China’s open-source AI edge threatens US position: advisory

A US congressional advisory body said China’s open-source AI ecosystem is giving it a self-reinforcing edge even as US export controls limit access to the most advanced AI chips.

The US-China Economic and Security Review Commission said cheaper models from firms such as Alibaba, Moonshot, and MiniMax rank highly on platforms including Hugging Face and OpenRouter, while industrial digitisation in factories, logistics, and robotics generates data that helps improve models.

The report said Chinese labs have narrowed performance gaps with leading Western models, and it cited estimates that around 80% of US AI startups use Chinese open models, while DeepSeek’s R1 overtook ChatGPT as the most downloaded model on the US App Store, and Qwen has surpassed Meta’s Llama in global cumulative downloads, according to Hugging Face.

The commission said embodied AI could favor China and noted security and bias concerns, while Siemens CEO Roland Busch said he sees no disadvantage in using Chinese open models for industrial automation.

🔗 Source: Reuters

🧠 Food for thought

Implications, context, and why it matters.

The cost and performance of Chinese models are forcing a market reckoning

“Cheaper” does not capture how hard low-priced models are squeezing budgets.
DeepSeek’s R1 reasoning model wiped more than $750 billion off the S&P 500 index after it challenged the belief that AI leadership depends on massive hardware spending 1.
Pricing gaps can be stark. Moonshot AI’s Kimi K2.5 model ran at about one-seventh the price of Anthropic’s comparable Claude Opus on some early benchmarks 2.
For AI startups, the difference can decide survival. Tight cash flow pushes teams toward Chinese models when costs fall that far 3.

Siemens’ strategy reveals a pragmatic, not political, approach to industrial AI

A Siemens CEO quote can sound partisan, yet the company is spreading bets across platforms.
Siemens is widening its partnership with Alibaba Group 4 while also growing its work with NVIDIA on the Industrial AI Operating System 5.
The plan centers on a tailored “Industrial Foundation Model.” The idea is that models like ChatGPT do not grasp the language of engineering and manufacturing 6.
The approach fits a wider shift in industry. Manufacturers mix the best tools they can get, no matter where they come from, to tackle factory-floor tasks without ripping out whole systems 7.