The morning briefing.
While you slept: China's AI chip sector pushes independence, OpenAI considers US government stake, and enterprise AI faces new governance challenges.
Geopolitics. China is intensifying its efforts to achieve self-sufficiency in advanced AI hardware, with a new startup emerging to leverage 3D stacking to bypass US export controls. This move highlights the ongoing technological arms race and the strategic importance of chip independence. Meanwhile, OpenAI has reportedly proposed giving the US government a stake in the company, raising questions about national interest and the future governance of frontier AI. The global landscape sees major players like Alibaba taking cautious steps, reportedly banning employees from using Claude Code due to perceived high risks.
Enterprise AI. The focus for enterprises is shifting from simply generating AI output to ensuring its usability, trust, and measurable business outcomes. This evolution is accompanied by growing concerns over data privacy and the control exerted by developers of proprietary AI models, as highlighted by Mistral CEO Arthur Mensch's warning that labs gain intimate access to business processes. In healthcare, the NHS is integrating AI into its app for patient triage, while Australia's government has issued privacy warnings regarding doctors' use of AI scribes.
Hardware Innovation. Significant advancements are emerging in AI hardware, with a Peking University and CAS team developing the world's first neurodynamic chip based on phase-change memristors, offering substantial speed improvements for brain modeling. Huawei is also pushing the boundaries of CPU design with its Tao's Law V2 paper, detailing LogicFolding to dramatically boost Kirin CPU clock speeds. However, the development of AI glasses faces an "impossible triangle" of cost, performance, and battery life challenges at the chip level. The broader AI boom is also reshaping infrastructure, with data centers increasingly locating based on access to cheap power.
AI Agents & Creative Industry. Research indicates that AI search agents primarily fail not in searching, but in their inability to ask clarifying questions when faced with ambiguous queries. This highlights a key limitation in current agent design. Meanwhile, the creative industry continues to grapple with the implications of AI, as Hollywood studios are reportedly divided over ByteDance's Seedance video tool, seeking its ban while some quietly use it. This tension is further evidenced by Midjourney's legal efforts to compel studios to disclose their own AI usage.
Chinese AI Chip Startup Exits Stealth Mode, Bets on 3D Stacking to Bypass US Controls
Dongfang Suanxin, led by industry veteran Wei Shaojun, has launched its corporate website and social media, positioning itself as a new player in China’s AI computing. The company aims to use 3D stacking technology to circumvent US tech export restrictions.
OpenAI Wants to Give the US Government a Piece of the Company
OpenAI's reported proposal to offer the US government a stake in the company raises questions about who should profit from AI. This move could influence national policy and the future of AI governance.
Alibaba Reportedly Bans Employees From Using Claude Code
Alibaba has reportedly classified Claude Code as high-risk software, leading to an internal ban on its use by employees. This action reflects growing corporate caution regarding proprietary AI model security and data handling.
Doctors’ Soaring Use of AI Scribes Prompts Australian Government Warning Over Privacy
The federal health department has raised concerns about AI scribe tools that record, transcribe, and summarize patient-doctor conversations. Regulators are monitoring implementation and potential pitfalls to ensure patient data protection.
NHS to Use AI on Its App to Direct Patients to Appropriate Services
The NHS in England will begin using AI on its app to triage patients and direct them to GPs, pharmacies, or A&E depending on their condition. This initiative is part of a £10bn package to overhaul NHS systems.
Mistral CEO Mensch Says Proprietary AI Models Give Labs a Front-Row Seat to Your Business Processes
Mistral founder Arthur Mensch warns companies against relying on closed AI models, claiming AI labs are storing customer data and have used it to compete with their own customers. Mistral emphasizes EU sovereignty as its strategic edge.
Citi CEO Says Two AI Races Are Shaping Future of Banking
Jane Fraser, CEO of Citi, states that the global financial sector is engaged in two critical AI races. She believes AI will cause job dislocations but also create new positions, though the transition may not be perfectly timed.
The AI Boom Is Increasingly Chasing Cheap Power for Data Centers
The physical location of the internet is being redrawn, with new data-center campuses chosen primarily for access to cheap, plentiful electricity. This prioritizes power over proximity to talent pools, transforming regions like Iowa and Ireland into data-center hubs.
Peking University and CAS Develop World's First Neurodynamic Chip Based on Phase-Change Memristors
A team from Peking University and CAS has created the world's first millisecond-level neurodynamic system chip using phase-change memristors. This innovation is 50-478 times faster than NVIDIA A100 for brain modeling.
Huawei's Tao's Law V2 Paper Details Logic Folding to Boost Kirin CPU Clock Speeds
Huawei's He Tingbo has published the Tao's Law V2 paper, introducing LogicFolding technology. This method boosts Kirin transistor density by 53.5%, enabling 4GHz+ CPU frequencies.
Tsinghua Alums Launch Guangxiang Technology, Raise Hundreds of Millions for Embodied AI in Automotive
Tsinghua-incubated Guangxiang Technology has secured hundreds of millions in angel funding. The startup focuses on developing physics-native embodied AI models specifically for automotive manufacturing.
Hollywood Wants Seedance Banned But Reportedly Also Wants to Keep Using It
ByteDance's AI video tool Seedance is causing a divide in Hollywood, prompting the Motion Picture Association's first cease-and-desist against an AI company. Despite public opposition, studios are reportedly using the tool quietly.
AI Search Agents Fail at Asking Right Questions When Queries Are Ambiguous
A new benchmark, DiscoBench, shows that AI search agents struggle with ambiguous queries, performing worse when searching repeatedly instead of asking for clarification. Accuracy jumps significantly when ambiguity is removed.
AI Glasses Face Impossible Triangle: Cost, Performance, Battery Life in Chip Design
As AI glasses enter mass production, they face critical chip-level challenges including overheating, latency, and battery life. Dedicated SoCs are needed to resolve this "impossible triangle" of design constraints.