Huawei announces new chip plans to compete with Nvidia in computing power race

SoftBank Vision Fund is cutting 20% of its global workforce in its third staff reduction since 2022, now operating with just 44 employees. This restructuring aligns with founder Masayoshi Son's pivot toward high-risk AI investments, including a $9.7 billion stake in OpenAI and a $500 billion "Stargate" data center partnership. The company maintains strong finances with 4 trillion yen in reserves despite previous setbacks with investments like Coupang and Didi. This strategic shift toward leaner operations and concentrated AI ecosystem investments across infrastructure, software, and semiconductors may influence broader tech investment trends.

The Trump administration released "America's AI Action Plan" in July 2025, outlining over 90 federal policy actions to position the US as a global AI leader. The plan focuses on accelerating innovation by removing regulatory barriers, building AI infrastructure through modernized permitting and semiconductor manufacturing support, and strengthening global leadership through technology exports to allies. President Trump emphasized deregulation to advance AI development while implementing procurement guidelines requiring objectivity and preventing ideological bias in AI systems. The administration prioritizes a market-driven approach over centralized regulation and considers winning the global AI race essential for American prosperity and security. A more detailed "2025 National AI R&D Strategic Plan" is forthcoming.

Microsoft will invest a total of $7 billion in Wisconsin for AI data centers, including a second facility to complement its $3.3 billion Mount Pleasant center opening next year. The project will create approximately 800 jobs once both centers are operational. The expanded site will host what's planned to be the world's most powerful AI supercomputer using Nvidia chips, supporting advancements in AI technologies like ChatGPT and innovations in medicine and science. Microsoft is implementing environmentally responsible measures, including pre-paying for electrical infrastructure and using an efficient cooling system that minimizes water usage.

DeepSeek has developed its R1 large language model for just $294,000, significantly less than typical US AI model costs. The model was trained on 512 Nvidia H800 chips available in China despite US export restrictions. R1 is notable for being the first major LLM to undergo peer review, with its methods published in Nature. It uses an innovative automated reinforcement learning approach where it rewards itself for correct solutions rather than imitating human examples. The model has become widely accessible with over 10.9 million downloads on Hugging Face, making it the platform's most popular model. AI researchers praise DeepSeek's transparency and suggest it may influence more cost-effective, open approaches to AI development in the future.