SemiAnalysis: Musk Breaks Through Again, How Will SpaceX Compete for the World's Largest AI Cloud?

TL;DR

· According to SemiAnalysis tracking, SpaceX has raised its 2027 data center scale target to approximately 13GW, with about 3.6GW currently under construction and identified project capacity exceeding 9GW.

· Expansion relies on additional construction at the Colossus campus, reuse of rocket bases, and parallel processing across multiple stages, with some projects expected to take only 4 to 5 months; Tesla may further supply modular mechanical and electrical equipment.

· On-site gas power generation helps projects shorten grid interconnection wait times, while nearly 6GWh of energy storage and public grid access handle subsequent dispatch tasks, though the sustained power supply gap still needs to be resolved.

· Some already-sold compute capacity commands high premiums, but short-term contracts, customer exit clauses, and no-prepayment arrangements make it difficult to directly extrapolate current revenue levels to the long term.

· SemiAnalysis expects SpaceX may support expansion by lowering prices and increasing prepayment ratios; adding over 7GW of additional capacity in 2027 alone could require hundreds of billions of dollars in funding.

· Tightening chip supply and financing difficulties for competitors may support its pricing power, while gas turbine permit litigation and regulatory rule changes will still affect construction plans.

Two months ago, when SpaceX proposed a plan to reach 10GW of data center scale by 2027, the market still had substantial doubts. Now, SemiAnalysis has tracked approximately 3.6GW of capacity under construction across multiple existing and new campuses, and SpaceX has also raised its 2027 expected scale to approximately 13GW.

By SemiAnalysis's comparison, if this target is achieved, it would be equivalent to building in about a year and a half the entire data center capacity Google accumulated from 1999 to 2026. The firm previously believed the 10GW plan could create conditions for over $300 billion in annual recurring revenue (ARR), but this estimate does not equate to locked-in full-year revenue.

Supporting this judgment is a continuously expanding project pipeline and increasingly shortened delivery cycles. SemiAnalysis has identified over 9GW of data center project capacity while tracking over 8GW of potential self-supplied power capacity. However, between project scale, actual built capacity, and stable operational capability, there remain multiple hurdles including power supply, equipment, permits, and funding.

Compressing Timelines to Months, Rocket Base Begins Data Center Construction

SpaceX's most direct advantage is compressing construction work that would normally take a longer cycle into just a few months.

According to SemiAnalysis tracking, the first phase retrofit of Colossus 1 delivered 150MW of capacity in 122 days, while Colossus 2 completed 210MW in 91 days. On this basis, a project codenamed Project SPAM, located within the Colossus 2 campus, broke ground on October 5, 2026, with the goal of bringing the first batch of capacity online by December, potentially shortening delivery time once again.

The MiniHard project in Memphis has a construction cycle of approximately 5 months, and the newly acquired Gun Range site in Mississippi is expected to take about 4 months. By comparison, SemiAnalysis says many peers are still striving to keep construction cycles within 8 to 12 months. If the relevant projects are delivered on schedule, SpaceX will cut this cycle roughly in half.

Construction speed has also extended beyond the campus. Musk is attempting to purchase hundreds of homes around Colossus—on one hand to alleviate community tensions caused by noise, and on the other to provide housing for construction workers.

New projects have also extended beyond the existing Colossus campus. The Starbase in Texas, originally tasked with rocket manufacturing and launch operations, is building two data centers. SemiAnalysis tracking shows that the site went from land grading at the end of August to above-ground structural construction by the end of September, and is subsequently expected to move into interior equipment installation.

Combining building volume, the 2027 construction timeline, and assumptions about NVIDIA Vera Rubin platform deployment, SemiAnalysis estimates that the site's capacity could significantly exceed 1GW. Steel structure erection and roofing construction proceeding in parallel also reflect its approach of compressing total timelines through multiple concurrent work processes.

TeraFab may provide the next batch of capacity. SemiAnalysis says this chip manufacturing project—planned at approximately 100 million square feet with power consumption exceeding 10GW—was initially launched as a data center project before pivoting to wafer fab planning. Local documents and August 2026 official communication records indicate that the campus may deploy computing capacity, and relevant representatives have not ruled out the possibility of leasing to third parties.

However, TeraFab's data center use has not been finalized. On-site engineering could serve one or more data centers, or could support wafer fab ancillary facilities such as gas supply facilities and central power stations. SemiAnalysis leans toward the former interpretation, while also suggesting that the slower main construction progress may be related to ensuring relevant agreements and permits are compliant. If data center use is not approved, the early-graded land can still be repurposed for other uses.

Another Starbase campus in Louisiana has also come under scrutiny. Local meeting minutes did not rule out the possibility of building a data center, and the campus also involves on-site power generation planning. SemiAnalysis said dredging of the waterfront has begun, with heavy road construction expected to start in Q1 2027, and speculated that the site could contribute over 1GW of capacity in 2027. This remains an early-stage project assessment.

Tesla's Supply Chain and Self-Generated Power Get Construction Moving First

Behind the rapid construction is a combination of engineering organization, equipment supply, and power arrangements.

