The latest evidence sharpens the delivery question.
The September Xinhua Finance account identifies the Dongying 500kV project as a dependency for VNET’s Ulanqab development. Its potential 1,160MW IT power-supply figure is not commissioned or contracted capacity. China’s computing–electricity policy increases the strategic relevance of that link, but grid commissioning, installed customer hardware and paying utilisation remain the decisive evidence.
VNET / INVESTMENT THESIS · UPDATED 17 OCTOBER 2026
A data-centre company in the middle of a transformation
The most revealing way to understand VNET today is to begin with the business it has built, not the shareholder who arrived in September. Over several years, the company has moved steadily toward large wholesale data-centre campuses while investing in the networks, power systems and operating capabilities needed to serve increasingly demanding AI customers. CATL's arrival makes that transition more interesting, but it did not create it.
The financial consequences are less settled. Building a bigger estate is not the same as earning a satisfactory return on it, and the proposed western projects bring questions about customer commitments, electricity supply and financing into sharper focus. The opportunity lies in whether VNET can turn its existing relationships and technical work into a business that captures more value without shouldering every yuan of construction cost.
The transformation was underway before CATL arrived
The clearest evidence of change is already operating. At the end of June 2026, VNET reported 1,007MW of wholesale capacity in service, with 744MW utilised and a 96.3% contracted rate. That installed estate is different in kind from the gigawatt-scale projects announced later in the year: it shows that VNET has built facilities and attracted substantial customer commitments, even as the economics of the next wave remain to be established.
The company has also been working on the harder question of how such facilities operate. Its source-grid-load-storage ideas predate the CATL transaction, from the SPEAR demonstration concept in 2021 to the later Powerlet architecture. In July 2026, VNET helped lead an open technical report on gigawatt-scale AI data centres. The significance is not that every proposed technology is commercially proven; it is that VNET has spent years developing an operating approach in which electricity, cooling, interconnection and computing demand are designed together.
Q2 2026 operating disclosure · VNET's compute-energy development history · GW-scale technical report
A customer network that is becoming an infrastructure ecosystem
VNET's relationships already reach across China's major technology platforms. Its investor materials name Alibaba, Tencent, JD.com, Kuaishou, Kingsoft Cloud, Huawei and ByteDance among wholesale customers. Those names establish breadth, not the identity of the undisclosed buyer behind any particular order. The more interesting evidence lies in the physical and technical connections that can be traced beyond a presentation slide.
Huawei illustrates why the customer list alone is inadequate. It appears in VNET's orbit as a wholesale customer, equipment supplier and technical collaborator. Those roles need not collapse into one commercial contract, but together they show how VNET can sit between cloud demand, network architecture and the physical data-centre estate.
That customer base makes VNET a credible participant in large projects, and long-term commitments could help lenders assess the next generation of campuses. Yet the public record does not identify who will occupy the roughly 3GW proposed in Qingyang and Zhongwei. Until specific contracts and delivery schedules emerge, it would be a mistake to assign those megawatts to ByteDance, Huawei or any other familiar name.
What the ByteDance procurement record actually shows
A procurement trail in Ulanqab, Inner Mongolia, gives the ByteDance relationship a physical location. A 29 June 2026 design tender and a 31 July construction notice name a “ByteDance (Envision–VNET) transmission duct and optical-fibre construction project”. The construction notice describes approximately 1.2 kilometres of ducts, 32 runs of 288-core optical fibre and roughly 64 kilometres of fibre installation, with self-funded project expenditure of RMB18.38 million. These are unusually specific details: not simply a logo in an investor presentation, but a named transmission infrastructure project involving the three organisations.
It is also important to say what the documents do not show. They do not establish which company will own or operate the finished fibre, the commercial division of responsibilities, whether ByteDance has reserved VNET data-centre capacity at the site, or how many MW—if any—should be attributed to ByteDance. The tender evidence is a meaningful connection in the network layer, not a licence to allocate VNET's unnamed wholesale orders.
Original procurement notices: 29 June design tender · 31 July construction tender.
Huawei and VNET: from supplying equipment to testing distributed AI networks
The relationship between VNET and Huawei is more revealing when read as a sequence of projects rather than a partnership announcement. Huawei first appeared in VNET’s infrastructure through power equipment, including its SmartLi UPS technology. In October 2024, the companies expanded the relationship into integrated networks, green computing and the foundations of a distributed computing system. By 2026, the work had become substantially more specific: Huawei published a case study describing a jointly developed AI-native networking solution with VNET and DYXnet, VNET’s network-services subsidiary.
The June 2026 Huawei case study provides something unusually valuable in a sector dominated by architectural promises: reported test results. In a two-node configuration separated by approximately 10 kilometres and linked by dual 400 Gbps connections, Huawei reported a training-performance difference of less than 2% against its same-node reference, and model-distribution performance improvements of five to eight times under the stated test conditions. These are vendor-published results, not an independently audited benchmark, and they should not be extrapolated to national-scale scheduling. But they show the companies working on the practical problem of making physically separate computing resources behave more like a connected system.
