SpaceX’s ambitions in artificial intelligence are undergoing a significant evolution in southeastern Arizona, where an affiliated entity has proposed what may become one of the world’s largest dedicated supercomputing and AI infrastructure campuses. This 747-acre development in Cochise County features a substantial projected electrical demand of up to 6 gigawatts, to be fulfilled across four construction phases.
If successfully executed, the project would represent an unprecedented concentration of computing infrastructure, necessitating millions of advanced accelerators, extensive electrical grid upgrades, large-scale energy storage, and a robust supply chain capable of hardware procurement at a scale rarely attempted.
However, there is a fundamental distinction between the proposal of a 6 GW supercomputing facility and the practical capability to construct, power, and equip such an installation. This reality is central to the application submitted by Tombstone Resources LLC, a SpaceX-affiliated entity operating as SpaceXSI. The proposal outlines a development near Cochise, Arizona, comprising up to nine buildings of approximately 493,000 square feet each, phased into four 1.5 GW stages.
This project raises questions that extend beyond a single technology firm's objectives, underscoring how the global race for AI infrastructure is increasingly testing the limits of regional power systems, hardware manufacturing capacity, transmission infrastructure, and environmental resources.
Six gigawatts is an extraordinary computing commitment
To put the proposed scale into perspective, 6 GW represents approximately 19 percent of Arizona's stated 32 GW electrical grid capacity. The project's projected demand would also be roughly 2.5 times the combined demand attributed to data centers across Maricopa County in the reporting cited by the vetting brief.
The development would be positioned adjacent to the Apache Generating Station, an existing power complex with approximately 626 MW of coal, natural gas, and solar generation capacity. The proposal envisions expanding the surrounding energy infrastructure to support a campus whose ultimate demand could exceed 6 GW.
The difference in scale is remarkable.
This is not simply another hyperscale data center expansion. It is a proposal to establish an enormous computing destination whose electrical requirements would rival those of major industrial regions.
The application identifies approximately $3.6 billion in grid infrastructure upgrades, including an initial phase estimated at $500 million. Construction could begin in March 2027 if the required zoning approvals are obtained.
However, the proposed electrical infrastructure is not the same as secured, operational computing capacity. The application describes an ambitious development schedule, but the availability of the necessary generation, transmission equipment, construction resources, and computing hardware remains a central question.
For the supercomputing industry, that distinction matters.
A building capable of housing millions of processors is not a supercomputer. The actual computing system depends on accelerator availability, memory capacity, high-speed networking, storage, power distribution, cooling, and the software infrastructure required to operate the machines efficiently.
At 6 GW, every one of those requirements becomes a major engineering and procurement challenge.
The GPU question: Where do millions of accelerators come from?
One of the most revealing aspects of the proposal is what it does not disclose.
The application does not identify the intended computing workloads, a specific customer, an accelerator supplier, or the hardware architecture that would populate the campus.
That omission makes it difficult to independently evaluate the project's ultimate computing capacity.
A useful way to understand the scale is to perform an illustrative hardware calculation.
Using the approximately 1.9 kW per GPU figure cited in the vetting brief for NVIDIA's GB300-class hardware, a 6,000 MW IT load would correspond to approximately 3.16 million GPUs.
That is an order-of-magnitude illustration, not a procurement forecast. Future accelerator generations will have different power characteristics, and the actual facility electrical demand would also depend on cooling, networking, storage, and other infrastructure loads.
Nevertheless, the calculation exposes the extraordinary scale of the proposed undertaking.
Millions of accelerators would require an equally substantial supporting infrastructure. High-bandwidth memory, advanced packaging, high-speed interconnects, network switches, optical connectivity, storage systems and power-delivery equipment would all have to arrive on a coordinated schedule.
The challenge is not merely purchasing GPUs.
It is assembling a complete, operational computing platform at a scale that could strain multiple segments of the semiconductor and data-center supply chains.
SpaceXSI has not publicly identified the accelerator procurement strategy for the Cochise development. Nor has it disclosed a workload roadmap explaining whether the campus would support large-scale model training, inference, AI cloud services, or a combination of computing operations.
Without those details, the 6 GW figure describes an infrastructure ambition rather than a verified measure of future computing performance.
A 24 GWh battery cannot manufacture electricity
The proposal also includes a 24 GWh battery energy storage system intended to stabilize power delivery and reduce the impact of the campus on the surrounding electrical grid.
That is a substantial installation, but its role needs to be understood correctly.
At a continuous 6 GW draw, a fully charged 24 GWh battery would theoretically provide approximately four hours of electricity before depletion, assuming the entire rated capacity were available and ignoring conversion losses, operating reserves, and other limitations.
That makes the battery a potentially valuable grid-management component. It does not, however, represent an independent source of sustained electricity.
The distinction is important because a computing campus operating at this scale requires dependable power around the clock.
Battery storage can help manage fluctuations, provide short-duration backup, and smooth the interaction between generation and demand. It cannot replace the generation and transmission infrastructure needed to supply a continuous multi-gigawatt computing load.
The project's ultimate success will therefore depend on whether its proposed electrical upgrades and associated generation resources can be delivered on schedule.
There is also a financial question: who bears the cost if the campus is delayed, substantially reduced in size, or never reaches its proposed capacity?
The application identifies potential electricity-bill reductions of 10 to 40 percent for members of Sulphur Springs Valley Electric Cooperative. Those figures are applicant projections, not independently established outcomes, and are subject to regulatory approval.
The Arizona Corporation Commission's review of the proposed Energy Services Agreement will be an important test of the project's financial structure and protections for existing electricity customers.
