DeepSeek V4 Is Here, but Don’t Expect Another ‘DeepSeek Moment’
We view V4 as a competent follow-up, not the watershed that R1 was.

DeepSeek released a preview of its V4 model, 15 months after introducing R1—an event widely known as the “DeepSeek moment.” DeepSeek claims V4’s capabilities largely match those of US models.
Why it matters: We view DeepSeek V4 as a competent follow-up, not the watershed that R1 was. R1 shocked markets because no one expected a Chinese open-source model to compete at the frontier. V4 is the next arc of the same story, and follow-ups don’t spook markets the way surprises do.
- While V4 performs strongly in various benchmarks, we think fixating on these scores misses the bigger picture. Value in AI is created when capability is embedded in revenue engines—something DeepSeek lacks, but which several of our covered firms have accomplished.
- V4’s use of Huawei Ascend chips is consistent with our view that Chinese AI firms are gradually shifting to homegrown silicon. This should benefit China-listed GPU names and foundries like SMIC and Hua Hong in the long run.
The bottom line: The V4 release doesn’t change our views on listed Chinese tech names, but it reinforces two existing themes. (1) Chinese models continue to put up numbers that keep the pressure on US models, and (2) China’s chip self-sufficiency is improving despite US export controls.
- Tencent remains our top pick in Chinese tech. It doesn’t have the best AI model, but its integrated ecosystem generates vast amounts of proprietary behavioral data, which serves as a built-in testing ground that’s already turning AI advancements into shareholder value.
- On our semiconductor coverage, the biggest difference between the market’s view and our 1- or 2-star ratings for SMIC and Hua Hong is whether this rise in domestic AI demand will be profitable in two or three years, amid elevated capex and low yields.
Between the lines: DeepSeek’s own benchmark table includes other Chinese models—Kimi K2.6 and GLM-5.1—alongside Opus, Gemini, and ChatGPT. That framing didn’t exist with R1, and shows how much domestic competition has intensified.
The author or authors do not own shares in any securities mentioned in this article. Find out about Morningstar’s editorial policies.