SemiAnalysis estimates that for every 1GW of capacity built by Musk's team, staffing can be kept at around 3,000 people. The firm also believes Tesla may be manufacturing modular mechanical, electrical, and plumbing (MEP) equipment, reducing on-site installation work and external procurement wait times by increasing prefabrication levels.

This approach is similar to some data center developers manufacturing their own switchgear to shorten delivery cycles; however, the SpaceX-Tesla collaboration could cover a larger scale. If this assessment holds, Tesla's role in SpaceX's expansion will extend further from energy storage supplier to data center infrastructure supplier.

Power is being secured on a build-as-you-go basis. SemiAnalysis has tracked over 8GW of potential behind-the-meter power capacity, involving suppliers such as FTAI and Solaris. Behind-the-meter refers to configuring generation facilities on the consumption side to supply power directly to the campus, reducing dependence on new public grid connections.

At the Starbase site in Texas, APR mobile power generation equipment began deployment about one month after main construction started. SemiAnalysis estimates the site may use over 1GW of APR equipment, but given the building scale, this may still be insufficient to meet ultimate demand. Two natural gas pipelines under construction lead the firm to believe the campus's subsequent power capacity could reach several gigawatts.

The site also involves laying fiber optic channels through the Port of Brownsville and leasing waterfront space. Based on this, SemiAnalysis speculates that some of Colossus 2's temporary gas turbines may be relocated to Starbase in the future to support the new campus's launch.

Equipment orders are also progressing. According to information obtained by SemiAnalysis, the first 380MW Doosan gas turbine is in transit, with two more planned for delivery in 2026 and three more in 2027, totaling 1.9GW of nameplate installed capacity. Job postings suggest these units may be intended for the Texas Starbase, but the specific location has not been confirmed.

The institution also inferred from recruitment and project descriptions that SpaceX's gas turbine stationary blade and rotating blade casting project may aim to supply core components with long delivery cycles in-house, then cooperate with complete machine manufacturers. The related cooperation has not yet been announced.

A longer-term path is photovoltaics and energy storage. SemiAnalysis's ChipBook team tracked 1,730 monocrystalline furnace imports and expects the first batch of photovoltaic module installations could be seen as early as the end of 2027, with its calculated full-capacity annual module production capacity at approximately 25GWp.

GWp here measures module peak power; annual module production capacity cannot be directly equated to already-commissioned power generation capacity. The institution further estimates, based on an effective load carrying capability (ELCC, a measure of a power source's contribution to reliable supply) of approximately 20%, that the related deployment could contribute approximately 4 to 6GW of reliable capacity once fully rolled out, but this result depends on actual construction, energy storage configuration, and system operating conditions.

Nearly 6GWh of energy storage cannot replace continuous power sources

As Colossus 2 scales up, the power supply plan is shifting from temporary gas-fired generation to coordination among the public grid, on-site generation, and energy storage.

According to SemiAnalysis citing information from the Tennessee Valley Authority (TVA), Macrohard and MiniHard have applied for 700MW and 450MW of power service, respectively. The institution temporarily assumes that this grid connection can cover the full demand of both facilities.

A 500kV gas-insulated substation near MiniHard is also under construction, planned for completion by December 31, 2026, and connection to the Freeport substation. This timetable is particularly critical for subsequent adjustments to gas-fired units.

SemiAnalysis said that, according to binding requirements signed on July 30, 2026, the Southaven power station currently has an operational nameplate capacity of 2.7GW, but this will drop to 1.24GW by July 2027.

Even if the public grid provides approximately 1.2GW of power, supply and demand may not necessarily match. According to SemiAnalysis's calculations, after deducting grid supply, the remaining IT equipment load is approximately 1.3GW, higher than the 1.24GW generation nameplate capacity at that time. Considering unit derating, operational losses, and power consumption for supporting facilities such as cooling, the gap may exceed 300MW.

This is also why large-scale battery energy storage systems (BESS) are drawing attention. SemiAnalysis said Colossus 2 already has 5.9GWh of energy storage under installation, approaching 6GWh, and may become one of the largest energy storage deployment projects in the world. The institution estimates that SpaceX may be Tesla's largest Megapack buyer, with procurement volume reaching approximately 8GWh under its high-end estimate, and speculates that Tesla may still hold large orders not yet reflected in revenue.

SemiAnalysis envisions that the campus can coordinate gas turbines with energy storage, differentiate power supply priorities between customer service and internal training, and set different service-level agreements (SLAs) for different clusters, thereby improving compute utilization under existing power conditions.

If the energy efficiency metric PUE can be maintained between 1.10 and 1.15, power consumption by supporting facilities can also be kept in check. PUE refers to the ratio of total campus power consumption to IT equipment power consumption; the closer to 1, the lower the share of auxiliary power use.

