, establishing AI as its third business segment. In the second quarter, total revenue reached approximately $7.8 billion. The AI segment contributed $2.56 to $2.6 billion, representing a year-over-year increase of about 247%. The connectivity business, including Starlink, generated approximately $4.29 billion, while aerospace operations brought in about $962 million. During an August all-hands meeting, Musk stated that AI revenue would "probably surpass the rest of the company" in September and would significantly exceed it in the fourth quarter. He claimed that AI could account for roughly 99% of the company's value within four to five years. Musk set a target of 10 gigawatts of computing capacity by the end of 2027, with a valuation of $30 to $50 per watt, potentially generating annual revenues of $300 billion to $500 billion.

The Memphis and Southaven cluster has a nameplate capacity of approximately 1.4 gigawatts. Anthropic leases capacity from Colossus at roughly $1.25 billion per month. Musk clarified in May that the initial lease runs for 180 days, after which either party can terminate with 90 days' notice—a short-term structure SpaceX requires to reclaim capacity during periods of tight computing power availability. SpaceX's agreement with Google is valued at approximately $920 million per month for roughly 110,000 NVIDIA GPUs, with leasing expected to begin in the fall. Model training will remain on the ground, while daily inference is planned to transition to an orbital data center. Grok version 4.5 has entered internal testing at SpaceX and Tesla, with plans for an accelerated release schedule. Musk highlighted Starlink and Starship engineers as sources of acceleration.

The strategic logic centers on computing power that ties power plants, clusters, and launches to the same balance sheet, while the competitive advantage depends on rapid lab expansion driven by cloud vendor speed. The core claim is that the timeline for "catching up with models" has compressed to six months, with emphasis placed on self-use and the ability to reclaim leases. Revenue from tenants like Anthropic and Google funds cluster depreciation and the next generation of cabinets, enabling SpaceX to expand using its own power and land resources. This puts pressure on tenants relying on shared cluster capacity for training—short-term lease terms effectively convert their capacity into retrievable inventory. Whether model rankings shift within six months ultimately depends on whether hardware delivery continues to outpace competitors' algorithmic advancement.

Musk frames the earlier founding dates of OpenAI and Anthropic as a disadvantage: they started earlier, but SpaceX compensates with three years of launch and power infrastructure capabilities. The six-month timeline functions as a rallying call for shareholders and engineers rather than a claim that rankings have already shifted. Current revenue in the AI segment is primarily from computing power rentals—the $2.6 billion in Q2 derives mainly from cluster contracts, not Grok subscriptions. A monthly rental of $1.25 billion, if annualized, would exceed the entire company's 2025 revenue, but the 180-day plus 90-day exit clause indicates this revenue is not locked in as a long-term annuity through 2029. When Google enters the market in the fall, the AI segment will have a better opportunity to exceed Starlink revenue in a single month.

SpaceX's strategy follows a sequenced capital path: first sell computing power, then train its own models. Colossus generates external revenue and timelines while retaining the right to reclaim capacity internally, effectively positioning competitors as transitional tenants. Integration of coding tools like Cursor would supplement Grok's developer offerings rather than prove foundational model superiority over competitors like Opus. The 10-gigawatt target and $30-to-$50-per-watt valuation directly convert power infrastructure into the revenue model, while planned orbital inference incorporates Starship's capacity into future marginal costs. Engineers transferred from Starlink and Starship to AI bring the rocket company's execution culture to cluster deployment timelines.

This mirrors how Amazon uses AWS to support retail, Microsoft leverages Azure to back OpenAI, and Google deploys TPUs for both internal use and external leasing—all infrastructure companies discover that training becomes their highest-value customer. SpaceX's structural advantage lies in simultaneously owning launch capability, satellite networks, and ground power stations, allowing it to integrate power sourcing and hardware placement into unified timelines. The industry phase is shifting from comparing model parameters and benchmark scores to determining which company can first deliver incremental gigawatts monthly. The six-month window bets on execution speed, not research breakthroughs.

The competitive advantage hinges on technological substitution overlapping with capital concentration. Pricing power is shifting from the release cadence of model laboratories to integrated companies capable of delivering gigawatts monthly. The underlying mechanism is that the constraint in cutting-edge model training has migrated from algorithms to power infrastructure, land, and supply chains. Whoever can offer 180-day leases can toggle between internal use and external rental, adjusting to changing needs. If a six-month lead materializes, it will first appear in cabinet deployment and subsequently in model rankings.

Laboratories that started earlier do not necessarily outpace companies that can plug in power first. Short-term lease terms convert tenants' computing capacity into landlords' retrievable inventory. The timeline for model leadership is often determined by infrastructure timelines rather than research pace.