AMD announced a deep collaboration with Anthropic, under which Anthropic will procure up to 2 gigawatts of AMD Instinct MI450 computing capacity, while AMD will invest up to $5 billion in Anthropic based on milestone achievements. This agreement marks Anthropic's move to diversify its chip supply dependencies and reflects how AI competition is evolving from model capabilities to comprehensive battles involving infrastructure and hardware integration.

OpenAI has officially launched ChatGPT Voice on its desktop application, allowing users to control their computers via voice and coordinate multiple AI agents through GPT-Live. The feature enables cross-task execution within a single conversation—launching tasks across Codex and ChatGPT Work, checking threads, and scheduling workflows—to support continuous operations from drafting documents to engineering follow-ups. Greg Brockman noted that voice interaction is more natural than typing.

Anthropic announced a major upgrade to its Claude Voice mode, now running on the more powerful Claude Opus and Sonnet models. It can also access connected tools such as email and calendar during conversations. The update is rolling out publicly across all platforms and adds support for multiple new languages including Spanish, French, Hindi, and Japanese.

OpenAI announced the launch of ChatGPT Health for U.S. users, enabling secure connections to medical records and Apple Health data to provide personalized context. OpenAI outlined privacy safeguards and functional capabilities, and referenced the upcoming release of GPT-5.5 and GPT-5.6 models, focused on improving medical accuracy.

Uber co-founder Travis Kalanick's industrial robotics company, Atoms, has emerged after nearly eight years in stealth, raising $1.7 billion in equity funding led by Andreessen Horowitz, with Uber also appearing on the shareholder list. The company focuses on specialized automation machines for mining, logistics, and food production, rather than humanoid robots.

The U.S. Treasury has warned of potential sanctions against Chinese AI company Moonshot, following allegations by government officials that it improperly distilled Anthropic's Claude Fable 5 model. Its claim of discovering a Redis zero-day vulnerability has also been questioned as a misinterpretation of publicly available information. Industry figures like Suhail have criticized the feasibility of regulating model distillation, advocating instead for investment in open-weight U.S. models to ensure security.

Microsoft AI CEO Mustafa Suleyman revealed progress on the MAI model series. MAI-Voice-2-Flash is now in public preview, offering twice the speed and a 32% price reduction to $15 per million characters, with GPU costs cut by up to 89%. It is already deployed in Dynamics 365 Contact Center. The image model MAI-Image-2.5-Pro is launching in preview on Azure Foundry, delivering high-fidelity generation and precise in-image text rendering. MAI models are now integrated into products such as PowerPoint, Bing, and Dragon Copilot.

OpenAI announced an increase in its AI infrastructure investment plan from $600 billion to $750 billion, aimed at strengthening autonomous data center construction. Meanwhile, TSMC is adding $100 billion in investments, bringing its total in Arizona to $265 billion, reflecting the massive capital and energy demands driven by AI compute expansion. BloombergNEF predicts that by 2035, U.S. data centers will consume one-fifth of the nation's electricity.

Andrew Ng announced the launch of OpenWorker, an open-source AI agent capable of generating refined documents, sending Slack messages, and updating calendar entries by operating across user files and tools. It focuses on directly delivering completed work outputs rather than merely providing conversational assistance.

An empirical study based on 33,596 pull requests (PRs) and 2,807 repositories investigates concurrent AI coding agent PR submissions for the first time. It found that 40.2% of repositories exhibit overlapping PR timing from agents, with such concurrent pairs accounting for nearly 80% of all agent PR activity. Text conflicts in cross-agent PRs reached 41.7%, more than double the 19.8% rate in single-agent scenarios, and approximately 84.4% of conflicting files were source code.