AnswersPublished27 sources
Envision’s Inner Mongolia AI Campus vs. Musk’s Orbital Data Centers
Envision has commissioned a 120 MW first phase of a renewable powered AI campus in Inner Mongolia, designed to scale to 2 GW. SpaceX has sought approval for up to 1 million solar powered satellites and says initial AI satellites could launch in late 2027, with significant scale targeted for 2028.
Published byEdited with GPT-5.6 LunaImages generated with GPT Image 1.5
Research answer
Key takeaways
- Envision has commissioned a 120 MW first phase of a renewable powered AI campus in Inner Mongolia, designed to scale to 2 GW.
- SpaceX has sought approval for up to 1 million solar powered satellites and says initial AI satellites could launch in late 2027, with significant scale targeted for 2028.
- The central difference is not renewable energy versus solar energy: Envision moves compute to abundant power on Earth, while SpaceX would generate power and operate computing hardware in orbit.
AI Prompt
openai.comCreate a landscape editorial hero image for this Studio Global article: How does Envision Group’s operational and planned renewable-powered AI infrastructure in Inner Mongolia challenge SpaceX CEO Elon Musk’s pro. Article summary: Envision’s Inner Mongolia project is a practical counterexample to the claim that AI’s electricity constraint requires moving compute into orbit: it places large-scale AI hardware directly beside low-cost renewable gener. Topic tags: general, news, general web, user generated, government. Style: premium digital editorial illustration, source-backed research mood, clean composition, high detail, modern web publication hero. Use reference image context only for broad subject, composition, and topical grounding; do not copy the exact image. Avoid: logos, brand marks, copyrighted characters, real person likenesses, fake screenshots, UI text, readable text, watermar
The electricity bottleneck facing AI does not have only one answer. Envision Group’s Galaxy Campus in Ulanqab, Inner Mongolia, has begun operating with a 120 MW initial phase and is designed to scale beyond 2 GW using directly connected renewable power. 1
3 SpaceX, by contrast, has proposed a constellation of up to 1 million solar-powered orbital data-center satellites, but that system is not yet operational.
18
19
That makes Envision a useful real-world counterexample to the idea that AI compute must eventually leave Earth to escape grid constraints. It does not settle the long-term contest: terrestrial projects must manage variable renewable generation and land-based infrastructure, while orbital systems must solve launch costs, heat rejection, radiation, networking and hardware replacement.
Two ways to solve AI’s power problem
Envision’s approach is geographic. Instead of building a large data center near a heavily constrained demand center, it places compute in a region with substantial wind and solar resources, cooler conditions and available land. The Galaxy Campus integrates renewable generation, transmission infrastructure, energy storage and intelligent power management. 3
6
SpaceX’s proposal is architectural. Its filing describes solar-powered satellites in low Earth orbit that would operate as an interconnected orbital data-center system. The company argues that near-continuous solar exposure could reduce dependence on terrestrial grids and lower operating costs. 18
20
In simple terms:
- Envision moves the servers to the electricity.
- SpaceX moves the servers and the electricity system off the planet.
Both concepts use solar-derived energy, but their engineering and execution risks are very different.
How the projects compare
| Dimension | Envision Galaxy Campus | SpaceX orbital data centers |
|---|---|---|
| Location | Ulanqab, Inner Mongolia, on Earth and near renewable resources. |
Low Earth orbit, using a proposed satellite constellation. |
| Status | The initial 120 MW phase has been commissioned; the larger 2 GW capacity is planned. |
Proposed, with regulatory approval and large-scale deployment still ahead. |
| Scale | Designed to support up to 2 GW of power capacity and as many as 1 million AI accelerators once fully built out. |
SpaceX has applied for up to 1 million satellites, though the filing does not establish that all would be deployed or define a completed compute capacity. |
| Energy source | Land-based wind and solar, combined with storage and power-management systems. |
Solar arrays in orbit, with near-continuous exposure in suitable orbital configurations. |
| Cooling | Uses familiar terrestrial data-center cooling systems, with the region’s cool climate potentially reducing cooling demand. |
Must reject heat through radiators because a vacuum provides no air convection; this remains a central technical challenge. |
| Maintenance | Accessible servers, facilities and power equipment can be serviced or replaced on Earth. | Repairs and component replacement would require difficult and costly space operations. |
| Deployment timeline | An initial phase is already operating. |
Musk has said first AI satellites could launch in the fourth quarter of 2027 and reach “significant scale” in 2028. |
Why Envision challenges Musk’s premise
The strongest challenge is practical rather than theoretical. Envision shows that constrained grid capacity can be addressed by changing where compute is built, not necessarily by changing the planet on which it operates.
A renewable-powered campus can be developed around existing terrestrial technologies: standard servers, conventional networking, storage systems, transmission equipment and maintenance operations. Envision’s project is already using that model at an initial scale, even though its full capacity remains a future build-out. 1
3
This matters because the immediate constraint for many AI developers is not the global availability of energy. It is the difficulty of securing enough power at the right location and connecting a new data center to the grid. Envision’s design attempts to avoid that bottleneck by integrating generation and compute in the same infrastructure project. 3
6
SpaceX proposes a more radical solution: bypass terrestrial interconnection entirely. But that benefit comes with a new chain of dependencies—launch cadence, satellite manufacturing, orbital networking, thermal control, radiation protection and eventual replacement of computing hardware.
