Home › News › AI Daily October 21 Evening Edition

AI Daily | October 21, 2026 Evening Edition: CNCC2026 Opens in Chengdu with Embodied Intelligence as a Core Topic, Zhou Xiaochuan Warns of AI Infrastructure Financial Risks

Published: 2026-10-21  |  Category: AI Daily  |  Read: about 8 min

Top Stories This Evening

  • The 23rd China National Computer Conference (CNCC2026) opens in Chengdu, with embodied intelligence elevated to a core topic for the first time
  • NVIDIA GTC Berlin closes: the Isaac platform is fully upgraded and the open-source robot operating system IsaacOS is released
  • Shengshu Technology begins a full public livestream of Motubrain 2.0 training, lasting half a month
  • China Telecom-incubated embodied intelligence company Gamma Robotics is registered in Xuhui, Shanghai
  • Zhou Xiaochuan warns: massive capital flows into AI infrastructure, and failed investments could trigger financial turmoil
  • Ramp completes a $1.85 billion funding round at a $60 billion valuation, as the AI financial management track heats up
  • Amazon's 2026 capex is expected to reach $220 billion, with data centers facing power supply bottlenecks

1. CNCC2026 Opens in Chengdu: Embodied Intelligence Elevated to a Core Topic for the First Time

CNCC2026Embodied IntelligenceAcademic Conference

On October 21, the 23rd China National Computer Conference (CNCC2026) officially opened at the Western China International Expo City in Chengdu — the first time in more than two decades of CNCC's history that it has landed in Sichuan. The theme is "Data-Driven Intelligence, Computing the Future," and the event runs from October 21 to 24.

Notably, embodied intelligence has been elevated to a core topic, standing alongside AI foundation models and world models, marking that robotics-related research has officially moved out of the niche automation track and into the mainstream computer science view. The conference features 17 invited talks, 3 plenary forums, 167 special forums and more than 40 special events, covering 11 major themes including AI foundation models, machine learning theory, reinforcement learning and inference, vision, multimodality and world models, AI safety and governance, and robotics and embodied intelligence. The concurrent CCF Computer Exhibition covers 15,000 square meters, with 131 leading industry companies confirmed to exhibit.

Editor's note: Embodied intelligence moving from the margins to the center of the mainstream academic stage reflects the Chinese computer science community's strong consensus on the trend of "AI moving from the virtual into the physical world." Sichuan Province has made artificial intelligence its No. 1 provincial innovation project, also showing local governments' active positioning in the AI race.

2. NVIDIA GTC Berlin Closes: Isaac Platform Fully Upgraded, IsaacOS Released

NVIDIAGTC BerlinRobotics

From October 20 to 22, NVIDIA GTC 2026 Berlin took place — the AI Super Bowl's first stop in Europe. The conference held in-depth discussions on core tracks including physical AI and robotics, generative and agentic AI, AI factories, and digital twins and accelerated computing.

The biggest announcement was the full upgrade of the Isaac platform: on one hand, IsaacOS was officially released as the world's first open-source robot operating system, integrating the ROS2 ecosystem, real-time kernel scheduling, low-latency AI inference pipelines and safety-certified HALOS interfaces into a unified platform, so developers can launch digital-twin simulations with a single command; on the other hand, the GR00T N1.7 model continues to iterate, and together with the Jetson Thor edge AI chip (2070 FP4 TFLOPS) provides a complete "brain plus body" development chain for humanoid robots. The reference humanoid robot system is built by NVIDIA together with Unitree Robotics and Sharpa, and is expected to launch in Q4 2026.

Industry significance: NVIDIA is pushing AI compute down from the data center into the physical world. The open-sourcing of IsaacOS means robot development is moving from "each doing its own thing" to a "unified operating system," which is expected to greatly lower the R&D barrier for humanoid robots and accelerate the entire process from simulation training to real deployment.

3. Shengshu Technology Livestreams the Full Motubrain 2.0 Training Process

Shengshu TechnologyPhysical AIOpen Source

On October 21, the world-model team at Shengshu Technology officially began a full public livestream of Motubrain 2.0 training, planned to last half a month, showing the public the complete process of an embodied model from training and testing to capability iteration. It is the first large-scale public disclosure of a full training process in the Physical AI field.

The livestream covers four stages: pre-training (observing how training scale affects zero-shot generalization), post-training (improving execution on specific tasks), agent task execution (exploring experience transfer), and capability evaluation. Motubrain 2.0 will be open-sourced soon, and the team chose to disclose the training process before the official release, including failures, trial and error and adjustments, making model R&D more transparent.

Editor's note: Following Xiaomi MiMo's public reinforcement-learning training, the embodied intelligence field has reached a milestone of "transparent training." This open approach helps build public trust in AI and provides the industry with a methodology reference for Physical AI training.

4. China Telecom-Incubated Embodied Intelligence Company Gamma Robotics Lands in Shanghai

China TelecomEmbodied IntelligenceHumanoid Robots

On October 21, Gamma Robotics, an embodied intelligence company incubated by the China Telecom Artificial Intelligence Research Institute (TeleAI), was registered in Xuhui District, Shanghai, and moved into "Model Space." It is the first market-oriented innovation company incubated by TeleAI.

