DeepSeek’s latest hiring push offers a window into the infrastructure behind China’s AI race. The Hangzhou-based company is recruiting for an IDC data-center team across Beijing, Hangzhou, and Ulanqab, covering planning, construction, testing, and operations. Its careers site lists roles not only in model research and agent infrastructure, but also in supercomputing clusters, distributed storage, high-performance communications, training and inference frameworks, AI platform operations, and data-center work. TechNode’s report says the new infrastructure roles span facilities from megawatt-scale computer rooms to gigawatt-scale campuses, including air- and liquid-cooling systems.
The language should be read as a hiring signal, not proof that DeepSeek owns a named gigawatt campus or has announced a new data-center project. The company has not publicly identified a site, budget, construction timetable, or cloud partner. But the range of roles indicates that its ambitions increasingly extend beyond training and releasing models. To serve AI at scale, DeepSeek needs people who understand power, cooling, networking, storage, operations, and the software that turns raw accelerators into a usable inference platform.
That shift is important because the cost and reliability of AI are moving from research questions to infrastructure questions. Model releases can capture attention overnight. Serving those models to millions of users, developers, and agents requires a sustained industrial system. EastFrontier has examined how China’s domestic chipmakers are gaining AI market share. DeepSeek’s recruitment illustrates the complementary challenge: even the best chips and models need facilities and operations teams that can keep them available.
The IDC team covers the physical foundation of AI
The IDC data-center team is one of several technical groups listed by DeepSeek, but it is distinct because it connects AI software to the physical world. Data centers must secure electricity, remove heat, manage networking, maintain equipment, protect uptime, and plan capacity years ahead. Those tasks become harder as AI workloads increase power density and require faster communication between chips.
TechNode’s report says DeepSeek is looking for expertise in electrical engineering, heating, ventilation and air conditioning, automation, energy, communications, computer science, environmental engineering, and civil engineering. That range shows why AI infrastructure is not a conventional software project. A cutting-edge model may be developed by a comparatively small research group, but its deployment can depend on a much larger chain of specialists.
The company’s hiring spans Beijing, Hangzhou, and Ulanqab. Beijing and Hangzhou are major centers for Chinese AI research, technology companies, and engineering talent. Ulanqab, in Inner Mongolia, has become known as a data-center location because of available land, energy considerations, and its place in China’s broader computing-network strategy. DeepSeek has not stated why it selected these locations, so it would be speculative to assign a specific function to each. Their inclusion nevertheless suggests that the firm is thinking about a distributed infrastructure footprint rather than a single office-based engineering operation.
Cooling is one of the most consequential areas. AI accelerators generate concentrated heat, and the economics of an AI data center depend partly on how efficiently that heat is managed. The reported roles include air and liquid cooling, reflecting an industry-wide move toward more sophisticated thermal systems as workloads become denser. DeepSeek’s recruiting therefore aligns with the wider transition from general-purpose cloud capacity to infrastructure designed around AI training and inference.
Serving models requires more than a training cluster
DeepSeek’s jobs page shows a company building multiple layers of technical capability. In addition to its IDC team, it lists positions for supercomputing-cluster research and development, high-performance distributed storage, high-performance operators, communications and compilers, training and inference frameworks, and AI platform operations. It also has openings for agent infrastructure, model-data strategy, pre-training data, post-training, and multimodal research.
Together, those roles describe the stack required to move from a model demonstration to a durable AI service. A training framework needs to distribute work across many accelerators. Storage systems need to feed large datasets and checkpoints without becoming a bottleneck. Communications software needs to keep chips synchronized. Inference systems need to respond to users efficiently. Operations teams need to detect failures, manage capacity, and control costs.
For AI companies, inference is becoming especially important. Training a new model is an episodic event, but serving users is continuous. An API, coding assistant, agent platform, or consumer chatbot needs capacity every hour of every day. The performance of the underlying system affects latency, pricing, reliability, and the ability to handle spikes in demand. That is why a company known primarily for model releases would need to invest in operations and facilities talent.
The effort also complicates the simple idea that an AI lab either owns infrastructure or rents it entirely from a cloud provider. In practice, firms can use combinations of public cloud, dedicated capacity, colocation, partnerships, and their own operational teams. DeepSeek’s jobs do not reveal which model it has chosen. They do show that it wants internal expertise capable of evaluating and managing infrastructure at a significant scale.
Infrastructure hiring is becoming a strategic signal
China’s AI sector is being shaped by a tension between expanding demand and constrained access to the most advanced foreign hardware. Chinese companies are responding through domestic accelerators, model optimization, efficient architectures, cloud services, and new data-center construction. Talent is a critical part of that response. Building a competitive AI stack requires researchers, but it also requires engineers who can make systems work under power, cooling, networking, and supply-chain constraints.
DeepSeek’s hiring provides evidence of this broader change. The company is recruiting not just for isolated technical skills, but for an organization that can connect facilities, model systems, data, and operations. It is a move toward industrialization. The same pattern can be seen across China’s cloud providers, chip companies, model labs, and local governments, all of which are treating AI capacity as an infrastructure asset.
The expansion carries risks. Data-center projects are capital intensive, can be difficult to utilize efficiently, and may be exposed to shifts in hardware supply, electricity availability, and model demand. Recruiting for a data-center team does not guarantee that a company will achieve a profitable or resilient infrastructure footprint. The largest global AI companies have also shown how quickly capital expenditure can outrun near-term revenue.
Still, DeepSeek’s job listings indicate where the competition is heading. The next advantage may not come from a single benchmark result. It may come from the ability to operate a model platform cheaply, reliably, and at scale. That requires an AI lab to think like a systems company, a cloud operator, and an industrial planner at the same time.
DeepSeek has not announced a new campus, a power target, or an investment total. It does not need to for the hiring move to matter. The recruitment itself demonstrates that the company sees infrastructure as part of its core capability. In China’s AI race, models may remain the public face of competition. The data centers, cooling systems, storage networks, and engineers behind them are becoming the decisive machinery.