Shaoguan Data Centers: Alibaba, Tencent and Huawei Build Big While Locals Complain, FT Reports

Where does China put its AI compute? Increasingly, in a hilly corner of northern Guangdong. The Financial Times reports that Alibaba, Tencent and Huawei are pouring billions into data centers in Shaoguan, a region with cheap power and a cool climate, and that the boom is angering some of the people who live there. The piece is described as the first installment of a three-part series on China's AI build-out by Eleanor Olcott and Will Langley, covering Shaoguan and Ulanqab.

The FT article is paywalled, and we could not read it in full. This post therefore does what our "claimed vs verified" posts do: it states what the FT reports, as relayed by secondary summaries, then checks the buildout itself against analyst and official sources. The scale and the reasons for choosing Shaoguan are well documented. The detail of local grievances is not, and we flag it as the FT's account.

TL;DR: reported vs verified

Why Shaoguan?

The reason is mostly power and policy. SemiAnalysis's China data-center analysis says Shaoguan is the only site in Guangdong where datacenter energy quotas were reopened. The province stopped issuing them in 2021 to 2022, and applications resumed only for the Shaoguan hub under China's national "East Data West Compute" scheme, which steers compute to regions with cheaper power. Any project above 10 MW needs an energy-consumption review before construction, which makes that quota the binding cap on facility size.

The geography helps. SemiAnalysis reports roughly 1.3 milliseconds one-way latency to Guangzhou, 1.66 to Shenzhen and under 3 to Hong Kong, thanks to about 600 million dollars (4 billion yuan) of network spending by the local government since 2022, including 13 400G all-optical routes. A planned fiber corridor should cut latency by roughly another 30 percent from 2027. The analysts note that Tencent's later capacity went to Shaoguan and argue that training workloads care more about power price than latency.

There is also a regulatory backstory. According to the same analysis, Guangdong's Energy Bureau found in mid-2023 that Tencent's two Qingyuan campuses were running at roughly 13 to 14 times their approved scale and ordered rectification within six months. When quotas are scarce, a place that can legally add load becomes valuable.

What is being built

Three layers of reporting give a picture, with different levels of reliability.

The local government's own numbers. A 2026 press release distributed by third parties, and carried by outlets such as Macau Business, says Shaoguan has attracted 20 large computing projects from Alibaba, Tencent, China Mobile and China Unicom, worth over 97 billion yuan. It cites installed capacity near 30,000 petaflops, projected above 110,000 by the end of 2026, and says Alibaba's domestic 10,000-card cluster and a Huawei Ascend super node are already running. These are promotional figures from a city that is courting investment, so they are best read as targets.

Alibaba and China Telecom. Macau News reported on April 10 that China Telecom will own and operate a facility that starts with 10,000 of Alibaba's own Zhenwu chips, with plans to scale to 100,000, built for training and inference of models with hundreds of billions of parameters. It followed the activation of a Huawei Ascend 910C-based cluster the month before. The project fits Beijing's push for chip self-sufficiency under US export limits, a theme in our posts on China's AI chip market share and DeepSeek and Huawei's chip co-design.

The macro picture. SemiAnalysis puts China's fleet above 24 GW across more than 1,000 facilities and 60 operators, and describes the build-out as chip-gated rather than power-gated: electricity is abundant, but export controls limit chips. We looked at the national comparison in China 24 GW vs US 56 GW, and at the grid side in the 3.9 trillion yuan grid buildout claim.

What the FT says about local reaction

Here is the part we cannot verify. The headline says the investment is "angering some locals". A summary of the piece on World Today News, which credits the FT, describes residents as feeling a mix of cautious optimism and anxiety: they hope the AI push brings the boom that has long eluded the region, while worrying that rural areas get left behind as talent and capital go to cities. Another report on the FT story, as relayed by a search summary, mentions pollution, slow job creation and delayed land compensation. We could not open the original to confirm those specifics or any quotes.

Two things are worth keeping separate. First, the economics: data centers are capital-heavy and employ few people once built, so a region that hosts billions in servers can still see few jobs. That is a structural point that holds in China, the US and India alike. Second, the politics: in China, local governments typically court these projects, so the friction is more likely to be about land compensation and who gets the benefits than about a permit fight in the way Western moratorium campaigns work.

How this compares with data-center opposition elsewhere

The contrast is useful for readers following the global pattern. In the US, opposition has produced moratoriums such as San Francisco's data center moratorium. In India, residents near Thane protested over water and power in the Amazon Thane data center dispute. In both, the objection is open and legal. In Shaoguan, reported discontent is quieter and framed as economic divide, with no verified protests that we found.

For companies, the cross-border angle matters too. Chinese hyperscalers also rent chips abroad, as in Tencent's reported 100,000-chip lease in Southeast Asia, so domestic sites like Shaoguan are one part of a wider compute strategy.

What people are asking

Is Shaoguan a new AI hub or just overflow from Guangzhou and Shenzhen? Mostly the second, at least so far. Its pitch is proximity: low latency to the Greater Bay Area lets operators keep inference near customers in Hong Kong and Guangdong while running power-hungry training further north where electricity is cheaper. That is why the SemiAnalysis latency figures matter more than the climate.

Is cooler weather a real advantage? It helps, because cooling is a large share of a facility's power bill, but the quota system and the price of electricity matter more. A cool site without an energy quota cannot be built at scale, which is exactly the constraint Guangdong imposed on the rest of the province.

Does the Zhenwu and Ascend hardware match Nvidia? The sources we found do not give benchmark comparisons for these clusters, only chip counts and plans. Alibaba and Huawei have said little about utilization. Chip counts of 10,000 scaling to 100,000 describe ambition, and whether supply keeps up is the open question flagged in the analyst coverage.

What would a verified local complaint look like? Named residents, a village or township, a specific land dispute, or an official response from the Shaoguan government. The FT series is reportedly built on on-the-ground interviews, so the full text should answer this; we simply have not been able to read it.

What this means if you build or invest in AI

• Chinese model costs are tied to this build-out. Cheap power and state-backed sites help explain why Chinese labs can price aggressively; see how DeepSeek has priced its models.
• Capacity numbers are targets until audited. City press releases and analyst models differ; wait for operator or regulator disclosures.
• Watch the social license. Whether Shaoguan's residents feel they benefit may decide how fast the next sites can be approved.
• Chip supply is the real limit. If Zhenwu and Ascend volumes fall short, megawatts will sit idle.

What would change our read

We would treat the local-anger claim as verified if the full FT article, or a second outlet with on-the-ground reporting, documents specific complaints, named residents or official responses. We would upgrade the capacity figures if Alibaba, China Telecom or Huawei publish installed numbers. Until then, the buildout is real and large, and the human cost is the FT's reporting.

Related reading

• China 24 GW compute vs US 56 GW (SemiAnalysis)
• China's 3.9 trillion yuan grid buildout: claim vs verified
• Nvidia 34%, Huawei 28%: China AI chip share
• DeepSeek and Huawei Ascend chip co-design
• Tencent and Oracle 100,000-chip lease
• San Francisco data center moratorium
• Amazon Thane data center protests

Details are accurate as of October 20, 2026. The Financial Times article is paywalled; local-reaction details come from summaries and are not independently verified.