However, energy storage can shift the timing of power supply and buffer load fluctuations, but it does not itself create electrical energy. If the 300MW shortfall exists continuously over the long term, simply expanding battery capacity cannot eliminate that shortfall; additional generation, surplus power available for charging, or reducing and staggering part of the load would still be needed. Therefore, whether this scheme can hold up still depends on the units' actual output, the load curve, and the dispatch design.

High-priced orders bring cash flow, but expansion still requires hundreds of billions of dollars

Whether construction speed can translate into a profitability advantage also depends on how long customers are willing to pay a premium to obtain compute earlier.

According to SemiAnalysis citing SpaceX Chief Financial Officer Bret Johnsen, the company has sold more than 1GW of capacity at an annualized price approaching $50 million per MW, corresponding to a payback period of less than one year. The market price reference for five-year contracts given by the institution is approximately $12 million to $16 million per MW per year. Because contract terms, exit rights, and delivery conditions differ, this price gap cannot be directly regarded as a premium under equivalent conditions.

SemiAnalysis also said that some customers received GB300 quotes exceeding $15 per hour, but the material did not further clarify the full billing basis for that quote.

The high prices are underpinned by short-term scarcity. Some of SpaceX's current transactions have terms of about 6 months, with customers able to exit with 90 days' notice and no prepayment required; the agreements with Anthropic and Google are irrevocable only until January and March 2027, respectively. Such contracts can quickly monetize scarce compute, but they also mean revenue continuity cannot simply be extrapolated at current prices.

SemiAnalysis expects that more than 50GW of global data center capacity will be added in 2027. Against this supply backdrop, SpaceX needs AI labs to continue exceeding revenue expectations and to keep underestimating their own compute needs in order to sustain strong additional purchases.

The cash burn has already begun. According to the institution, SpaceX had nearly $100 billion in net cash when it launched its cloud business, but this capital is rapidly depleting. Adding more than 7GW of additional capacity for 2027 alone could require hundreds of billions of dollars.

To broaden its customer base and support GPU procurement financing, SemiAnalysis expects that SpaceX may accept lower prices in exchange for higher prepayments—for example, adopting 6-to-12-month contracts, over 50% prepayment, and in a low-end scenario, pricing of approximately $30 million per MW per year. This is the institution's forecast of future contract structures, not a published uniform quote.

Chip supply and the competitive landscape may ease pricing pressure. SemiAnalysis believes that Vera Rubin supply is tightening, and by late 2027 or early 2028, AI chips may replace power and data centers as the primary bottleneck for compute expansion.

Meanwhile, some neoclouds, despite securing NVIDIA allocations, still need to simultaneously find customers, data center space, and financing. As capital markets tighten, they may become more reliant on long-term offtake agreements with hyperscale customers and find it difficult to directly participate in the high-premium on-demand market where SpaceX operates. If this assessment holds, reduced competition and chip scarcity will help SpaceX sustain pricing power for longer.

Federal support favors power supply, but litigation is not over

Gas turbine permits are another hurdle the expansion plan still needs to clear.

According to SemiAnalysis, the NAACP sued xAI in April 2026 under the Clean Air Act, targeting the use of unpermitted mobile gas turbines at the Southaven site powering Colossus 2, and seeking shutdown until permits are obtained.

The institution says the U.S. Department of Justice sought to intervene in June on national, economic, and energy security grounds, citing reasons including Grok's support for defense-related operations, and subsequently referred related procedural issues to the Fifth Circuit Court of Appeals. On October 19, the judge allowed the federal government to participate in the case but did not end the litigation on that basis; on October 30, the judge rejected xAI's claim that vehicle-mounted turbines could be exempt from permitting as "mobile pollution sources." The case still faces a preliminary injunction dispute.

Rulemaking could also change permitting requirements. SemiAnalysis says the COMET coalition of gas turbine manufacturers and users has petitioned the U.S. Environmental Protection Agency (EPA) to classify vehicle-mounted turbines as "non-road engines," thereby changing the applicable permitting framework.

The agency said the proposed rule was sent to the White House for review on October 15; if ultimately adopted and applied as expected, it could benefit xAI, but its legal effect still depends on the final rule and subsequent judicial proceedings.

SemiAnalysis also cited a July 2025 executive order requiring accelerated permitting for data center power supply, as well as Trump's statements on protecting Musk's companies, to conclude that there is political momentum at the federal level to support the project's continued operation. This does not mean gas turbines can ensure continuous operation; community disputes, emissions permits, and court rulings may still affect the power supply arrangements.

SpaceX's logic for competing in the AI cloud market thus becomes increasingly clear: by delivering computer rooms and power faster, it can seize the opportunity to take on customers' urgent additional computing demand, and then use high-priced orders and customer prepayments to support subsequent construction.

But 13GW remains the target. What will determine whether it can become stably operating, continuously revenue-generating capacity is whether power supply continuity, chip deliveries, contract renewals, and financing progress can all keep pace at the same time.

Welcome to join the official BlockBeats community:

Telegram Subscription Group: https://t.me/theblockbeats

Telegram Discussion Group: https://t.me/BlockBeats_App

Official Twitter Account: https://twitter.com/BlockBeatsAsia