TREE NEWS update: DeepSeek released a technical paper detailing DSec (DeepSeek Elastic Compute), its agent-training sandbox platform, with founder Liang Wenfeng among more than 130 listed authors. One production unit comprises about 160 CPU nodes, 30,000 cores and 250TB of memory, hosting petabyte-scale images. The platform serves roughly 3 million sandboxes per day, peaks above 380,000 concurrent sandboxes, creates over 5,000 per second, and can spin up as many as 32,000 sandboxes for a single training task.
DeepSeek publishes DSec agent-training sandbox paper with 130+ authors
AI take
The scale here is the signal: a sandbox platform sized for millions of daily agent environments suggests agent training, not model training alone, is now an infrastructure problem. The author list, founder included, points to this being treated as core engineering rather than a side project. For anyone building or funding agent tooling, the interesting question is whether this elastic-compute pattern becomes a standard layer others must match, or stays an internal advantage.
Generated by AI for reference only.
Share on WeChat
Open WeChat → Scan → then tap "…" to send to a chat or Moments.
Tap "…" in the top-right corner to send to a chat or share to Moments.
Related News
42m ago
Supermicro Begins Shipping NVIDIA Vera Rubin NVL72 Racks
46m ago
Verizon Adds $50M to AI Skills Training Program, Total Reaches $70M
48m ago
SoftBank Launches Over $11B High-Yield Bond Sale to Fund AI Expansion
1h ago
China’s MIIT Pushes AI Integration Into Telecom Networks
2h ago
Qualcomm to Acquire US Embodied AI Firm Picknik
2h ago
Chengdu and China Telecom Launch ‘Tianfu Token Smart Factory’ Computing Platform