AI Daily Insights No. 100 · November 12, 2026

AI Tracker

Follow AI products, research, funding, acquisitions, earnings and major executive changes through sourced briefs and linked editorial coverage.

Source publication time · Beijing time

Checked every 6 hours · Last checked (Beijing) · 10 new in this 6-hour window

All updates

180 updates in the archive

According to Xinhua, China Mobile validates 1.6 Tb/s single-wavelength transmission over 960 km

According to Xinhua, China Mobile and partners completed real-time single-wavelength 1.6 Tb/s transmission over 960 km on standard G.652.D backbone fiber using 230GBaud key technology. The company said the test set a new record for its longest single-wavelength 1.6T transmission and marked the industry’s first long-distance 1.6T validation based on 230GBaud. China Mobile also said the result will provide measured evidence for standard work on 1.6T backbone and DCI scenarios.

AI-generated Reported IT之家 ↗ #中国移动

AWS Machine Learning details enterprise rollout paths for Codex on Amazon Bedrock

AWS Machine Learning published guidance on using Codex with Amazon Bedrock to access supported OpenAI models while keeping AWS identity, IAM, regional, and observability controls in place. It recommends starting with a small least-privilege pilot, then choosing Native AWS Access, AgentCore Gateway, or an LLM Gateway depending on policy needs. The examples come from the open source Guidance for Codex on AWS repository and include quick starts for Native AWS Access, AgentCore Gateway, LiteLLM on ECS, Portkey, ADOT, and the CloudWatch setup used later in the post.

AI-generated Technical overview AWS Machine Learning ↗ #OpenAI#Codex#Amazon Bedrock

AWS Machine Learning describes HyperPod checks for silent data corruption in distributed training

AWS Machine Learning describes a set of checks for detecting silent data corruption in PyTorch DistributedDataParallel training on Amazon SageMaker HyperPod. The approach adds three guards: non-finite gradient checks, ring replay audits, and weight fingerprint comparisons. When a check fires, the system maps the suspect rank to a node and lets HyperPod replace the bad hardware. The post says the method was tested with software-injected faults on a 4-instance ml.p5.48xlarge cluster with 32 NVIDIA H100 GPUs.

AI-generated Technical overview AWS Machine Learning ↗ #Amazon SageMaker HyperPod#Amazon EKS#AWS

Wccftech reports AMD’s MEXT memory system for EPYC servers

According to Wccftech, AMD has introduced MEXT for its EPYC server processors, a software-managed memory system that combines DRAM and NVMe SSDs to cut memory costs while aiming to keep performance close to an all-DRAM setup. The stack includes a Linux driver, an AI engine and the RAM View monitoring tool, and the AI engine reportedly runs on one EPYC core without a GPU. AMD’s tests showed about 99% of baseline performance in Azure Spark TPC-DS and about 95% in Synopsys PrimeTime.

AI-generated Reported IT之家 ↗ #AMD#MEXT#EPYC

AWS details event-driven sync for MCP server updates in Agent Registry

AWS Machine Learning described an event-driven architecture for Amazon Bedrock AgentCore’s AWS Agent Registry. The pattern uses AWS CloudTrail, Amazon EventBridge, and AWS Lambda to detect UpdateAgentRuntime changes and automatically refresh the matching MCP server record. It is designed for both single-account and cross-account setups, helping keep registry metadata current within an existing governance workflow. The post also notes that updated records revert to DRAFT and must re-enter approval before discovery.

AI-generated Technical overview AWS Machine Learning ↗ #AWS#Amazon Bedrock AgentCore#AWS Agent Registry

AWS Machine Learning says DXC Technology built DXC OASIS on Amazon Bedrock with 95% faster resolution times

AWS Machine Learning says DXC Technology built DXC OASIS on Amazon Bedrock to support mission-critical IT operations for regulated customers. The system uses one managed interface to access multiple foundation models, while Bedrock Guardrails, AWS PrivateLink, and serverless scaling provide policy controls and private inference traffic. According to the post, DXC has seen average resolution times fall by 95% across 56 measured use cases, and it is evaluating Amazon Bedrock AgentCore for future expansion.

AI-generated In use AWS Machine Learning ↗ #DXC Technology#Amazon Bedrock#DXC OASIS

According to TechCrunch, Meta’s longtime infrastructure chief Santosh Janardhan is stepping down

Meta said longtime infrastructure chief and co-head of engineering Santosh Janardhan is stepping down after joining the company in 2009. He will stay through the transition, move into an advisory role on Meta Compute, and is expected to leave next year. Meta will replace him with Dave Brown, who joined earlier this year as vice president of infrastructure. Brown previously spent 19 years at Amazon working on compute, cloud, and AI, according to his LinkedIn profile.

AI-generated Reported TechCrunch ↗ #Meta

TechCrunch reports Fractile raises $720 million Series C for AI chips

According to TechCrunch, British AI chip startup Fractile has raised a $720 million Series C, with backing from Britain’s Sovereign AI Fund. The round was led by Thrive Capital, Founders Fund, Lightspeed Venture Partners and Redpoint, and the UK government said the deal is meant to give Britain a bigger stake in hardware for its future economy. The company says it is building a full-stack system that pairs chips with a machine learning compiler, and CEO Walter Goodwin said the funding should help Fractile ship its first chips and systems next year.

AI-generated Reported TechCrunch ↗ #Fractile

Reuters via Ars Technica AI: TSMC plans $1 billion stakes in Anthropic and OpenAI

According to Reuters, as reported by Ars Technica AI, TSMC’s board approved plans to invest up to $500 million in OpenAI preferred stock and another $500 million to buy Anthropic common shares at its expected IPO price. The move would give the Taiwanese chipmaker direct stakes in two leading AI developers. The Wall Street Journal said such circular financing may help bolster confidence, but it also makes it harder to gauge real demand for AI hardware and services.

AI-generated Announced Ars Technica AI ↗ #OpenAI#Anthropic#TSMC

Google launches Gemini Decisions API with Gemini Decision 1 for fast multimodal scoring

Google has released the Gemini Decisions API, powered by its new Gemini Decision 1 model and available through the Gemini API in Google AI Studio. The endpoint can evaluate up to 1,000 yes/no, choice, and rubric-scoring questions in one pass across text, images, audio, video, and PDFs. Google says it returns probabilities and confidence scores with 150–250 ms end-to-end latency and costs $0.10 per 1M input tokens, with no output-token charges.

AI-generated Released Google ↗Techmeme ↗ #Google#Gemini Decisions API#Gemini Decision 1