Expired60% confidenceFactExact time
The chips DeepSeek is developing in-house are aimed at inference rather than training.
2
Sources
60%
Confidence
Medium-term (~90 days)
Relevance
7/16/2026
First Seen
This claim has expired. It is kept for the record and no longer describes the current state.
Valid until: 10/14/2026(expired)
Sources
Related Claims
UnverifiedDeepSeek's chips are dedicated to inference rather than training, and inference chips can typically adopt low-precision quantization formats such as INT8 or even INT4.82% similarVerifiedThe DeepSeek-R1 technical report indicates that this training approach can cause the model to spontaneously develop complex reasoning behaviors such as self-correction and re-thinking, without requiring explicit human annotation guidance76% similarUnverifiedDeepSeek's marginalization of Pro may stem from freeing up compute resources, migrating to domestically produced inference chips, or prioritizing its flagship product.75% similarUnverifiedDeepSeek open-sourced DSpark (repository name: DeepSpecs), which achieves 50% to 400% inference speedup without retraining or quantization.73% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/530826API
curl https://kongchang.com/api/v1/knowledge/claims/530826MCP
get_claim(id=530826)