TLDR IT 2026-08-25

Nvidia's Groq 3 LPX accelerator is entering production as part of its Vera Rubin platform, targeting ultra-fast token generation for agentic AI. Nebius is the first cloud provider committed to deploying it.

IBM is developing a dual-architecture processor that can natively execute both Arm and IBM Z instructions. The design could let enterprises run modern AI and cloud-native software directly alongside critical mainframe workloads.

OpenAI appears to be regaining ground against Anthropic among US corporate customers, according to spending data from Ramp covering more than 70,000 organisations. Anthropic led in May and July, but third-quarter trends suggest faster growth for OpenAI, while increasing overall spending on paid AI services indicates that both providers may expand even as corporate loyalty remains fluid.

Stream processing is valuable for continuous event-driven computation but is often the wrong architecture for request-response interactions, direct state queries and corrections, or work requiring locks. As evidence, Kestra rebuilt its messaging layer around plain queueing, and Confluent moved Control Center's metrics pipeline from Kafka Streams to Prometheus.

AI cybersecurity capabilities are inherently dual-use: the same ability to discover and exploit vulnerabilities is needed to defend infrastructure, with intent shaped by incentives and whoever controls the system. Defenders must automate the full loop from vulnerability detection through patching, deployment, and rollback.

Zoom Phone lives inside the Zoom app your team already uses, so replacing a legacy PBX or another provider doesn't mean a new tool, new logins, or retraining anyone. Enterprise reliability and built-in security features, plus AI that summarizes calls and transcribes voicemail. From $10.50/user/month.

See Zoom Phone plans →

Atlassian Code Context is a Teamwork Graph capability that lets developers and AI coding agents search connected GitHub and Bitbucket repositories alongside Jira, Confluence, Loom, and other organizational data. Atlassian says internally benchmarked agents using this context produced 44% more accurate results with 48% fewer tokens, while access remains controlled by existing repository permissions.

Claude Tag can now follow entire Slack conversations and proactively join discussions instead of waiting to be explicitly invoked. Anthropic describes the approach as “multiplayer AI,” where agents participate directly in team workflows.

Thomson is Thomson Reuters' first proprietary large language model. It is designed around professional legal workflows and the company's specialized data.

Google is adding Antigravity to Gemini Enterprise with centralized budgets, quotas, and overage controls. The move gives IT teams more visibility over rapidly growing AI usage and spend.

AI-focused clouds such as CoreWeave, Lambda, and Nebius offer cheaper, highly optimized GPU infrastructure, but hyperscalers still hold advantages in security, managed services, and enterprise integration. Gartner expects neoclouds could capture 20% of the AI cloud market by 2030.

ClickHouse describes how its LogHouse observability platform evolved to handle OpenTelemetry ingestion at 50 million events per second without losing telemetry during ClickHouse backpressure. After finding in-memory queues and local collector write-ahead logs inadequate, the company designed a priority-based failover pipeline that keeps ClickHouse in the hot path and uses blob storage as a durable overflow buffer.

Cursor uses S3 as its source of truth and local NVMe repositories for latency-sensitive operations to overcome Git scaling limitations.

A design for building a source-neutral, continuously maintained MTTR timeline across PagerDuty, Jira, Slack, observability systems, and other incident tools.