What happened
Envision Group, a Chinese company whose core businesses include wind turbines and batteries, says it will finish building the Galaxy Campus data center in Ulanqab, Inner Mongolia, by the end of 2028. Ricky Zheng, general manager of Envision’s AI Data Center business, gave that timeline in an interview with Bloomberg TV reported on October 30, 2026. The company says the campus could eventually reach 3 GW of capacity.
The facility is not entirely prospective. Envision said in August that it had begun operating the Ulanqab data center. The newly reported statement supplies the larger build-out timetable and eventual capacity figure. Ulanqab is roughly 350 kilometers northwest of Beijing and has abundant renewable-energy resources, according to the report. Envision’s existing position in renewable energy is central to the pitch: it is entering AI infrastructure through power and supporting systems rather than through ownership of the computing workloads themselves.
That division of responsibility matters. Envision says it will provide the power and associated infrastructure, while technology-company tenants lease the facilities and install their own chip clusters. In other words, the company is describing a colocation-style operating model rather than a cloud service that processes customers’ data. Zheng said Envision has lined up offtake agreements through 2029 or 2030, but the report does not name counterparties, disclose contracted capacity, or specify the terms of those agreements.
The intended tenant base extends beyond hyperscalers. Zheng said Envision is talking with companies in autonomous driving and content generation that are seeking leases, alongside larger technology firms. He attributed their demand to business growth from using AI. The company also says it plans further data centers in Inner Mongolia and other parts of northwest China. Longer term, Zheng said Envision is looking outside China, with an eventual global target of as much as 10 GW and potential locations in Latin America, Southeast Asia, Europe, and Africa where renewable resources are available.
The announcement also illustrates a change in who is seeking a role in the AI buildout. The report notes that such facilities have historically been developed mainly by specialist developers, telecom operators, and internet companies. Envision says sector momentum has helped it obtain what Zheng characterized as “very fair” financing rates. That is a company characterization, not a disclosed financing cost or a completed financing transaction.
What the archive adds
This is a new fact on a known infrastructure story. Financing the AI Buildout examined how financing determines where capacity and demand risk sit. Datacenters 2027: The Year of Acceleration, +50GW in a Single Year covered a forecast of more than 50 GW of capacity coming online in 2027; Envision adds one company’s stated 3 GW campus plan and 2028 deadline.
Trend signal
No supplied pattern directly fits this announcement. It concerns the physical supply of power and colocation capacity, rather than a demonstrated shift in agent architecture, software capabilities, security, or organizational practice.
Why now
Envision said the Ulanqab site began operating in August, while Zheng’s October interview added an end-2028 completion target and a 3 GW eventual scale. The timing therefore turns an operating facility into a more specific multi-year expansion plan. Its claimed offtake runway through 2029 or 2030 also frames the buildout as a response to expected demand rather than an unallocated speculative site, although the underlying agreements are not public in the supplied material.
Connections to other events
The campus plan connects to Financing the AI Buildout, which focused on the allocation of leverage and demand risk in infrastructure expansion. Envision’s claimed access to favorable financing and its decision to supply facilities rather than operate customer compute place those issues at the center of this project. It also sits alongside the scale implied by Datacenters 2027: The Year of Acceleration, +50GW in a Single Year: a 3 GW campus is meaningful, but it is one announced project within a much larger projected global capacity buildout.
Why it matters
For AI builders, electricity, cooling, sites, and financing increasingly constrain where large training and inference clusters can be placed. Envision’s stated model separates those constraints from the tenant’s model and chip choices: a customer would bring its own cluster while leasing an energy-backed facility. That can broaden the pool of infrastructure providers available to companies that need large deployments, especially where renewable energy is plentiful.
The practical implication is less about a new developer service than about the structure of future capacity procurement. A tenant in Envision’s proposed model would still need to source chips, design and operate its cluster, connect it to its systems, and establish the commercial and reliability terms required for production workloads. The provider’s role is the powered physical environment and related support systems. That separation can matter to organizations that want control over their computing stack while avoiding the cost and time of developing a data-center site themselves.
But the announcement is not evidence that capacity is available to a particular team today. It gives no tenant names, present usable capacity, chip availability, pricing, service-level commitments, network characteristics, or allocation process. Teams evaluating capacity should treat it as a market-supply signal and continue to diligence actual power delivery, commissioning dates, lease terms, and operational resilience.
Limits and uncertainty
The reported plan rests on statements from an Envision executive. The material does not establish how much of the eventual 3 GW is built, energized, contracted, or available for tenant use today. “Offtake agreements” are described without counterparties, volumes, durations, or conditions. It also does not say whether the stated 10 GW global ambition has a schedule, committed funding, identified sites, or permits.
The report gives no independent validation of construction progress, renewable-power sourcing, financing terms, or tenant demand. It does not define whether the 3 GW figure refers to site power, IT load, or another capacity measure, which limits comparisons with other data-center projects. Nor does it establish how the site will handle grid connection, cooling, network connectivity, hardware supply, or the operational requirements that determine whether nominal capacity becomes useful compute.
Reader takeaway
Envision’s 3 GW, end-2028 target is a sizable announced addition to China’s AI infrastructure pipeline, with a tenant-owned-chip-cluster model backed by a renewable-energy company. It is useful evidence of expanding supply ambitions and of energy companies entering colocation, not proof of delivered capacity, public lease availability, or accessible compute.