That effort also has a public industry dimension. At the AIDC Industry Development Conference in Wuhu on 15 September 2026, Huawei presented its 3D data-centre architecture while VNET executive Yan Kun discussed “Hyperscale 2.0” and autonomous AIDC operations. Their appearance in the same conference programme is not, on its own, proof of a jointly delivered campus. Read alongside the published case study, however, it shows that the technical collaboration sits within a broader shift in how both companies think about AI infrastructure: beyond buildings and racks, toward power-aware, connected and increasingly automated systems.
For VNET, the significance is commercial as well as technical. A neutral data-centre and network operator that can interconnect AI clusters, distribute models and integrate heterogeneous hardware may be able to earn more than rent on powered space. Huawei could also capture much of the networking, accelerator and software value itself. The question is therefore not whether the companies know one another; the evidence establishes that they do. It is whether their technical work leads to paying customers, separately identifiable network or orchestration revenue, and a durable economic role for VNET.
Sources: Huawei's June 2026 joint VNET–DYXnet case study · Wuhu conference programme · Conference reporting · VNET–Huawei relationship archive. Test results are Huawei's reported figures.
VNET customer portfolio, 4Q25 investor presentation · 2026 orders and utilisation
Ulanqab: the difference between a campus and a powered computing system
A September Xinhua Finance report offers a practical counterweight to the language of large-scale AI capacity. VNET's chief engineer at its Ulanqab campus said the first phase was operating and the second was advancing, but pointed to the unfinished Dongying 500kV electricity project as an important next step. The report described a potential 1,160MW of IT power-supply capacity across the development area at full build-out. That is a development-scale figure, not evidence of 1,160MW energised, occupied or generating revenue.
The distinction has become more consequential as China's leadership places suàn diàn xiétóng (算电协同), coordination between computing and electricity, within the AI Plus agenda. The 16 October State Council account reinforces a policy direction already underway; it does not mean the Dongying line has been commissioned. For VNET, the next measurable milestone is not another campus capacity announcement but the availability of grid power, followed by equipment installation, contracted use and cash returns.
Why Qingyang and Zhongwei matter
Qingyang, Gansu: a western computing hub, not simply another data-centre address
Qingyang sits in Gansu province in north-western China and is part of the national East Data–West Compute effort to place large computing workloads closer to western land and energy resources. VNET announced a GW-scale compute-and-energy base there in September 2026. This matters because the proposition is not just server halls: it envisages coordinated computing capacity and electrical infrastructure at campus scale. The public materials reviewed do not establish the final asset-owning entity, the customer contracting the capacity, the construction financing or the electricity supply contracts. Qingyang's broader regional power and computing projects should not be represented as VNET-owned assets or committed VNET energy supply.
Zhongwei, Ningxia: a more specific three-party energy agreement
On 11 October 2026, VNET signed a strategic cooperation framework agreement with the Zhongwei municipal government and China Nuclear Huineng Ningxia New Energy (中核汇能宁夏新能源有限公司). Zhongwei is another national computing hub in north-western China, at the intersection of important electricity and communications routes. VNET described a proposed GW-scale AI data-centre and Powerlet MicroGrid project on approximately 2,000 mu of land (about 133 hectares), combining very large computing loads with energy storage and active balancing.
The energy partner is not a decorative name. China Nuclear Huineng Ningxia New Energy is explicitly a signatory to the framework, placing an energy-sector counterparty alongside the municipality and VNET. The proposed architecture integrates generation, grid connection, storage, AC/DC distribution and intelligent load management, potentially extending from grid interfaces to MW-scale racks. But a framework agreement is not a signed power-purchase agreement, project-finance commitment or construction completion notice. Its commercial significance depends on who owns the power assets, sells electricity, manages balancing and retains any savings or market revenue.
VNET says the Qingyang and Zhongwei announcements together represent roughly 3GW of proposed new GW-scale projects. That is an indication of ambition and site pipeline, not 3GW of delivered revenue-generating capacity. Read VNET's Zhongwei announcement (Chinese).
The investors behind Green Alchemist—and what CATL stands to gain
Who is Green Alchemist?
The buyer behind the September stake transfer is unusually consequential. PJ Millennium acquired 650,424,192 VNET ordinary shares, approximately 38.1% of the company, from existing holders. Behind the acquisition sits a partnership structure involving Green Alchemist, identified in the underlying research with Zeng Yuntian, son of CATL founder Robin Zeng. The economic ownership of that partnership and the governance rights exercised through its general partner are separate questions; the SEC filings describe Lochpine Capital's ownership, including CATL's 45% interest. The transaction brought a strategically connected shareholder group into VNET, but because it was a secondary sale it did not itself fund new construction.
The distinction matters. Founder-family exposure may give the investment a longer strategic horizon than a routine portfolio purchase, but it does not make VNET a CATL subsidiary or establish that the battery manufacturer controls every decision of the buyer. The legal structure, the economic interests and the direct CATL–VNET cooperation agreement each tell a different part of the story.