Gamma Robotics focuses on full-model innovation and application of self-evolving humanoid robots and will build a humanoid robot laboratory with TeleAI. The company is strategically invested by the China Telecom (Shanghai) AI Industry Private Equity Fund; Professor Li Xuelong, CTO of China Telecom Group, serves as scientific advisor, and Dr. Bai Chenjia serves as CEO. The fund is capitalized by the central SOE, the Shanghai Guotou Guiding Fund, the national strategic emerging industry fund and Xuhui District, forming a multi-tier capital structure of "central SOE + local state capital + national strategic emerging industry fund." Gamma Robotics will focus on advancing R&D of "self-evolution" and "whole-body models" for humanoid robots, enabling robots to keep learning through real interaction.

5. Zhou Xiaochuan Warns: Failed AI Investment Could Trigger Financial Turmoil

Financial RiskAI InvestmentBund Annual Conference

On October 20, Zhou Xiaochuan, former governor of the People's Bank of China, issued a warning at a roundtable on "AI and the Productivity Revolution" at the 2026 Bund Annual Conference: massive capital is currently flowing into AI and related infrastructure, and "once part of it fails, it could trigger financial turmoil and brew a crisis."

Zhou divided AI investment into three categories: large models and compute data center infrastructure, the AI application layer, and supporting industries such as semiconductors and power. He judged that productivity gains mainly come from the application layer, while compute infrastructure is the riskiest segment. Data shows the five major overseas tech giants plan nearly $700 billion in AI capex for 2026, and McKinsey estimates global AI data center investment could reach $5.2 trillion from 2025 to 2030, far exceeding the scale of the 2000 internet bubble. He called for "conducting research and preparing contingency plans, and stocking corresponding policy tools in the policy toolbox."

Risk note: The key phrase in Zhou Xiaochuan's remarks is not "investment failure" — it is normal for a batch of companies to exit during a technological revolution — but "bailing out financial institutions." This implies AI investment risks could be amplified through credit chains and financial institutions' balance sheets, repeating a systemic risk transmission path similar to 2008.

6. Ramp Completes $1.85 Billion Funding at a $60 Billion Valuation

RampFundingAI Finance

Enterprise expense management platform Ramp announced the completion of a $1.85 billion new funding round, with its valuation rising to $60 billion, doubling from $44 billion in four months. The round was co-led by Dragoneer, Thrive Capital and Founders Fund, bringing cumulative funding to about $5 billion.

Ramp's annualized revenue has topped $2 billion ($1.5 billion four months ago), its customer count has grown from 70,000 to 85,000 (including Visa, Uber, Shopify and Figma), and annualized procurement volume exceeds $275 billion (up 175% year on year). The company is expanding comprehensively around AI: launching a corporate credit card designed for AI agents, entering model routing (saving up to 40% on inference costs), and releasing the Glass collaboration space. Founder Rabois said bluntly: "We want to compete with Anthropic for token spending management."

7. Amazon's 2026 Capex Expected to Reach $220 Billion

AmazonCapexData Centers

According to financial disclosures and multiple reports, Amazon's 2026 capital expenditure is expected to reach $220 billion, pushing free cash flow negative. Meanwhile, some AWS data centers face power supply bottlenecks, constraining the pace of compute expansion.

The figure reflects the intensity of the AI infrastructure "arms race." The five major cloud providers (Amazon, Microsoft, Google, Meta and Oracle) are expected to spend a combined $769.2 billion in capex in 2026. However, data center construction faces not only funding pressure but also multiple constraints including power supply, land approval and environmental assessment. Amazon's power bottleneck shows that physical infrastructure limits are becoming a key bottleneck for AI expansion.

Today's AI Data at a Glance

MetricDataChange
CNCC2026 exhibition area15,000 sq m131 companies exhibiting
NVIDIA Jetson Thor compute2070 FP4 TFLOPSEdge AI chip
Motubrain 2.0 livestream durationHalf a monthFull process public
Gamma Robotics registered capitalNot disclosedCentral SOE + local + strategic emerging fund
Global AI data center investment (2025-2030E)$5.2 trillionMcKinsey estimate
Ramp annualized revenue$2 billionUp 33% in 4 months
Amazon 2026 capex$220 billionFree cash flow negative
Five major cloud providers' 2026 capexAbout $769.2 billionAI arms race

Sources

  • China Computer Federation (CCF) - official CNCC2026 information
  • NVIDIA official - GTC Berlin 2026 agenda and announcements
  • Leiphone, ifanr - coverage of Shengshu Technology Motubrain 2.0
  • The Beijing News, Jiefang Daily - coverage of Gamma Robotics' founding
  • 21st Century Business Herald - Zhou Xiaochuan's remarks at the Bund Annual Conference
  • IT Times - coverage of Ramp's funding
  • SegmentFault AI Daily, Sina Finance - comprehensive coverage