DeepSeek’s newly launched V3.1 model has stirred attention across technology and investment circles, particularly after introducing the UE8M0 floating-point standard that Chinese chipmakers are expected to embrace.
According to UBS analysts, this move is less about displacing U.S.-based GPU leaders and more about aligning China’s ecosystem as it pushes toward AI self-sufficiency.
The company’s latest release merges capabilities from earlier DeepSeek-V3 and DeepSeek-R1 models into a hybrid architecture. UBS compared the approach to OpenAI’s GPT-5, noting it is “more evolutionary than revolutionary.”
While seen as an incremental step, the standout feature is the adoption of UE8M0, an FP8 floating-point format that offers reduced cost through exponent use.
Floating-points function as the numerical language of AI, determining how data is stored and processed. The industry has long relied on formats like FP32 or FP16, with FP8 gaining traction for balancing speed and cost at the expense of precision.
DeepSeek’s software-first and open-source approach aims to standardize FP8 adoption in China, providing much-needed compatibility across domestic hardware.
UBS strategists argue that the near-term disruption to global GPU markets is limited.
“Despite market concerns, at this stage, we see DeepSeek’s latest standard more as a way to address ecosystem alignments in China than as technology supremacy over U.S.-based GPU companies,” a team led by Sundeep Gantori wrote..
Still, they stress that investors “cannot be complacent” as such initiatives accelerate China’s path to AI self-sufficiency.
Source :
https://www.investing.com/news/stock-market-news/will-deepseeks-latest-innovation-pose-a-threat-to-gpu-makers-4211109