Reliability: intermittency on Earth versus complexity in orbit
Envision’s renewable supply is not automatically equivalent to uninterrupted power. Wind and solar output varies, so the campus’s reliability depends on the performance and scale of its storage and power-management systems. Public information supplied for this comparison does not establish the site’s firm 24/7 renewable capacity, backup arrangements or outage performance. 3
6
Orbital solar could offer more consistent access to sunlight, but continuous sunlight does not eliminate reliability problems. Space-based hardware must operate in a radiation environment, communicate across a distributed network and survive without ordinary physical access. Experts have identified cooling, launch logistics, harsh conditions and the rapid evolution of AI chips as unresolved issues for orbital data centers. 46
47
48
Cooling is especially counterintuitive. Space is cold, but a vacuum does not carry heat away through air. Ground data centers can use air, water or other cooling infrastructure; an orbital satellite must radiate waste heat away as infrared energy. 45
46
Cost: an operating-cost promise versus a deployment-cost burden
Envision’s economic argument rests on several terrestrial advantages: abundant local renewable generation, reduced dependence on a congested grid connection and potentially lower cooling requirements in a cool region. However, the supplied evidence does not establish a verified all-in cost per kilowatt-hour or per unit of AI compute for the campus.
SpaceX’s economic case depends heavily on future launch costs. The European Parliament identifies launch economics as the main barrier to space-based data centers and notes that current reusable-launch costs remain on the order of several thousand euros per kilogram. 43
One non-peer-reviewed analysis illustrates the sensitivity of the calculation. Assuming a future launch cost of $200 per kilogram, it estimates that putting 1 GW of AI compute into low Earth orbit could require 10,000 tonnes of satellites, roughly 50 Starship V3 launches and $2 billion in launch costs alone. These are model assumptions, not achieved SpaceX prices or a complete estimate of project costs. 44
Neither approach has a settled lifetime cost. For Envision, the unanswered variables include:
- accelerator and server prices;
- storage duration and replacement costs;
- transmission and network expenses;
- utilization rates; and
- the delivered price of renewable electricity.
For SpaceX, the list is longer. It also includes satellite mass, launch cadence, solar arrays, radiators, radiation protection, communications, failures, insurance, replacement launches and the risk that AI chips become obsolete before satellites can be upgraded. Analysts have specifically warned that a fixed orbital design could struggle with the rapid pace of chip development. 46
47
Timelines favor Envision—but only for the near term
Envision’s first 120 MW phase is already in operation, while the planned 2 GW scale has not yet been completed. 1
3 That distinction is important: the project demonstrates deployment, but not yet the performance or economics of a fully built multi-gigawatt campus.
SpaceX’s timeline is more conditional. Musk has stated that initial AI satellites could launch in late 2027 and reach significant scale in 2028. 17 Those targets depend on regulatory approvals, manufacturing, launch readiness and the ability to sustain a high-cadence, low-cost deployment program.
The comparison is therefore between a terrestrial project with an operating first phase and an orbital architecture whose most ambitious benefits remain prospective.
What Inner Mongolia says about China’s computing strategy
Envision’s campus fits a broader terrestrial pattern in which AI infrastructure is concentrated in regions with plentiful energy, land and favorable cooling conditions. Reporting describes Ulanqab and Inner Mongolia as growing centers for AI and green-computing projects, with other technology companies also operating or considering facilities in the region. 10
37
That pattern treats energy geography as a computing advantage: build large clusters where renewable power is available, then connect those clusters to users and demand centers. The evidence supports Inner Mongolia’s importance as an AI-infrastructure location, but it is not sufficient to quantify Envision’s precise role in China’s national computing strategy or to attribute wider policy outcomes to the Galaxy Campus alone.
The biggest unknown is future demand
The economics of both models depend on what AI workloads look like several years from now. Training and inference have different infrastructure requirements; model efficiency could reduce the amount of compute needed for a given task; and accelerator utilization will determine how much capacity generates revenue.
Rapid hardware improvements create a further tension. A terrestrial data center can generally refresh servers through established supply chains. An orbital system may have to keep hardware in place for much longer, making it vulnerable to the falling cost and rising performance of newer chips. 46
The result is that nominal capacity—2 GW on Earth or a proposed million-satellite constellation in space—does not by itself show which model will deliver cheaper or more reliable AI compute. The meaningful comparison will require lifecycle costs, utilization, uptime, network performance and hardware-refresh economics.
Bottom line
Envision does not disprove the possibility of orbital AI computing. It does challenge the stronger claim that Earth-bound electricity constraints leave no scalable path forward.
The Inner Mongolia model offers a relatively familiar solution: move compute to renewable power, integrate storage and power management, and use terrestrial infrastructure that can be maintained and upgraded. SpaceX’s model could eventually gain an advantage from near-continuous solar exposure and freedom from terrestrial grid bottlenecks, but it must first prove that launch, cooling, radiation protection, networking and hardware replacement can work economically at scale.
For now, Envision represents a deployed terrestrial answer to AI’s power problem. SpaceX represents a high-upside but high-risk bet that the economics of operating computers in orbit will improve faster than the practical alternatives on Earth.
Studio Global AI
Continue your research
This page includes a source-backed answer you can continue inside Studio Global.
People also ask
What is the short answer to "Envision’s Inner Mongolia AI Campus vs. Musk’s Orbital Data Centers"?
Envision has commissioned a 120 MW first phase of a renewable powered AI campus in Inner Mongolia, designed to scale to 2 GW.
What are the key points to validate first?
Envision has commissioned a 120 MW first phase of a renewable powered AI campus in Inner Mongolia, designed to scale to 2 GW. SpaceX has sought approval for up to 1 million solar powered satellites and says initial AI satellites could launch in late 2027, with significant scale targeted for 2028.
What should I do next in practice?
The central difference is not renewable energy versus solar energy: Envision moves compute to abundant power on Earth, while SpaceX would generate power and operate computing hardware in orbit.