One year on from the shock that DeepSeek delivered to the global tech sector, China’s AI landscape is entering a period of intensified competition as companies ready new low-cost models and consumer-focused upgrades timed to the Spring Festival holiday, which officially begins on February 15.
DeepSeek - a Hangzhou-based startup that vaulted to prominence early in 2025 - precipitated a reordering of priorities across China’s AI ecosystem when it released a low-cost, open-source model during the country’s New Year holiday. That initial release came amid U.S. export controls that limit access to advanced semiconductors, yet still managed to force a reappraisal of assumptions about the resource requirements for high-performing AI systems.
Industry observers say the market reaction to DeepSeek’s debut has left rivals better prepared and eager to match or exceed the company’s combination of performance and affordability. Alfredo Montufar-Helu, a managing director at Ankura Consulting in Beijing, framed expectations succinctly: "The surprise would be if some of these new models end up being underwhelming. I think there are high expectations here."
Over the past week several Chinese AI players have signalled or revealed their next moves. Zhipu AI released an updated model that it says improves coding capabilities and can run long-running tasks without repeated prompts from users. ByteDance formally introduced Seedance 2.0, described by the Chinese state-backed Global Times as a video-generation AI able to produce cinematic-style outputs in seconds, and is preparing upgrades to Doubao - currently reported as China’s most popular AI app with 155.2 million weekly active users, according to QuestMobile.
DeepSeek is also preparing to ship a next-generation model, V4, while Alibaba is expected to release a Qwen 3.5 series, a reported upgrade that would bring enhanced mathematical reasoning and coding skills. Evidence that Alibaba’s release may be imminent appeared when Qwen developers submitted supporting code to the open-source repository Hugging Face earlier this month - a common precursor to a formal launch. Alibaba, ByteDance and DeepSeek have not announced formal release dates and did not respond to requests for comment.
DeepSeek’s January 2025 debut had immediate market effects beyond China. The model’s release coincided with a severe selloff in global tech stocks, including an intraday loss of $593 billion in market value for AI chipmaker Nvidia. The episode also prompted some of DeepSeek’s Chinese competitors to accelerate upgrades to their own models.
Central to the ongoing competition is a persistent emphasis on low-cost deployment and open-source distribution. Over the last two years, DeepSeek’s pricing strategy repeatedly undercut rivals, putting usage costs far below many U.S. offerings. A RAND report on U.S.-China AI competition cited in industry commentary found that Chinese models run at about one-sixth to one-fourth the cost of comparable U.S. systems, a finding that has bolstered the narrative that high performance does not necessarily require the deepest pockets for computing infrastructure.
Analysts highlight the significance of that cost-performance mix. Lian Jye Su, chief analyst at tech research firm Omdia, said: "DeepSeek showed the industry that you can create a very good model even when you’re resource-constrained." He added that "the combination of open-source access, strong reasoning capabilities and low deployment costs has become a defining model for how Chinese vendors now approach foundation models."
Before DeepSeek’s breakout, several industry leaders favored closed-source approaches. The shift toward openness has been visible in recent months: within days of DeepSeek’s assistant surpassing ChatGPT in App Store downloads in the U.S., major Chinese firms including Baidu began opening portions of their own models. Hugging Face has seen a growing volume of releases from Chinese technology firms and startups, including Baidu, ByteDance, Tencent and Moonshot.
Notably, while several competitors pursue broader commercialisation - integrating generative AI into consumer-facing services and transactions - DeepSeek’s corporate structure remains distinct. The company’s parent is a quantitative hedge fund controlled by founder Liang Wenfeng. That ownership model allows DeepSeek to emphasise research rather than immediate monetisation, avoiding some of the external shareholder pressures that larger public companies face.
Other firms are taking a different path. Alibaba’s Qwen chatbot has been trialling commerce-focused features that let users buy products directly through conversational prompts, a pivot that reflects mounting pressure on public companies to convert AI investments into revenue-generating services while they continue to scale costly infrastructure.
As China’s longest and busiest holiday approaches, market participants will be watching closely to see which companies succeed in balancing cost, openness and consumer appeal. The coming weeks promise a concentrated set of product releases and upgrades that, market observers say, could further entrench low-cost, open-source models as a dominant design in China’s AI landscape.
Note: Some companies mentioned have not provided formal release dates for their products; requests for comment were not answered by Alibaba, ByteDance or DeepSeek.