ByteDance is reportedly considering one of China’s largest new AI computing projects, with preliminary plans to add between five and six gigawatts of capacity in Ulanqab, Inner Mongolia. South China Morning Post reported that the TikTok parent had been discussing additional data centers with local vendors in Ulanqab’s Jining district and was aiming for delivery by early 2028.
The scale of the reported plan is extraordinary. SCMP cited a Soochow Securities estimate that one gigawatt of AI data-center capacity requires roughly 160 billion yuan of upfront investment. Applying that estimate to a five-to-six-gigawatt buildout produces a broad 800 billion to 960 billion yuan range, but that is a third-party calculation, not a disclosed ByteDance budget.
ByteDance did not comment on the reported expansion. The capacity figure, site, and timetable all come from an unnamed person familiar with the matter. The report should therefore be understood as a well-sourced account of preliminary discussions, not a final construction announcement. Even so, it shows how Chinese internet groups are looking beyond model releases and consumer apps toward the physical infrastructure needed to run AI at scale.
Ulanqab Has Become a Major Destination for AI Capacity
The reported project would be located in Jining district in Ulanqab, a northern Chinese city that has become a data-center hub. SCMP said the area has already attracted computing-facility investment from Huawei, Kuaishou, Alibaba, and Apple. A growing concentration of customers can make a location more attractive by supporting specialized power, networking, construction, operations, and equipment-supply ecosystems.
The broader numbers are already large. The source card for this article records that Ulanqab’s committed data-center capacity reached 12.5 gigawatts by June 2026, up from 3.3 gigawatts in July 2025. That does not mean all capacity is operating or that all of it is dedicated to AI training. It does indicate how rapidly the city is being positioned as part of China’s national computing network.
ByteDance’s potential addition would put it at the high end of that buildout. A five-to-six-gigawatt cluster would require large amounts of land, power, cooling infrastructure, network equipment, chips, servers, and operational support. The early-2028 target reported by SCMP also signals that the company would be thinking in multiyear infrastructure cycles rather than only near-term model launches.
The location is relevant to the economics of AI. Data centers need stable power and connectivity, and companies often balance those factors against proximity to users and corporate offices. Ulanqab’s attraction has been its role as a large computing base. The report does not disclose the specific hardware or workloads ByteDance would use, so it would be premature to assume the site will run a particular model, product, or domestic accelerator.
Earlier EastFrontier coverage described how Guizhou’s data centers are testing China’s AI development model. Ulanqab represents another version of the same national question: whether large compute hubs can translate power and construction capacity into durable AI services, rather than simply a headline count of server halls.
The Reported Cost Is an Estimate, Not a ByteDance Commitment
The 800 billion to 960 billion yuan estimate has attracted attention because it illustrates the capital intensity of AI infrastructure. But it must be handled carefully. The calculation begins with Soochow Securities’ estimate of 160 billion yuan per gigawatt and multiplies it by the reported five-to-six-gigawatt range. It is not an amount that ByteDance has announced, approved, borrowed, or allocated.
A project’s eventual cost can vary based on the balance between buildings, power connections, cooling systems, chips, servers, networking, and whether some capacity is leased instead of owned. The report says ByteDance was in preliminary talks with local data-center vendors, which suggests the commercial structure had not been finalized in public.
Bloomberg separately reported that ByteDance secured a US$29.6 billion dollar-denominated loan from a group of banks. The financing was described as Asia’s second-largest dollar loan of the year. The two developments are connected by the company’s broader AI investment needs, but Bloomberg’s report does not establish that the loan proceeds are committed to the Ulanqab expansion. Treating them as one confirmed financing package would go beyond the evidence.
The infrastructure spending comes as ByteDance develops its Doubao product family and expands its AI operations. In August, EastFrontier reported that ByteDance created a new AI data and safety department. The reported Ulanqab plan would address another layer of the same strategy: the compute base required to train, serve, and scale AI products.
Large capital needs can create a strategic advantage for the companies that can fund them, but they also create execution risk. A company must match capacity with demand, manage equipment supply, secure electricity, and avoid building facilities that sit underutilized. That is especially relevant in China, where multiple internet platforms, cloud providers, and local governments are pursuing AI-related data-center projects at the same time.
China’s Compute Race Is Moving From Plans to Infrastructure
The reported ByteDance expansion fits a wider increase in Chinese AI computing investment. EastFrontier reported in August that China’s AI compute capacity had reached 2.45 million PFLOPS. That national figure measures a different concept from gigawatts of data-center capacity, but both point to the same competition over the physical resources needed for large-scale AI.
For ByteDance, the Ulanqab plan would be a major signal that it intends to control more of the infrastructure behind its AI services. The company is best known globally for TikTok, but in China it is also a major developer of AI products and models. Building or securing dedicated capacity can reduce dependence on third-party cloud providers and provide more flexibility as model workloads increase.
Still, the key facts remain provisional. ByteDance did not respond to SCMP’s request for comment. The five-to-six-gigawatt scale, the Jining location, and early-2028 schedule are all attributed to the source familiar with the discussions. The cost range is an analyst estimate, and the relationship between the reported bank loan and this particular project has not been confirmed.
The report is meaningful because it gives a rare numerical sense of the infrastructure ambitions under discussion. If it moves forward, the plan would strengthen Ulanqab’s position in China’s computing map and make ByteDance one of the most aggressive builders of AI capacity in the country. Until the company discloses more, however, the development should be read as an emerging plan, not an executed investment.