Forget DeepSeek. China’s already released 5 new AI models and UBS prefers this one
HS 01/03/2026
As speculation over a new version of the DeepSeek AI model grows, several other Chinese tech companies have released their own generative artificial intelligence models in the past few weeks. They range from Alibaba’s Qwen 3.5 to ByteDance’s video-generating Seedance 2.0. But what’s caught the eye of AI users — and UBS stock analysts — is MiniMax , which only went public in Hong Kong in January. The company in mid-February released its M2.5 model, with performance rivaling Claude’s latest Opus 4.6 at a lower price. That’s driven a surge of developers to choose MiniMax M2.5 over even DeepSeek’s current V3.2 model, as well as those from U.S. companies, according to data from OpenRouter. One-tenth the price Data from a mid-February report showed MiniMax’s AI usage had already reached one-third of Anthropic’s Claude, at one-tenth the price, according to UBS analysts. Recently, the investment bank initiated coverage of the Chinese stock with a buy rating and a price target of 1000 Hong Kong dollars ($127.83). That’s upside of more than 30% from the HKD763.50 where the Shanghai-based company traded Friday afternoon. “We are positive on MiniMax’s potential upside in gaining share in the global enterprise market,” the UBS analysts said. Also setting the Chinese company apart: an array of AI tools leading not just in text generation, but video generation, ElevenLabs-like audio generation and AI companionship, they added. That’s different from Zhipu , which also went public in Hong Kong in January. Its recent GLM 5.0 model focuses on coding. Rival Chinese startup Moonshot, which released a popular Kimi 2.5 model in late January, is still privately held and has also focused more on coding and agentic-powered task completion, rather than audiovisual generation. 2025 global shock Anticipation of a new DeepSeek model climbed around the nine-day Lunar New Year holiday that ended Feb. 23. The Chinese startup’s model release ahead of the same holiday last year shocked global markets with how advanced China’s generative AI capabilities had become, despite U.S. chip restrictions. This year, China’s internet companies Tencent , Alibaba , Baidu and ByteDance used the Lunar New Year to attract local consumers, with the AI race in China shifting toward user applications. After integrating their own AI models, or DeepSeek, with existing apps and services such as e-commerce, the companies spent hundreds of millions of U.S. dollars on red packet promotions. “In the near term, we believe this will accelerate AI penetration among users (especially in lower-tier [less developed Chinese] cities) and promote usage of more AI and agentic functions such as image and video generation, quick commerce, and other transactional bookings within these AI native apps,” UBS China internet services analysts said in a China AI report on Feb. 23, at the end of the holiday. Of the stocks highlighted in the report, MiniMax is the only one whose focus from the start was AI models. “We think the ongoing AI disruption narrative boosts sentiment for model providers while keeping investors cautious on vertical platforms and application companies,” the UBS analysts said. “We recently initiated coverage on MiniMax, and consider the company well positioned to benefit from the AI tailwinds in China and global markets.” 3% opportunity In an optimistic scenario, MiniMax could achieve 3% of the global market for enterprise services, for segment revenue of $41 billion, according to UBS estimates. Video generation offers MiniMax a roughly $5 billion revenue opportunity, while AI companionship is about $4 billion, the report said. New AI models can quickly disrupt existing rankings, and data shows users will swiftly shift to tools that are cheaper and smarter. But if MiniMax’s AI leadership grows faster than currently expected, the stock could climb even higher, to at least 1380 HKD, the UBS analysts said in their report initiating research coverage of the company. — CNBC’s Michael Bloom contributed to this report.
