Alibaba CEO Wu Yongming: machines will become primary source of thinking; DeepSeek hires 90s CFO ahead of IPO

Alibaba is assembling three key components—AI models, AI chips, and cloud computing—to pave the way for the "Machine Intelligence Era."

At the 2026 Cloud Summit, Alibaba Group CEO Eddie Wu used the story of Thomas Edison building the first power station in 1882 to remind the audience that the "light bulb" of the machine intelligence era is already lit, but the "appliances" that will truly transform daily life have yet to arrive. His assessment was even more striking: machines are becoming the primary drivers of thought, intelligence is becoming a commodity that can be supplied at scale, and in the future, the total volume of machine-generated thinking will be more than 1,000 times that of humans. Machines will handle 99.9% of the world's thinking. The three core infrastructures supporting this vision—AI models, AI chips, and AI cloud—are now being deployed one by one.

Qwen Team Achieves Breakthrough with Zero Human Intervention

The Qwen team was the first to showcase progress. Qwen3.8-Max, with zero human involvement, autonomously built its training pipeline, constructed training data, designed experiments, identified defects, and iterated continuously for over a month, completing 33 effective iterations. On the Artificial Analysis leaderboard, its score rose from 40 to 45, placing it in the top tier alongside Claude and GPT's strongest models.

Even more noteworthy is the breakthrough in chip-model synergy: Qwen3.8-Max autonomously adapted and optimized its inference framework on a previously unseen GPU from T-Head, achieving a 96% improvement in single-instance inference throughput. In chip design scenarios, it ran autonomously for over 60 hours, called EDA tools more than 10,000 times, and completed a physical implementation optimization that reduced chip area by 42%. The flywheel of models designing chips and chips feeding back into models has begun to spin.

T-Head Unveils New AI Chip, Alibaba Cloud Sets Ambitious Targets

T-Head released its next-generation training-inference integrated AI chip, Zhenwu V900, at the summit. Its performance is three times that of the Zhenwu M890, featuring 216 GB of high-capacity memory and 1200 GB/s of inter-chip interconnect bandwidth. Built on the proprietary ICN Switch interconnect chip, a single cluster can scale to 500,000 cards. The chip is scheduled to enter mass production and sales in Q1 2027.

Meanwhile, Alibaba Cloud announced a target to operate global data centers with a total capacity exceeding 20 terawatts by 2032. From 33 iterations with zero human intervention to a 500,000-card cluster, the Cloud Summit demonstrated the process of turning the "power grid" from metaphor into reality.

Eddie Wu: Machines Are Becoming the Primary Thinkers, Intelligence Will Be a Scalable Commodity

In his opening speech at the 2026 Cloud Summit, Eddie Wu first addressed the conventional understanding of AI. He argued that viewing AI as "artificial intelligence" may underestimate machine intelligence—just as early steam engines replaced humans and horses for pumping and hauling, but soon machine power far exceeded what humans and animals could do, creating an entirely different species. Today's machine intelligence is not a replacement for human intelligence but a fundamentally different way of working.

Wu put forward two key judgments:

1. The total volume of machine-generated thinking will be more than 1,000 times that of humans. Currently, machine thinking accounts for less than 3% of total human thinking, meaning there is room for tens of thousands of times more growth in scale and demand. He cited the example of rare diseases like neonatal progeria, which may have only a few hundred recorded patients globally, making it difficult to secure sufficient research resources. But when AI possesses professional research capabilities, as long as computing power is abundant and cheap enough, millions of AIs could simultaneously conduct research in every niche field. Top-tier thinking will transform from a luxury into a commodity that can be supplied at scale.

1. The representative product of the machine intelligence era has not yet arrived. Today's AI coding is very much like the early light bulb of the machine intelligence era—it replaces tasks that humans deliberately perform. But replacing existing work cannot create a truly new era.

On the strategic front, Wu made clear that Alibaba will firmly invest in three core infrastructures: AI models, AI chips, and AI cloud.

• AI Models: The technical path toward superintelligence (ASI) is becoming clearer, with recursive self-improvement at its core. The Qwen team is advancing research on model architecture and data optimization, with plans to train entirely new models.
• AI Chips: T-Head has fully deployed GPUs, CPUs, smart NICs, and ASIC interconnect chips. Starting this quarter, supernodes based on the M890 have been deployed at scale in Alibaba's data centers. The newly released Zhenwu V900 is described as the most powerful AI chip currently available in China.
• AI Cloud: Wu revealed that customer demand is extremely strong, with Alibaba Cloud revenue continuing to accelerate. However, medium- to long-term industry demand far exceeds Alibaba Cloud's current supply capacity. The goal is for Alibaba Cloud to operate global data centers with a total capacity exceeding 20 terawatts by 2032.

Yan Wentao Officially Joins DeepSeek as CFO

DeepSeek ends its three-year CFO vacancy as a second round of funding approaches.

On Monday, September 21, Yan Wentao, a former partner at Hillhouse Capital's venture arm (born in 1991), officially joined DeepSeek and reported to the company. The 90s-born investor, a graduate of Fudan University, ends DeepSeek's three-year period without a CFO. His career includes stints at Tencent Investment and H Capital before joining Hillhouse Capital in 2020, where he quickly became a representative young investor. He participated in investments in frontier tech companies such as MiniMax, Zhipu AI, ByteDance, J&T Express, and Webull. During his student years, he founded a campus discount platform covering 14 universities in Shanghai, achieving monthly profits exceeding 10,000 yuan and even securing funding.

Yan's arrival comes as DeepSeek's second round of funding nears completion. In June of this year, DeepSeek completed its first funding round of approximately 50 billion yuan, with a post-investment valuation exceeding 350 billion yuan. Liang Wenfeng personally contributed about 20 billion yuan. In July, news of a second funding round emerged, with a pre-investment valuation rising to 500 billion yuan. This round is expected to introduce new investors, primarily industrial investment institutions, including Shixi Capital and SMIC Juyuan.

According to The Information, DeepSeek recently held an investor meeting with extremely strict confidentiality requirements. Participants had to travel to offices in Beijing or Hangzhou, and were required to surrender their phones, electronic devices, and bags during the meeting. Multiple sources close to the deal revealed that over the past two weeks, DeepSeek has conducted a rigorous qualification review of its LP list, with some LPs—including state-owned capital, non-listed companies, and individual investors—being removed. "Everyone is now waiting for the qualification review results."

All signs point to DeepSeek's second funding round being finalized soon. The company is also reportedly preparing for an IPO on the STAR Market (Shanghai's科创板). Although the process is still in its early stages, the market already sees a trillion-yuan valuation for this former leader of China's open-source large model wave. The arrival of this 90s-born CFO has also sent shockwaves and inspiration through the many young professionals in the primary market.

This article was originally published on ebrun.com.

Disclaimer: This weekly report is entirely AI-generated. Please verify the information. If there are any errors or misunderstandings, please feel free to point them out. If you find any issues, please email ai-content-feedback@ebrun.com.