_________

The Chinese AI and cloud provider Alibaba has deployed a large model in space. "Qwen-3" has been uploaded to a cluster of several satellites and carried out multiple end-to-end inference tasks while in orbit, reported Chinese media.

The deployment happened in November 2025, but was reported for the first time now, after experts discussed it during a symposium about computing in space organised by the China Academy of Information and Communications Technology (CAICT), the reports said.

The satellite-based LLM is part of an ambitious Chinese programme to harness solar power in space and use it to power notoriously energy-hungry data centers directly in orbit. Guoxing Aerospace Technology (Adaspace), which is running the current pilot, plans to launch a total of 2,400 inference satellites into low-Earth orbit by 2035.

The "AI race" between the United States and China has now extended to outer space. A SpaceX rocket deployed the Starcloud-1 satellite with Nvidia GPUs in space last November. "It's a no-brainer building solar-powered data centers in space; the lowest-cost place to put AI will be space, and that will be true within two years, three at the latest," SpaceX founder Elon Musk said at the recent World Economic Forum in Davos.

China agrees and has now started to implement a similarly large project to build "giga-watt-class" hubs in an increasingly crowded orbit. To understand both the economic chances and military risks of this new AI race in space, read this week's "Must read" by clicking here:

_________

The Shanghai-based large-model startup StepFun raised more than RMB 5 billion (around US$700 million) from investors in its latest funding round, the largest to date.

It was the largest financing round for Chinese companies competing with the likes of OpenAI or Anthropic in the United States over the past twelve months.

To explore the logic behind this large valuation, the signals it sends about investors’ confidence in foundational models, and the background of StepFun, click here for our Highlights page:

With most industrial robots still of the programmed type, suitable only for clearly defined and repeatable tasks, and most smart robots relatively feeble, the startup Galbot from Beijing is now trying to shake things up with its new model, Galbot S1.

The industrial robot falls into the category of “dual-arm mobile manipulators,” with an upper body resembling a humanoid and a lower body made of a sturdy metal base.

With a continuous dual-arm payload of 50 kg, its makers claim that it breaks long-standing payload limits in the industry, and it operates autonomously.

There is a certain fatigue among observers watching all those dancing humanoids, while cobots and other workhorses of manufacturing still await smart AI features.

The Galbot S1 promises to bring embodied AI to production lines. First models are already deployed in a battery factory operated by CATL.

Baichuan Intelligence, a Beijing-based startup, unveiled a new medical large language model, Baichuan-M3 Plus.

The company, which reportedly received around US$691 million in funding in 2024, claims that this “Plus version” offers improved accuracy over its previous M3 model.

The hallucination rate is said to have dropped from 3.5% in the previous M3 model to 2.6%. This would reportedly be lower than that of OpenEvidence and more than 30% lower than that of GPT-5.2.

The new vertical AI model also introduces “evidence anchoring,” linking every AI-generated medical conclusion to a corresponding paragraph in a medical research paper.

Medical AI is a field where the fear of hallucinations is particularly acute, as errors based on false or fabricated findings can affect patients’ lives.

Doctors and hospitals are likely to adopt such AI tools only once they become more accurate and more transparent.

Moonshot AI, another Chinese startup developing large AI models, has released an updated version of its flagship multimodal model Kimi.

The open-source model is called Kimi K2.5, was trained on 15 trillion mixed text and image tokens, and can process text, images and videos as well as agentic workflows with multiple agents working in parallel (an "agent swarm").

Initial tests reportedly show that Kimi K2.5 beats GPT-5.2 and Gemini 3 Pro in some benchmarks when it comes to coding, such as the "SWE Bench Multilingual."

When handling agents, it reportedly even outperforms Claude Opus 4.5 and GPT-5.2 in some benchmarks.

Mirroring a worldwide fascination with agentic AI, there is now an arms race between Chinese AI companies to launch agents for everything from coding to copywriting and video creation with as little human intervention as possible. Moonshot AI is backed by Alibaba, among others, which holds around 36% of its shares.

