Chinese tech stocks are rallying following DeepSeek’s latest UE8M0 floating-point standard, which UBS says is “aimed at driving compatibility with ‘home-grown chips soon to be released’.” 

While market concerns exist, UBS notes that the initiative is “more a way to address ecosystem alignments in China than as technology supremacy over US-based GPU companies.” 

Still, the development should “further accelerate China’s path toward AI self-sufficiency,” the bank’s analysts wrote.

UBS explained that DeepSeek’s new V3.1 AI model, released on 21 August, merges the capabilities of DeepSeek-V3 and DeepSeek-R1 into a unified hybrid architecture. 

The bank described the model as “more evolutionary than revolutionary,” though it is built on the radical UE8M0 floating-point format that could be compatible with domestic chips.

Floating-point standards are critical in AI, serving as the numerical language that dictates how data is stored and computed. 

UBS added that FP8, used in DeepSeek’s UE8M0, “is clearly the favorite choice among emerging and low-cost models,” balancing speed and efficiency, though it is less precise than FP16 or FP32. 

Many US-based GPU makers already support FP8, whereas in China, the standard is still developing.

UBS emphasized that DeepSeek’s “software-first” and open-source approach should help establish an AI ecosystem domestically. 

The analysts said the initiative could contribute to China’s long-term growth potential, noting that “China typically accounts for 20-25% of major tech addressable markets” and that AI compute represents “almost 60% of overall AI capex.”

While UBS does not expect an immediate disruption, the bank recommends that investors view these initiatives as a pathway toward AI self-sufficiency, maintaining “a balanced positioning across semiconductors, software, and internet” amid the volatility in tech markets.