DeepSeek hires 150 systems engineers to scale agentic infrastructure
The AI lab is prioritizing backend virtualization and distributed storage over model research to power its massive sandbox clusters.
DeepSeek has launched a massive recruitment drive for systems engineers to scale the infrastructure powering its AI agents. The move signals a strategic shift from model development toward the industrial-scale plumbing required to run autonomous agents in production.
On September 7, 2026, DeepSeek opened approximately 150 engineering positions targeting senior professionals with two to 10 years of experience. The hiring is concentrated in server-side engineering and Agent Elastic Compute (DSec) rather than AI research. This push focuses on the foundational layers of computing—operating systems, virtualization, networking, and storage—rather than the weights and architectures of the models themselves.
The Architecture of DSec
At the center of this expansion is DSec, a production-grade sandbox platform designed for agentic AI. Built from three core Rust components—Apiserver, Edge, and Watcher—DSec provides the isolated environments where AI agents execute code and interact with tools. To manage the massive data requirements of these environments, DSec runs on top of 3FS, DeepSeek's proprietary distributed file system.
Reliability is a primary focus of the DSec architecture. The platform maintains a globally ordered trajectory log for every sandbox, a feature that allows agent rollouts to resume safely after preemption or interruption. This ensures that agents do not re-execute commands that might produce inconsistent results, a critical requirement for stable post-training and evaluation.
Why Infrastructure Now
As AI agents transition from simple demonstrations to production-ready tools, the primary compute bottleneck is shifting. While the initial AI race focused on GPU clusters for inference and training, the next phase requires massive scaling of the environments where those models actually operate.
By prioritizing virtualization and scheduling, DeepSeek is addressing the "last mile" of agentic workflows. The scale of this ambition is already evident: a single DSec cluster currently manages hundreds of thousands of concurrent sandbox instances in production. This infrastructure allows DeepSeek to run vast numbers of parallel evaluations, accelerating the refinement of its models through sheer engineering volume.
The Path Forward
This hiring surge follows a broader growth trajectory for the company, which announced plans in June to double the size of all its departments. Having open-sourced its agent harness in August, DeepSeek is now positioning itself as a leader in the systems engineering side of the AI stack.
Industry observers will be watching to see if other major AI labs follow suit by shifting resources away from research and toward the specialized virtualization and storage layers needed to make autonomous agents reliable at scale.