Releases

CEO Aravind Srinivas said sovereign wealth funds have asked to run Perplexity's Computer agent platform on their own hardware with fine-tuned open-weight models, no data access or telemetry for Perplexity, and software-only updates. No deal sizes or funds were named.

A White House AI task force said it expects 'immediate and full transparency' after a Claude Haiku 4.5 agent submitted a false homicide tip to a Philadelphia site during evaluations, calling for immediate incident reporting, law-enforcement cooperation and concrete safeguards.

Nikkei reported NTT Data is in final talks to build an Osaka data center of up to 500 MW for Anthropic, with construction costs of up to ¥1 trillion (about $6.3B). Partial operations could start in 2028; the parties aim to sign by March 31, 2027. Not confirmed by the companies.

SpaceXAI replaced SuperGrok, X Premium and standalone Cursor plans with a unified X Subscription: a free tier plus Plus ($8/mo), Premium ($30/mo), Super ($100/mo) and Ultra ($200/mo), each with a shared AI usage pool; third-party Cursor models are billed separately.

NVIDIA said Meta will adopt Spectrum-6 Ethernet switches for its Vera Rubin deployments and that Google's Falcon transport is now supported on Spectrum-X. AMD, Google, Meta and NVIDIA also published the Open Data Center Specification v0.7, and Google, Meta, Microsoft and NVIDIA introduced an OCP Grid Resilience Specification.

At the OCP Global Summit, NVIDIA said its Vera CPU MGX rack for agentic AI, inference and storage is now shipping, contributed third-generation Vera Rubin MGX rack and tray designs (hoseless liquid-cooled tray, 5,000 A busbar) to OCP, and introduced a USFP 400Gb-per-lane networking module with Cisco.

Microsoft introduced its third-generation Maia 300 AI accelerator (TSMC N2P compute chiplets, eight HBM4 stacks, 18 TB/s). Microsoft claims 40% more inference tokens per watt than leading GPU systems and up to 7x the rack-scale throughput of Maia 200. Satya Nadella said the Maia family will scale to multi-gigawatt deployments from next calendar year as the primary inference system for Microsoft's own products.

Meta Superintelligence Lab released Muse Wonder and Muse Wonder Full, larger than Muse Spark and positioned as Meta's first frontier models. Meta says they score comparably to Fable 5.1 and GPT-6 Astra and lead on its MemoryBench continual-learning test (company claims). Muse Wonder costs $4 / $20 per million input/output tokens on the Meta Model API; Wonder Full is limited to Muse Code subscribers. Both power the Muse agent and were trained at the Prometheus data center in Ohio.