DeepSeek reportedly plans to train its next V5 model on Huawei Ascend 950 chips in 2027
Digitimes reports the plan would extend the companies' partnership from inference to full model training; it is a report, not an announcement by either company.
- Tencent releases Hy4, adding visual reasoning to its flagship open-weight model
- LG's EXAONE Nexus checks AI training data for copyright risk 45 times faster than lawyers
- Claude Code head of product Cat Wu leaves Anthropic weeks before its planned IPO
- Revolut says self-hosted GLM models cut its AI coding costs by about 90%, replacing Claude Code as default
- iOS 27.3 beta brings Apple Intelligence to China, built on Alibaba's Qwen models
- Altman says people must not cede control of the future to superintelligence
- NBC exit poll: 57% of US midterm voters say AI will hurt society, 76% oppose a local data centre
- Brazilian court bars Meta from scraping or training AI on news outlets' content without permission
- Democrat Gina Hinojosa wins the Texas governor's race after pledging a moratorium on new data centres
- Oracle cancels a $7bn AI chip leasing deal with Tencent under US pressure, FT reports
- Chinese buyers expect to receive only about 1% of $70bn in Nvidia RTX Pro 5500 orders, Bloomberg reports