China's new five-year plan for information and communications sets a 2030 target of 9,800 Eflops in intelligent computing capacity, with clusters of at least 100,000 accelerators.
South Korean data puts a DRAM 1b wafer at $0.654 per square millimetre, compared with $0.424 for a TSMC N2 wafer. The AI-driven memory shortage has inverted standard pricing logic.
At the Yunqi conference on September 22, Alibaba announced a chip it calls the most powerful in China. A day later, T-Head expanded its open-source stack with migration tools and acceleration libraries.
An OpenAI agent accessed a health data portal run by the Australian government. The government learned of the breach months later, a development that complicates the AI safety debate.
The V4.1-Flash repository and weights are under the MIT license. On 31 August 2026, DeepSeek opened its first multimodal model, Flash-Vision-Exp, in the same way.
Digitalcourage awarded the Big Brother Awards 2026. The EU Commission, Interior Minister Dobrindt, Deutsche Telekom, Fielmann and Apollo, and the Evangelical Northern Church faced criticism.
After Microsoft cut off the ICC chief prosecutor, The Hague launched the DAWO project. Eight municipalities are already testing NixOS on old laptops that Windows 11 refuses to run.
Since 2 August, creators of generative systems must mark AI content in a machine-readable way. Anthropic is extending watermarking to more Claude models, but the tool has limits.
Alibaba's chip unit T-Head is expanding SAIL, its open software package, and rolling out the new Zhenwu V900 chip. The Chinese ecosystem is trying to loosen its dependence on CUDA.
Changan targets mass production of humanoid robots by 2028 and a robot that builds robots. In Hangzhou, the robot school issued its first diplomas after practical trade tests.
Measurements and simulations show that waste heat from large data centers noticeably warms the surrounding area. Germany and France already require operators to reuse it. At EU level, operators only have to check whether reuse is possible.
A Columbia economist estimates the US AI investment effort at 3.6% of GDP, nearly double the previous record. Financing structures are becoming increasingly opaque.
Chinese operator Meta-Infer pushed DeepSeek-V4.1-Flash throughput from 1,932 to 13,274 tokens per second on eight PCIe cards. The model stayed the same. Only the software stack changed.
At the Yunqi Conference, Alibaba announced the Zhenwu V900 chip, billed as the most powerful Chinese AI accelerator. A day later, T-Head released the components of its SAIL software stack as open source.
Following layoffs at the US ocean and atmosphere agency, European institutes are taking over copies of key datasets. A 2025 survey found that 69 percent of German users have doubts about US cloud services.
Enisa's report identifies ransomware as the threat with the highest short-term impact in the Union, noting that 57% of analysed incidents were driven by ideological motives.
At the Yunqi Conference, Alibaba's CEO presented the Zhenwu V900 as China's most powerful AI chip, three times faster than the M890. A day later, subsidiary T-Head expanded access to its SAIL software stack.
Unitree's founder says the barrier is matching a model to physics, with errors running to a few millimeters. In the first half of 2026, 35,000 quadruped robots shipped worldwide, three quarters of the whole of last year.