With Qwen3-Max-Thinking, Alibaba Cloud released its latest flagship reasoning model on January 26, 2026.

Chinese media call it the foundational model that comes closest to international top models to date.

Chinese tech media outlets report that it has achieved top scores in 19 benchmark tests, including in the areas of complex reasoning, factual knowledge, and agent capabilities.

Its overall performance is said to be comparable to GPT-5.2, Claude Opus 4.5, and Gemini 3 Pro.

With more than one trillion parameters, 36 trillion pretraining tokens, and a context window of 262,000 tokens, Alibaba calls it "China's strongest AI model."

Should the touted benchmark records be confirmed, this “Ferrari” among China’s large models would create new competitive pressure on US rivals such as Anthropic, Google, and OpenAI.

_________

_________

The number of users of generative artificial intelligence in China had reached 602 million as of December, 2025, an increase of 141.7% compared to the end of 2024.

Source: China Internet Network Information Center (CNNIC)

The number of AI-related companies in China exceeded 6,000 in 2025.

Source: Ministry of Industry and Information Technology (MIIT)

China's core AI industry is projected to exceed 1.2 trillion yuan (about €146 billion) in 2025.

Source: Ministry of Industry and Information Technology (MIIT)

850 AI-related financing deals in China in 2025, with a total value of RMB 69.23 billion (about €8.5 billion) .

Source: The Beijing News

_________

On January 21, 2026, the provincial government in Guangdong issued the “Policy Measures on Artificial Intelligence Empowering High-Quality Development of Transportation in Guangdong Province.”

Policy support for the application of AI in the transportation sector. One focus area is freight transport with unmanned vehicles. Another is joint platforms for logistics companies and railways.

Authorities in Sichuan named artificial intelligence as one of 25 emerging fields that deserve priority support in a "Notice on Key Directions for New Industrial Tracks in Sichuan Province" published on January 19, 2026.

Chosen as one focus area among others like biotechnology, the low-altitude economy, or intelligent connected vehicles, artificial intelligence and foreign companies active in this segment can expect more local government support and favourable investment rules.

The Provincial Department of Economy and Information Technology in Zhejiang, in eastern China, released a draft of its new “15th Five-Year Plan for the Digital Economy and Digital Infrastructure" on January 9, 2026.

Already home to promising AI startups such as DeepSeek and industry giants such as Alibaba, Zhejiang province plans to build three "digital industry clusters" with a strong focus on AI. The province seeks to leverage its initial success, hoping to attract more investment and talent to win China's internal AI race against Shenzhen and Beijing.

On January 5, 2026, Beijing released the “Action Plan for Building a "Beijing Artificial Intelligence Innovation Hub” that clearly states the ambition to scale AI-related industries and services beyond a total market size of RMB 1 trillion within two years.

The Chinese capital has decided to position itself as a "global AI innovation hub" and in this plan lays out a clear roadmap for the buildout of computing clusters, the scaling of unicorn startups, and other measures to achieve this goal.

The Reuters news agency reported on January 28, 2026, that China is considering giving the green light to three of its largest tech companies to buy Nvidia's H200 chips, quoting unnamed sources.

While the official decision has not been announced, early signals indicate that Beijing might give conditional approval for the import of the previously banned AI chips. There may be caps on the total number of imports or provisions on matching purchases from Nvidia with orders from domestic chipmakers.

_________

Even many Chinese observers, technology-loving as they may be, have had enough of dancing, flip-flop-jumping humanoid robots on every channel.

"We do not need a million dancing humanoid robots!" read a typical headline in China recently.

However, while the hype about humanoids as entertainers is in full swing, a quiet revolution has begun in China's factories.

Humanoids are showing up at production lines in battery factories of CATL, where they perform complex and dangerous tasks.

Other two-legged robots help assemble EVs or washing machines. Their numbers are still small, but predicted to grow exponentially from now on.

For the full analysis of this first phase of validation for humanoids as industrial tools, please click here to read our complete Deep Dive.