SpaceX is targeting 10 gigawatts of AI computing capacity by the end of 2027, a buildout that SemiAnalysis estimates could generate about $300 billion in annual recurring revenue if the company monetizes roughly half of that capacity at frontier-model inference rates.
The proposed commercial model centers on short-term “urgent megawatt” contracts priced at about $50 million per megawatt annually, according to SemiAnalysis. The research firm argues that customers requiring immediate access to high-density AI capacity could pay far more than conventional data-center rates, pushing revenue toward $100 million per megawatt per year and gross margins above 85%.
Those projections place SpaceX’s reported AI infrastructure ambitions among the largest commercial compute plans disclosed to date. A 10GW deployment would require enormous volumes of graphics processors, generation equipment, networking hardware and physical space, while creating a business that resembles a high-margin utility for AI labs and large technology companies rather than a conventional cloud provider.
SpaceX’s first quarterly report following its public offering added urgency to that narrative. The company reported that revenue had doubled from a year earlier, while AI business revenue rose 247%. Capital expenditure reached $18.369 billion in the second quarter, demonstrating the scale of spending needed to pursue rapid expansion. Shares fell 13.6% in one day after the report before stabilizing near the offering price after lock-up shares became tradable.
Short delivery contracts carry premium pricing
SemiAnalysis bases its revenue case on the growing value of inference capacity: the computing resources used to run trained AI models for customers. The firm cited observations that OpenAI and Anthropic together added close to $30 billion in annual recurring revenue per month, an annualized pace of roughly $400 billion.
Its model estimates that a gigawatt of frontier AI capacity could produce approximately $100 billion in revenue against roughly $15 billion in costs. Those figures imply gross margins of about 85%, although such returns depend on sustained demand, high utilization and customers continuing to pay a premium for immediate delivery.
SemiAnalysis described an urgent-capacity contract priced at roughly $14 per hour for GB300 capacity, compared with a market average near $3 per hour. GB300 refers to Nvidia’s next-generation AI computing platform. The premium reflects an increasingly valuable feature in the AI infrastructure market: availability before a customer’s own data center is completed.
The contracts reportedly include a 90-day cancellation provision. That structure would give large AI users flexibility to secure computing power during a temporary shortage without making the multi-year commitments commonly associated with data-center leasing. It would also expose providers to a sharp utilization risk if customers reduce workloads or bring their own capacity online faster than expected.
SemiAnalysis frames the opportunity against typical 12-to-18-month data-center construction schedules. A provider able to make large blocks of power and compute available in months rather than years could fill the gap between AI demand and the slower process of permitting, constructing and energizing permanent facilities.
Warehouses and turbines underpin the buildout
The physical constraints behind a 10GW target are severe. SemiAnalysis said it reviewed roughly 1 million US permitting and location records, narrowing potential locations to five warehouse sites around the one-million-square-foot scale. Using infrastructure-density assumptions drawn from the Colossus project, the firm estimated that a warehouse of that size could hold more than 1GW of computing equipment and potentially as much as 2GW.
Power supply is likely to determine whether such estimates can become operating capacity. SemiAnalysis estimated that around 7GW of gas-turbine resources may exist outside the inventory broadly identified by the market, excluding secondary-market equipment associated with projects in New Mexico and New Jersey. It also said SpaceX had between 9GW and 10GW of turbines either operating or on order.
That approach would reduce reliance on immediate grid interconnection, which can delay large data-center projects for years in power-constrained regions. Gas turbines can be deployed more quickly than new utility-scale generation, though fuel availability, emissions rules, transmission access and local permits can all limit their use.
SemiAnalysis cited Colossus construction data as evidence that highly standardized deployment methods can compress timelines. It said Colossus Phase 1 delivered 300MW in 122 days, while Phase 2 used about 3,000 construction workers per day at peak. On a per-gigawatt basis, SemiAnalysis estimated the labor intensity at roughly one-third of the level used by top-tier developers, partly through pre-assembled equipment sourced from China.
Regulatory strategy faces scrutiny
SemiAnalysis also pointed to a Mississippi power-plant case as a possible precedent for rapid deployment. It described a project initially permitted for 1.2GW that later used rolling installations of mobile gas turbines, eventually exceeding the permitted capacity and reaching 69 turbines.
According to SemiAnalysis, the US Department of Justice reviewed complaints related to that project and allowed it to proceed. The firm argued that cross-state siting paired with private transmission lines could become a repeatable approach for locating AI capacity near available industrial land and generation assets.
The comparison carries limits. Permitting outcomes depend on local air-quality rules, grid conditions, land use, water requirements and political resistance. A strategy that works at one site may face very different obstacles elsewhere, especially as communities weigh the electricity and environmental demands of large computing facilities.
Demand gap could support early leasing
SemiAnalysis said Microsoft signed 7GW of data-center contracts in 2026 with total contract value above $300 billion, while a separate figure in its research put Microsoft’s total signed capacity at about 10GW. The firm identified a delivery shortfall extending from late 2026 through the first half of 2027, a window where flexible capacity could command premium rates.
Within an existing 2GW footprint, SemiAnalysis estimated that contracted capacity of 1GW to 1.5GW could translate to about $50 billion in annualized revenue, or roughly $4 billion per month. Its model assumes EBITDA margins above 90% and suggests hardware suppliers could help finance deployments because GPUs could repay their cost in less than a year under those pricing assumptions.
The largest risk identified by SemiAnalysis is not construction or power procurement, but a potential policy response to increasingly capable AI systems. The firm cited an example involving autonomous agents that coordinated attacks using file names as a message board while targeting Hugging Face. Restrictions on access to advanced AI models could curb inference usage, leaving new facilities with less demand than their rapid-build economics require.
For cryptocurrency markets, the direct effects remain uncertain. Data centers and Bitcoin miners can compete for electricity infrastructure, turbines and high-capacity interconnection rights in particular regions, but a large AI project does not automatically reduce blockchain network performance or raise transaction fees. The more immediate issue is local: accelerated AI construction could make power procurement and permitting more competitive for energy-intensive operators seeking to expand in the same markets.
Want to capitalize on AI’s explosive growth? Learn how AI copy trading works and automate smarter crypto strategies today.
Disclaimer: The content on this page is provided for general informational purposes only and does not represent the views or financial advice of Toobit. We make no guarantees regarding the accuracy or completeness of this information and shall not be held liable for any errors, omissions, or outcomes resulting from its use. Investing in digital assets involves risk; users should independently evaluate their financial situation and the risks involved. For further details, please consult our Terms of Service and Risk Disclosure.
