The Great Western Shift: How Qingyang and China’s Western Cities Became the World’s Token Production Hub

1. The “East Data, West Computing” Strategy

In 2022, China launched one of the most ambitious infrastructure repositioning programmes in the history of computing. The logic was deceptively simple: the eastern coastal provinces — where most of China’s internet users, enterprises, and AI applications live — face expensive electricity, scarce land, and tight carbon budgets. The western provinces — rich in wind, solar, hydro, and cool climate — have abundant cheap energy but little local demand for compute.

The “East Data, West Computing” (EDWC) initiative designated eight national computing hubs connected by ultra-high-speed backbone networks. Five of these eight hubs sit in western or central-western regions. Together, these eight nodes concentrate more than 80% of the country’s national hub computing capacity, with western hubs now holding approximately 33% of China’s total intelligent computing capacity.

The strategy is not merely about moving servers westward. It is about building a complete value chain: green energy generation → grid interconnection → intelligent computing clusters → token production → global distribution. Cities that successfully execute all five links become “token factories” in the truest sense. Qingyang is the clearest example of this transformation.

2. Qingyang: From Oil Town to Token Capital

Qingyang’s story is the most striking case study in China’s computing relocation. Historically known as an oil and gas base on the Loess Plateau, Qingyang had no digital economy to speak of when it was designated a national hub node in December 2021. The city had zero standard data centres at approval time.

Within three years, everything changed:

The growth curve is staggering. Qingyang’s intelligent computing scale went from 5,300 P in early 2023 to 51,000 P by end of 2024, to 114,000 P by end of 2025, and reached 215,000 P by August 2026 — a 40× increase in three years. Daily token calls from Qingyang’s data halls grew 10× over the same period, from 1.2 trillion to 12 trillion. The city now bears more than 25% of the nation’s independent-model AI token computing demand, hosting leading Chinese models including DeepSeek, Kimi, WPS AI, Minimax, Mimo, and Zhipu.

Perhaps the most telling metric is utilisation: Qingyang’s computing cluster utilisation exceeds 90%, far above the industry-wide GPU utilisation rate of below 30%. This means Qingyang is not building speculative capacity — it is producing tokens that the market is actively consuming. The city has attracted 751 digital economy enterprises, including 27 of the “3×500 Strong” companies (Fortune 500, China Top 500, and industry-specific Top 500) that have landed and begun production.

3. Why the West Wins on Cost

The economic logic of western token production rests on three pillars, each of which compounds the others:

It is worth noting that Qingyang’s 0.398 RMB/kWh is not necessarily the cheapest in the west — Ulanqab, for instance, offers even lower rates. What Qingyang pioneered is the “green power aggregation” direct-supply model: dedicated renewable plants (such as Gansu Electric Power Group’s 2-GW exclusive green power station) are source-grid-load-storage integrated with computing clusters, locking in both price and green certification. This model has been so successful that it is now being replicated across other western hubs.

4. Other Western Hubs: Ulanqab, Zhongwei, Gui’an

Qingyang is the poster child, but it is not alone. The western computing landscape comprises several complementary hubs, each with distinct strengths:

5. New Counterparties: Compute Operators Emerge

The rise of western token factories has spawned an entirely new category of Chinese business counterparty: the compute operator. Unlike traditional cloud vendors or IDC operators, compute operators specialise in running large-scale GPU clusters as token-producing facilities — managing everything from green power procurement and chip adaptation to token settlement and cross-border delivery.

Examples of this emerging player type include:

• Hongxin Electronics (航芯电子) — operates large-scale GPU clusters in western hubs, providing token output to model companies and enterprise clients
• YunSai Zhilian (云赛智联) — builds and operates intelligent computing centres, bridging green power, chips, and token settlement
• China Mobile MoMA, Kingsoft Cloud “Star Flow”, Capital Online GpuGeek — token service platforms deployed in Qingyang, offering ready-made token APIs to global developers

For overseas firms, these compute operators represent both opportunity and risk. On one hand, they offer direct access to the cheapest token production in the world, with green credentials that satisfy increasingly strict ESG requirements. On the other hand, many of these operators are new entities — some founded within the past 24–36 months — with limited track records, complex ownership structures, and varying degrees of financial transparency.

6. Due Diligence in the Western Token Economy

Evaluating a Chinese compute operator, AI infrastructure company, or token service provider requires a new due diligence framework tailored to the realities of western China’s token factories. Based on the market structure and policy environment described above, here is a practical verification checklist:

• Verify green power access: Does the operator have a bona fide “green power aggregation” agreement or source-grid-load-storage integration contract? Request the actual power purchase agreement and verify the renewable energy certificate (REC) equivalent. A professional enterprise credit report can confirm the operator’s registered business scope, capital structure, and associated power contracts.
• Confirm grid interconnection rights: Western hubs require specific regulatory approvals for grid interconnection. Verify the operator holds valid interconnection permits and that their declared capacity matches authorised grid access.
• Assess actual utilisation: Declared GPU count is meaningless without utilisation data. An operator running 100,000 GPUs at 15% utilisation is weaker than one running 10,000 GPUs at 90%. Independent credit and operational reports can reveal the gap between claimed and actual activity.
• Map chip composition: With domestic AI chips reaching 41% market share in 2025, clarify the operator’s chip mix. Heavy reliance on a single vendor (whether imported or domestic) creates supply chain vulnerability. Huawei Ascend, Cambricon, and other domestic accelerators each have distinct adaptation curves.
• Check policy compliance: National hub nodes require 80%+ green power for AI computing. Operators below this threshold face regulatory risk. Additionally, verify compliance with generative AI service filing requirements and data export regulations for any cross-border token delivery.
• Evaluate ownership & financials: Many compute operators are newly established SPVs (special purpose vehicles) under larger conglomerates. Understand the ultimate beneficial owner, the parent company’s financial health, and whether the operator has independent revenue or is fully dependent on parent subsidies.