Google · Original source
Google outlined how its AI research is being applied to understanding and treating disease, saying its teams have published 90 peer-reviewed papers on the subject over five years. Examples include a Barcelona institute using AlphaFold to screen 30,000 compounds for Chagas disease, genetic diagnoses in under eight hours with Stanford, and a mammography system that caught 25% of historically missed interval cancers across 125,000 women. Google.org has provided over $60 million in 2026 to groups applying such models.
NVIDIA · Original source
NVIDIA and Cisco announced plans for Ultra-Dense Small Form Factor Pluggable (USFP), an optical module designed to support 400Gb/s-per-lane signals and up to 12.8Tb/s of capacity for AI factory networks. The companies say USFP delivers four times the bandwidth density and about 1.3 times the power efficiency of 200Gb/s-per-lane optics. TE Connectivity and Amphenol are working with them on cages, connectors and modules, and an open industry specification is planned through a multi-source project.
NVIDIA · Original source
At the Open Compute Project Global Summit in San Jose, NVIDIA contributed Vera Rubin rack and compute tray designs to OCP through its open MGX architecture, including an MGX 1.2 specification supporting rack deployments of up to 250kW. NVIDIA also released open source DSX OS components for running AI factories and joined AMD, Google and Meta in publishing the Open Data Center Specification v0.7. Partner news included a USFP optics standard with Cisco and Falcon Transport support with Google.
Character.AI · Original source
Character.ai has started a voluntary Chapter 11 bankruptcy process to facilitate a court-supervised sale of the company, citing significant legal, regulatory and operational headwinds over the past year. The company said the platform remains up and running and paid subscriptions stay active, but it will be wound down if no actionable bid emerges. It expects the sale process to be complete by the end of November or early December 2026.
Meta · Original source
Meta has introduced Muse Wonder and Muse Wonder Full, which it describes as its most capable AI models to date, built on its earlier Muse Spark work and aimed at personal and professional agentic tasks. The models target expert domains including law, finance and health as well as coding, and add Active Memory and Active Focus capabilities. Muse Wonder is available now through the Meta Model API and Muse Code, with a fuller safety report due soon.
xAI · Original source
xAI has launched The X Subscription, a single plan that combines Grok, Grok Bot, Cursor, and X Premium. New individual tiers run from Free to Ultra at up to $200 a month, with Premium at $30 pitched as replacing about $90 of prior separate plans. Paid tiers share a Grok model usage pool across Grok and Cursor, with at least twice the coding usage of comparable Cursor plans. Existing SuperGrok, Cursor, and X Premium customers can keep current plans or switch with prorated refunds when the option appears in billing settings.
Google · Original source
Google unveiled Fitbit Edge, a slim fitness tracker that brings Google Health Coach and AI-assisted automatic workout recognition to an ultra-thin glanceable device between Fitbit Air and Pixel Watch. The tracker offers more than 50 exercise modes, precision GPS, Health Guardian trend features, up to seven days of battery life and a repairable modular design, with pre-orders at US$179.99 and shelf availability from 27 October. A Stephen Curry Special Edition is priced at US$219.99.
Salesforce · Original source
Salesforce signed a definitive agreement to acquire Loops, an AI-powered analytics platform that monitors business performance, explains why metrics change and surfaces improvement opportunities using machine learning. Salesforce said Loops will advance Tableau’s agentic analytics strategy, with headless capabilities that can be triggered from Slack and email to watch KPIs and flag drivers of change. The deal remains subject to customary closing conditions.
Sakana AI · Original source
Sakana AI introduced Multi-Agent Self-Supervision (MASS), a recursive self-improvement method in which one shared model acts as task solver, workflow optimizer and evaluator for teams of virtual subagents. After two cycles with a 27B open-weights model on synthetic research tasks, score per output token reached 1.2–1.6x the base model across four research benchmarks. The company released a paper, detailed blog and open-source code with UC Berkeley collaborators, and flagged homogeneous RSI safety risks when a model is also its own evaluator.
Microsoft · Original source
Microsoft published architecture details for Azure Maia 300, its third-generation LLM inference system co-designed across silicon, racks, networking and datacentres. Each accelerator uses a multi-chiplet design on TSMC N2P with eight HBM4 stacks providing 288 GB and 18 TB/s of memory bandwidth, while racks pack 96 accelerators within a 167 kW envelope and scale into a 12,288-accelerator superpod. Microsoft said Maia 300 delivers up to 7x rack-scale token throughput and more than 2.5x performance per watt versus Maia 200.
Microsoft · Original source
Microsoft said Azure Maia 200 is now in production for Copilot and other Microsoft AI workloads, and that Maia 300 will begin production deployments in 2027 as the company scales Maia to multi-gigawatt Azure capacity. Maia 300 is an inference-first custom chip system aimed at long-context and agentic workloads, with Microsoft projecting about 40% more tokens per megawatt versus leading GPU systems and up to 7x Maia 200 rack-scale token throughput in a 96-accelerator rack.
