AI news, 16 August: Nvidia in talks for $3 billion OpenAI data-center stake

Nvidia eyes $3 billion investment tied to massive OpenAI campus. According to a report by The Information, picked up by Reuters, Nvidia is in talks to invest as much as $3 billion in SB Energy, the SoftBank-backed developer of a huge planned data-center project in Ohio for OpenAI. The stake is being discussed as part of broader negotiations in which Nvidia would provide around $100 billion in credit support for the campus; half the equity investment could come at deal signing and half around SB Energy’s planned IPO. Nvidia and SB Energy did not immediately comment. The talks underscore the enormous capital and power needs of frontier AI and Nvidia’s deepening role not just as chip supplier but as financial backer of the infrastructure boom.

SpaceX completes $60 billion Cursor acquisition. SpaceX has officially closed its acquisition of AI coding startup Cursor (Anysphere), according to a company blog post and reports from TechCrunch and Bloomberg. The all-stock deal, valued at roughly $60 billion when advanced earlier this year, gives Cursor access to what it calls “the largest fleet of GPUs in the world” inside SpaceX’s growing compute operations. Cursor said the move will help scale useful intelligence on top of that capacity. The close marks one of the largest-ever tech acquisitions and further knits AI coding tools into Elon Musk’s broader SpaceX/xAI ecosystem.

Anthropic details Claude’s invisible text watermarks. Anthropic published a fresh explanation of how it will watermark text from Claude models to meet EU AI Act transparency rules. The company said it is using a SynthID-Text-style approach that embeds an undetectable pattern for holders of a detection key, plans a public detection API, and noted that light editing often leaves the mark intact while a full rewrite removes it. Code is less affected because of stricter constraints. The move, applied globally on newer models, has sparked debate; some users have publicly said they are canceling subscriptions, while Anthropic stressed quality is unchanged and other major labs are expected to follow similar practices.

Apple builds China-specific AI model with Alibaba help. Apple has trained its own large language model tailored for the Chinese market with support from Alibaba, three people familiar with the matter told Reuters. The step departs from Apple’s earlier reliance mainly on third-party Chinese models and aims to give it more control as Apple Intelligence rolls out in China in the coming months after regulatory clearance. Alibaba’s Qwen is also set to be integrated. Neither company commented. The dual approach navigates strict Chinese rules and competition from local device makers, and would make Apple the first foreign firm approved to offer a proprietary model there amid U.S.-China tech tensions.

IJCAI-ECAI 2026 workshops open in Bremen. The world’s premier joint AI research conference, IJCAI-ECAI 2026, began its workshop and tutorial program on 15 August in Bremen, Germany, with the main technical sessions starting later in the week. Thousands of researchers are gathering to present and debate the latest advances, underscoring Europe’s continuing role as a hub for foundational AI science even as commercial development accelerates elsewhere.