|
Getting your Trinity Audio player ready...
|
One of China’s largest state-backed telecom operators has just made its biggest bet yet on the physical backbone of artificial intelligence. China Mobile has completed an expansion of its intelligent computing hub in Hohhot, the capital of Inner Mongolia, raising total processing capacity to 20,100 petaflops, according to a report by Inner Mongolia Daily on Monday that was picked up by the South China Morning Post. Earlier coverage from 2024 put the facility’s capacity at 6,700 petaflops, which means the site has roughly tripled in size in about two years.
The numbers come with some context. A petaflop is a thousand trillion calculations per second, and 20,100 petaflops is the same as about 20.1 exaflops. Of that total, 16,700 petaflops are dedicated to artificial intelligence and smart computing tasks, according to the report as summarized by KuCoin News. The expansion cost about 2.7 billion yuan, roughly $400 million at recent exchange rates, and the project has a total power capacity of 120 megawatts. The facility began operating in 2024 and deploys around 20,000 AI accelerator cards, Inner Mongolia Daily said. That last detail is worth noting, because the reports reviewed do not explain how capacity tripled while the card count is described as roughly unchanged, which could reflect upgraded chips or a different way of measuring performance.
That measurement point matters when comparing computing hubs. Petaflop figures depend on the numerical precision used, and AI workloads usually rely on lower precision than traditional supercomputing, so a figure quoted for AI performance cannot be compared directly with other benchmarks. The reports do not state which precision China Mobile used, so readers should treat 20,100 petaflops as a company-reported number, not an independently verified one, and avoid drawing conclusions about how it stacks up against specific systems elsewhere.
The facility is described as the world’s largest single-site intelligent computing centre run by a telecoms operator. When it opened in April 2024, Chinese outlets reported that around 85 percent of its AI chips were made in China, according to Yicai, and that it used technologies including high-speed lossless networks and cold-plate liquid cooling, as Light Reading noted. Domestic chip supply matters because the United States has restricted exports of advanced AI processors to China, and Chinese operators have been under pressure to rely more on local suppliers. The latest reports do not say what hardware went into the expansion, so it is unclear whether the same domestic share applies.
Where the computing power goes is as important as how much there is. The South China Morning Post reported that the project would funnel 90 percent of its computing power to the Beijing-Tianjin-Hebei economic zone, which includes the capital and its surrounding industrial region. That fits the logic of China’s “East Data, West Computing” strategy, under which data-heavy processing is moved to western and northern regions with cheaper land, power and cooler climates, while demand stays concentrated in the wealthy east. Inner Mongolia is seen as a critical part of that plan, and Hohhot and its neighboring city Ulanqab have emerged as a key link in the infrastructure network, according to the same coverage.
The environmental claim is part of the pitch. Green electricity powers 88 percent of the facility, Inner Mongolia Daily said, in line with Beijing’s effort to align fast AI expansion with national carbon goals. Inner Mongolia has large wind and solar resources, and its cold weather cuts the energy needed to keep servers cool. Third-party analysis from the engineering firm GBC estimates that the Hohhot and Ulanqab corridor hosts more than 3,000 megawatts of hyperscale capacity across several operators, including China Mobile, Alibaba Cloud, ByteDance and China Telecom, which would make it the largest concentration of data center capacity in the world. That is an industry estimate, not an official figure, but it shows why the region features so prominently in China’s plans.
China Mobile has been signaling this direction for years. A 2021 annual filing described a “3+3+X” data center layout with three key regional centers, three inter-provincial centers in Hohhot, Harbin and Guiyang, and provincial centers across the country. A 2024 industry report said the company planned to lift its intelligent computing capacity from 10 to 17 exaflops, and the latest numbers show the Hohhot site alone now offering about 20 exaflops. The earlier plans may use different definitions, so the totals should not be added or compared without care. Chairman Yang Jie has said the operator “transitioned from cloud to computing,” and it has placed large server orders, including a 16 billion yuan order for 265,000 servers reported in 2024, mostly supplied by Chinese firms such as ZTE, H3C and xFusion.
Why would a telecom company build AI computing at this scale? The reasoning is partly strategic and partly commercial. Telecom operators already own networks, land and power contracts, and they sell cloud and computing services to enterprises and government. As AI demand grows, selling capacity to model developers, businesses and public agencies becomes a significant revenue opportunity. It also positions state-owned firms as national infrastructure providers, which is consistent with Beijing’s push to treat computing power as a public utility. China Mobile said the Hohhot center was designed to support sectors such as transportation, healthcare, education, energy and finance.
There are limits to what can be concluded. The reports do not identify customers, utilization rates or revenue, and capacity on paper is not the same as capacity in use. Training large AI models requires thousands of chips working together over fast networks, and performance depends on software, cooling and reliability as much as on raw peak numbers. Chinese data centers have also faced concerns about overbuilding in some regions, and a facility is only valuable if there is demand to fill it. Nothing in the coverage suggests the Hohhot expansion has a problem, but those are the questions analysts will ask in the months ahead.
For readers outside China, the development is another sign of how large the global race for AI infrastructure has become. This month alone, Samsung guided to an $80 billion quarterly profit on AI memory demand, SpaceX was reported to be seeking $40 billion in debt for Nvidia chips, and the IMF said AI investment is helping keep global growth near 3 percent. China Mobile’s expansion shows that state-led spending is part of the same wave, backed by policy and domestic supply chains rather than by public markets. The company publishes its results and announcements on the China Mobile website, and the South China Morning Post has the report on its site.
Readers who follow global technology and business stories can find more coverage at BusinessTech Nigeria. The headline number is impressive, and the more revealing details will be who uses the capacity and how quickly the next expansion arrives.