CATL and its founder-family investors have reasons to want the investment to work. VNET's campuses could become customers and reference installations for storage, high-voltage DC systems and intelligent electricity management. A smoothly delivered, well-financed estate would also make those products easier to demonstrate at scale. None of that amounts to a disclosed obligation to finance Qingyang or Zhongwei, and equipment suppliers can prosper even when a data-centre operator's shareholders earn disappointing returns.
Primary documents: Initial Schedule 13D and Lochpine ownership · September 13D amendment · Green Alchemist identity evidence and research.
Alongside the ownership transaction, VNET and CATL signed a separate strategic agreement concerning large compute-energy facilities, distributed infrastructure and related systems. That agreement is the industrial link; the share acquisition is the ownership link. Treating them as interchangeable would overstate what has been committed.
CATL's incentives extend well beyond a possible appreciation in VNET shares. High-density computing creates a market for storage, power electronics and grid-responsive systems, while operating campuses offer the chance to gather performance data and refine the technology. VNET, for its part, may gain access to expertise and potential financing partners. Whether those mutual interests translate into attractive terms for VNET remains an open commercial question.
The wider CATL investment pattern is worth following, particularly its interest in electrical equipment and reports of potential exposure to AI model developers. But the existence of interests at different points in the value chain does not demonstrate that workloads will be directed to VNET or that the companies have agreed to share project capital costs. The compelling possibility is industrial coordination; the missing evidence is the commercial contract.
Share acquisition closing · Direct CATL agreement
The financing question behind the expansion
VNET is not starting from zero in financing infrastructure. It has used private REIT and asset-backed structures to recycle capital from mature facilities, and management reported two private REIT listings in March 2026 with roughly RMB6.36 billion of combined offering size. That establishes a mechanism for releasing capital from completed assets. It does not, on its own, pay for the construction and power systems required by the proposed western campuses.
Where CATL could participate in project financing
VNET contributes development and operations; CATL-linked or other investors contribute equity to campus or energy SPVs. CATL could secure an equipment market and project returns. No specific JV has been disclosed for either western hub.
Storage, HVDC and generation or power-distribution equipment could sit with specialist energy owners under service or supply contracts. This could lower VNET's upfront capex, but may transfer energy upside to the asset owner.
Long-term customer commitments could support non-recourse or limited-recourse financing. CATL's interest would be smoother equipment deployment and a reliable operating reference site; lenders would still require bankable contracts.
Completed facilities could be sold or securitised, with VNET retaining management or operating contracts. This route is demonstrated in principle, although price, recourse, ownership retention and cash proceeds determine shareholder benefit.
A fifth possibility is customer prepayments or customer-funded fit-out, especially for very large AI deployments. Each mechanism has different economics: cheaper construction funding is only attractive if VNET does not surrender too much operating margin, residual value or control. An investment vehicle controlled by a strategic partner might accelerate delivery while also taking a significant share of the return.
A credible financing plan could also matter to CATL-linked shareholders through VNET's market valuation. Reliable project delivery, transparent ownership structures and manageable parent-level leverage would make it easier to attract additional capital and potentially improve the terms on which assets are recycled. But valuation support is an incentive, not a contractual commitment. CATL may rationally prefer arrangements that protect equipment sales or energy-asset returns even if VNET retains a smaller share of the economics.
Management's REIT and capital-recycling commentary
How the economics could extend beyond floor space
The landlord comparison understates parts of VNET's existing business. Connectivity and managed infrastructure are already commercial activities. Beyond them, AI-ready cooling and power density may command different economics from ordinary space; inter-site networking may generate recurring fees; energy optimisation could produce savings or service income; heterogeneous computing management could create another operating role; and development or asset-management fees might survive even when ownership of mature facilities is recycled. These are distinct revenue possibilities, not a single new business that can already be valued.
The question is how much of each opportunity VNET can actually retain. Huawei may supply the network technology; an energy partner may own the batteries; a cloud customer may control the orchestration software. The presence of VNET in the system does not mean it captures every layer's margin. Future disclosures need to show who invoices whom, who owns the assets and which savings or service fees accrue to VNET.
The evidence the next phase must deliver
The next disclosures should make the western projects less abstract. Investors need to see which customers have signed, what is binding, who owns and finances the power infrastructure, when facilities will be energised and how much capital remains at VNET's parent level. Beyond that, evidence of separately billed network, energy or operating services would help establish whether the company's role is genuinely broadening.
There is a less attractive outcome. Large project announcements could remain frameworks; equipment and energy partners could capture the best returns; or VNET could be left financing facilities faster than customers occupy them. A rising contracted backlog would not settle those questions if cash conversion and shareholder returns failed to follow.
The investment case, and its limits
VNET's story is stronger than a simple bet on Chinese data-centre construction. The wholesale estate is already substantially larger, the customer base includes leading cloud and internet platforms, and the Huawei networking work and ByteDance procurement trail reveal relationships with physical and technical substance. Qingyang and Zhongwei show how management hopes to extend that position into a national compute-energy architecture. CATL may help make the expansion financeable and commercially useful, but the decisive question is still how much of the resulting income and asset value VNET shareholders will keep.
Editorial analysis as of 15 October 2026. Disclosed operating results and signed agreements are distinguished from proposed projects, financing scenarios and uncontracted service revenues.