Monthly salary of 30K, working as an operations and maintenance engineer for DeepSeek's data center on the grassland.
DeepSeek has released new job openings—this time, not for algorithm engineers or large model researchers.
Instead, they are hiring Senior Data Center Operations Engineers and Senior Data Center Delivery Managers.
Key point: "Responsible for the end-to-end project management of data centers, from initiation and construction to delivery and operation."
This company, known for its "low-cost algorithmic miracles," is suddenly shifting toward heavy asset operations.
Faced with the reality of limited computing power in the past, DeepSeek did not wait for better hardware.
Liang Wenfeng once expressed opposition to the "brute-force miracles" approach of stacking resources, advocating instead for optimizing architecture and engineering efficiency to compensate for hardware limitations, allowing "small efforts" to achieve miracles.
However, this efficiency-first philosophy has, for the most part, remained at the software and model level.
Building their own data centers and personally managing physical computing power means Liang Wenfeng is extending this methodology from training code to data center operations, moving from the world of bits all the way to the atomic world.
What they are pursuing is no longer just model training and inference efficiency, but the optimal Total Cost of Ownership (TCO), from data center cooling to electricity costs.
AI Heart on the Grassland
DeepSeek has chosen Ulanqab as its location, which is one of the eight national hub nodes of the "East Data West Computing" project and a core component of Inner Mongolia's computing power cluster.
By the end of 2025, the total computing power scale in Inner Mongolia reached 237,000 P, with intelligent computing accounting for over 92%, ranking first in the country.
For every 10 P of new intelligent computing power added nationwide, over 9 P are deployed on this grassland.
For computing centers, Ulanqab's natural conditions are top-tier. The area's low annual average temperature naturally serves as "free air conditioning" for servers.
Low temperatures mean lower cooling costs, with newly built data centers strictly controlling PUE below 1.2. Clean energy installed capacity accounts for over 65%, and the nation's first integrated source-grid-load-storage project for data center green power has already been put into operation here.
More importantly, this place is already a gathering ground for computing power giants.
Huawei Cloud partnered with Ulanqab as early as 2013, building one of Huawei Cloud's largest global data centers, planned to accommodate millions of servers.
Alibaba Cloud plans to invest over 10 billion yuan, with a single campus covering 114,000 square meters, and its registered capital has increased from 70 million yuan to 3.5 billion yuan.
This mature industrial ecosystem has already lowered the barriers and risks for DeepSeek to build its own data center to a minimum.
DeepSeek's strategic shift is also no coincidence in timing.
On the Eve of V4, Computing Power is the Moat
According to leaks, DeepSeek founder Liang Wenfeng recently confirmed in internal communications that the new flagship large model, V4, will be officially released in late April.
Rumored parameters for V4 are explosive: a trillion-parameter scale and a million-level context window.
If true, this would be a leap in magnitude.
As the marginal returns of algorithmic innovation begin to diminish, large-scale, high-efficiency physical data centers become the key to victory.
Recently, DeepSeek's web version quietly launched two interaction options: "Fast Mode" and "Expert Mode." Fast Mode supports concise searches, while Expert Mode handles long-range, complex problems.
The servers on the grassland may soon face their first trillion-parameter test.
Reference links:
[1]https://app.mokahr.com/social-recruitment/high-flyer/140576#/job/d34b2175-e8e4-47a8-aa63-3c2db5501ce9
This article is from the WeChat public account "QbitAI," author: Meng Chen
