- The AI race is moving beyond GPUs, as companies compete for land, power and data center capacity
- DeepSeek’s search for infrastructure talent signals a shift from model development to physical AI infrastructure
DeepSeek, one of China’s most closely watched AI startups, has begun recruiting for its IDC (internet data center) infrastructure team, with positions spanning electrical engineering, HVAC, automation, energy, environmental engineering and even civil engineering.
The company’s recruitment page highlights a broader ambition: “Every generation faces its own infrastructure revolution. The past brought railways, power grids and the internet. Today, it is the hyperscale computing infrastructure powering AI development.”
The hiring push is more than a routine expansion. In late June, DeepSeek announced plans to at least double the size of every department, with its data center team already included in the expansion.
But the explicit inclusion of civil engineering roles reveals a more aggressive direction: AI model companies that once relied heavily on external cloud resources are increasingly moving toward controlling their own physical infrastructure.
From GPUs to land and power
For the past two years, the fiercest competition in AI has centered on access to Nvidia GPUs. But the battlefront is now extending into the physical world.
As AI computing demand explodes, data center deployment is becoming increasingly strategic. Training workloads, which are less sensitive to latency, are moving toward regions with abundant land and cheaper electricity, while inference workloads requiring real-time responses need to stay closer to major economic hubs.
DeepSeek’s hiring footprint reflects this emerging strategy. Hangzhou serves as its technology and operations base, Beijing remains a key research center, and Ulanqab in Inner Mongolia is emerging as a large-scale computing hub.
Known as China’s “grassland cloud valley,” Ulanqab has attracted 84 data center projects with total investment exceeding 500 billion yuan. Its annual average temperature of just 4.3 degrees Celsius also provides natural cooling advantages for large-scale computing facilities.
More asset-heavy
DeepSeek is not alone in this shift.
ByteDance has also expanded into data center infrastructure, establishing new companies in Zhongwei and Ulanqab in July with business scopes including “non-residential property leasing.”
The competition among technology companies is no longer limited to chips and algorithms. Increasingly, access to suitable land, reliable electricity supplies and high-speed networks is becoming a critical advantage.
Global real estate firms such as JLL and CBRE have identified data centers as one of the fastest-growing asset classes. Industry analysts expect electricity availability to become the biggest constraint on future AI infrastructure expansion.
That explains why DeepSeek lists electrical engineering ahead of many other infrastructure disciplines: building gigawatt-scale AI computing campuses requires power systems expertise as much as construction capability.
Why it matters globally
The AI race is entering a new phase. The winners may not only be those with the strongest models, but also those with the most reliable access to computing infrastructure.
For AI companies, owning or controlling data centers, power resources and physical capacity can translate into lower costs and greater supply certainty.
DeepSeek’s search for civil engineers marks a symbolic transition: AI companies are moving from writing algorithms to building the infrastructure that allows those algorithms to operate at scale.
