During this National Day holiday, AI companies both domestically and internationally have been busy, with several major financial moves and product launches capturing attention. The biggest one came from DeepSeek—the Reuters reported on October 6 that the company is advancing a new round of funding aiming to raise over 80 billion yuan, expected to be completed within October. Notably, the company had only started this round of fundraising in July, initially targeting 50 billion yuan and discussing a valuation of 500 billion yuan; just a few months after its first external funding, Reuters said DeepSeek had recently completed a financing round of about 7.4 billion US dollars in June. If this round proceeds at the current scale, it will secure two rounds of hundreds of billions of yuan in funding. Bloomberg previously noted that Tencent and CATL are among the institutions that have made significant investment commitments, although DeepSeek has not yet officially confirmed the funding size or investor arrangements.
On the same day, Bloomberg also released another funding news: the parent company of Kimi, Yuedao Dark Side, has completed its final private funding round before listing, with a valuation of about 50 billion US dollars. This comes just over two months after it was reported on July 29 that it had secured about 3.5 billion US dollars in funding with a valuation of 35 billion US dollars, nearly doubling its value. Yuedao Dark Side is aiming to go public in the first quarter of 2027, planning to raise up to 5 billion US dollars through an IPO and preparing early meetings with potential investors, with the earliest possible start in October to gauge market interest for this listing.
The shortage of computing power has pushed Tencent to sign a five-year lease agreement with Oracle. According to media reports, Tencent signed a lease agreement earlier this year, locking in computing power from multiple data centers in Southeast Asia, involving approximately 100,000 advanced AI chips. An informed source estimated the total value of the agreement at about 7 billion US dollars, with a prepayment of about one-third—Tencent is not acquiring the chips themselves but rather the right to use computing power in overseas data centers.
On the product side, Google updated two product lines during the holiday. The new flagship model Gemini4Argon targets professional tasks such as software engineering, law, finance, and cybersecurity defense. The most noticeable change is that the single output limit has increased from 64,000 tokens in the previous version to 1 million, meaning the model can now generate longer code or documents in one go. It was first made available through the Fairwind program to approved cybersecurity defense organizations, while more paid API users and Google AI Ultra subscribers will have to wait a bit longer. On October 6, Nano Banana 2.1, which targets posters, e-commerce images, and daily photo editing, was also updated. Google focused on improving text presentation, infographic layout, and consistency of people and objects during continuous editing—when users repeatedly change backgrounds, adjust text, or modify layouts, the model should try to preserve the parts already edited in the previous round.
In a long interview published in Vanity Fair on October 5, Altman referred to his new assistant Dots as the product he had been waiting for. He said this AI could handle his emails and to-do lists, and gave his own example: after talking on the phone and giving a set of ideas, when he returned to his computer, Dots had already sent the results via Slack, even pointing out some issues. However, he also set boundaries for the AI—trivial tasks can be assigned to Dots, but conversations with friends should still be done by himself. He emphasized that AI shouldn't replace human connections. When asked about the delay of Astra 6.1, he admitted that the version ready at the time carried too high a risk of failure. OpenAI still wants to open it to more people, but must first improve safety and cannot release it as recklessly as they did a year ago. Coincidentally, Jason Kwon, Chief Strategy Officer of OpenAI, apologized on October 6 during an Australian parliamentary hearing for the incident where an AI system had accessed government websites and promised to accelerate incident reporting. As for the IPO, Altman said he isn't in a hurry. Even if it goes public, it will still be governed by a non-profit organization, and if there's a conflict between mission and shareholder interests, the mission will take precedence. When asked about potential stock price drops, he replied, "Let it drop."
During the holiday, Tibo from OpenAI issued a 28-day ultimatum to Codex and ChatGPT Work users: in the next 28 days, either launch an improvement that clearly benefits most users each day, or reset the usage quota. Among the initial improvements, the default output speed of GPT-6Astra and GPT-6.1Sol was increased; users who log in with a ChatGPT account can also use automatic review without using their plan quota—another intelligent agent checks potentially risky operations.
