English text is being prepared.

DeepSeek's new funding round approaches 100 billion yuan, plans IPO for early 2027

DeepSeek's new funding round drew subscriptions exceeding the original 500 billion yuan target, with the term sheet pointing to nearly 100 billion yuan and a post-investment valuation of about 500 billion yuan. Tencent and CATL pledged the largest contributions, and the deal is expected to close within October, followed by a restructuring of the corporate structure to pave the way for an IPO in early 2027. External investors hold shares indirectly through limited partnerships, with no voting rights and a five-year lock-up; Liang Wenfeng controls a combined roughly 84% of equity and nearly all voting rights. CATL has invested a cumulative total of more than 10.5 billion yuan across two consecutive rounds, and has also taken a stake in 21Vianet and increased capital in Zhongheng Electric, integrating green power and power distribution solutions for DeepSeek's gigawatt-scale data center in Ulanqab.

AI helps you plan the National Day holiday, but don't let algorithms define "how to spend your money"

During the National Day holiday, AI interactions related to travel on Fliggy increased by 297% month-on-month, and visits to Tongcheng Travel's DeepTrip agent during peak hours on the first day of the holiday grew more than threefold. AI is becoming a "travel buddy" for more and more people. However, experts point out that AI travel guides have a more hidden advertising tendency than traditional search platforms, and their recommendations have "black box" characteristics. In actual tests, flight tickets recommended by AI were about 10 yuan more expensive than on other platforms. Platforms should draw a clear boundary between "recommendation" and "charging," and consumers should treat AI plans as drafts rather than decisions.

English translation is being prepared.

After Nvidia chips became hard to buy, what keeps China's large models running?

A Huxiu analysis points out that domestic computing power can now cover large model operation: Zhipu's domestic-chip inference cluster has reached a scale of 100,000 cards, and Meituan's LongCat-2.0 has also completed training on domestic computing power. However, Ascend 910C achieves an MFU of about 30%-35% in trillion-parameter post-training, lower than H100's 40%-50%. The article breaks computing efficiency into three variables—chips, software, and models—and compares the four paths of Zhipu, MiniMax, DeepSeek, and Kimi, concluding that Zhipu has the highest short-term certainty, while DeepSeek has the greatest medium-term elasticity. The author proposes three tracking indicators—capacity, efficiency, and cost—and argues that only when all three improve simultaneously can domestic computing power shift from a backup option to a cost advantage.

English translation is being prepared.

English translation is being prepared.

English translation is being prepared.

English translation is being prepared.

English translation is being prepared.