Chinese AI startup DeepSeek is moving to double its headcount across all departments, according to multiple Chinese media reports on June 25, 2026. Following its first round of external funding, the company plans to expand hiring across all areas, including Agentic AI, infrastructure, and research and development.
June 25, 2026 · DeepSeek
DeepSeek Plans to Double Its Staff — Across Every Department
The Hangzhou AI firm behind the "DeepSeek moment" is reported to be doubling headcount as it scales model development and shifts toward self-owned data centers. Scale and timing await official confirmation.
2×
Planned increase in staff, across all departments
1M
Token context length supported by V4-Pro
GW
Scale of self-owned data centers now being planned
A Doubling of the Workforce
Based on the listed workforce band, the same proportional jump applies at both ends of the range.
The 2026 Model Lineup
DeepSeek-V3
671B total (MoE, 37B active) · MLA attention
DeepSeek-R1
Reasoning-focused, o1-class · R1-Zero open source
DeepSeek-V4-Pro
1.6T total / 49B active · up to 1M-token context
DeepSeek-V4-Flash
285B · efficient FP8 implementation
Well Received
State-of-the-art open-source results on math, coding & reasoning
Strong cost-performance and efficiency
Wide framework support (vLLM); local & cloud use
Popular distilled versions on Qwen and Llama
Noted Drawbacks
R1: endless repetition and poor readability
R1: language-mixing issues
Large models like V4-Pro hard to run on a single GPU
Requires multiple nodes or heavy optimization
Scaling on Its Own Terms
A flat, highly autonomous culture and a growth strategy distinct from U.S. players — now expanding from rented to self-owned data centers, from O&M staff in Ulanqab to MW-to-GW-scale planning. The reported doubling extends that approach. Details await official confirmation.
Continue reading The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.
Already purchased? Sign in ✓ Signed in — this article isn’t included in your current plan.Unlocking the full article…