The water issue is bigger than the cooling system
SpaceXSI's application emphasizes a dry-cooling design that would avoid conventional cooling towers and evaporative spray during normal operation.
The company estimates annual direct water consumption of approximately 7 to 10 acre-feet at full build-out, primarily for domestic use by facility personnel. It also states that no new groundwater well would be required for industrial cooling.
On the surface, those figures appear relatively modest for a development of this size.
But direct facility water consumption is only part of the equation.
The proposed campus would require enormous quantities of electricity. Generating that electricity can also consume water, depending on the generation technologies, cooling systems, and fuel sources involved.
Reports cite estimates that power generation can increase the overall water footprint associated with data centers by 60 to 90 percent. Yet the application does not quantify the indirect water consumption associated with supplying the full 6 GW campus.
That is a significant information gap.
The proposed development is located in the Willcox groundwater basin, one of Arizona's most severely depleted groundwater regions.
According to the Arizona Department of Water Resources, groundwater declines have reached approximately 400 feet in parts of the basin between 1940 and 2015. An estimated 5.7 million acre-feet of groundwater has been removed from storage.
The state's hydrologists have warned that even if pumping stopped entirely, recovery could take more than 280 years.
The issue is therefore not simply whether the data center itself uses a relatively small amount of water. It is whether the full energy infrastructure required to operate the campus can be supported without creating additional pressure on an already stressed regional resource.
Dry cooling addresses one part of that challenge. It does not answer the larger question.
The Apache power connection deserves scrutiny
The proposed campus would be located beside the Apache Generating Station, an energy complex whose existing generation includes coal, natural gas and solar resources.
The station had previously been expected to retire by 2028, but its operating outlook changed following federal support for upgrades.
Reports identify approximately $21 million in federal funding intended to support continued operation.
That creates an interesting contrast with the project's presentation as a modern, battery-supported, dry-cooled AI infrastructure development.
The proposed computing campus would be built alongside an energy complex whose continued operation includes fossil-fuel generation.
This does not establish the project's eventual emissions profile. That will depend on the actual generation mix, power contracts, grid arrangements, and any additional infrastructure ultimately constructed.
But it does raise an important question for an industry increasingly making claims about efficient and sustainable AI infrastructure: What is the real energy and environmental footprint of the computing systems once the electricity required to operate them is accounted for?
The answer cannot be determined from the campus's cooling design alone.
Is this the next stage of SpaceX's AI infrastructure ambitions?
The Cochise proposal arrives amid SpaceXSI's broader expansion of AI computing infrastructure associated with xAI and Grok.
The company's existing Memphis-area computing operations have been described as exceeding 1 GW of capacity, although the precise scale has been disputed in outside reporting.
The Arizona proposal would represent a substantial escalation in ambition.
However, there is an important distinction between the Memphis operations and the Cochise application. The Arizona filing does not identify the campus as a new Colossus facility, does not establish a direct connection to the Colossus 2 expansion, and does not disclose a specific customer or workload.
It should therefore be treated as a separate proposed development rather than automatically labeled Colossus 3.
The broader strategic direction is nevertheless apparent: access to large-scale computing capacity is becoming increasingly dependent on the ability to secure dedicated energy infrastructure.
For AI developers, owning or controlling the electrical supply can become just as strategically important as securing advanced accelerators.
But electricity alone does not produce competitive computing performance. The quality of the system will ultimately depend on how efficiently the power is converted into useful computational work.
That means accelerator utilization, memory bandwidth, network performance, storage throughput, thermal management, and software efficiency will determine whether an enormous facility delivers corresponding computing value.
A 6 GW electrical commitment is not a benchmark.
The December hearing is the first major test
The Cochise County Planning and Zoning Commission is scheduled to consider the proposal on December 9, 2026, at 2 p.m. in Bisbee.
Written comments intended for inclusion in the hearing packet are due November 27 at noon.
The county has explicitly stated that the project figures are based on information supplied by the applicant and have not been independently verified or endorsed by the county.
That qualification is essential.
The proposed 1,000 jobs, projected tax benefits, electricity-bill reductions, infrastructure costs, and construction schedule should all be treated as applicant claims until independently substantiated.
The zoning hearing is also not the only regulatory consideration. The proposed energy agreement will face a separate review by the Arizona Corporation Commission.
The project therefore has multiple milestones ahead before its proposed computing capacity can become an operational reality.
The real benchmark is delivery, not gigawatts
The SpaceXSI proposal exemplifies the extraordinary scale of the infrastructure competition currently emerging within the artificial intelligence trade. Establishing a 6 GW supercomputing campus necessitates a level of coordination across power generation, grid transmission, semiconductor manufacturing, and hardware integration that remains largely unprecedented.
While this project serves as a compelling case study regarding the integration of large-scale computing and energy infrastructure, the industry's tendency to quantify ambition in gigawatts should not overshadow critical engineering and operational uncertainties. Fundamental questions remain regarding the procurement of millions of accelerators, the definition of target workloads, the actual conversion efficiency of electrical capacity into computing power, and the long-term financial accountability for grid upgrades. Furthermore, the project must address substantial risks concerning construction timelines and the indirect water consumption necessitated by such a massive energy demand.
Ultimately, these are not peripheral concerns but are essential to determining the project's viability. Cochise County currently holds a development proposal, not a functioning supercomputing installation. The next generation of AI infrastructure will not be defined by the magnitude of power commitments, but rather by the capacity to transform energy into reliable, efficient, and economically sustainable computing performance. For SpaceXSI, the 6 GW ambition represents only the initial phase of this rigorous challenge.