AI News August 26, 2026: Anthropic Hits $65B Revenue, Cursor Challenges GitHub, Groq Pivots to Neocloud

🗞️ Top 5 AI Stories — August 26, 2026

The revenue race flipped, a coding-editor giant picked a fight with GitHub, and agents got their own always-on computers. Here are the five stories that matter this Wednesday.

1. 💰 Anthropic Hits $65B Annualized Revenue — $25B Ahead of OpenAI

Anthropic told investors that its annualized revenue run rate reached $65 billion at the end of July, CNBC and Bloomberg confirmed — up from $47 billion in May, a sevenfold jump from a year ago, and roughly $25 billion ahead of OpenAI’s $40 billion run rate. Preliminary Q2 revenue came in at $11.5 billion, a 14-fold increase year-over-year.

The engine is enterprise: API pay-per-token traffic and tools like Claude Code drive most of the growth, with 300,000+ business customers and over 100,000 running Claude on Amazon Bedrock. Investors reportedly expect Anthropic to finish 2026 between $100 billion and $120 billion in revenue.

The numbers land as the company gears up for what could be the biggest tech IPO in years — it confidentially filed its prospectus in June at a reported $965 billion valuation, with some investors floating as much as $2 trillion. (Sources: TechCrunch, CNBC, Sacra)

2. 🔄 Cursor Launches “Origin” — A GitHub Rival Built Into the Editor

Cursor, now officially part of SpaceX after its $60 billion sale in June, launched Origin this week: a Git-based code-hosting platform designed to do everything developers use GitHub for — repositories, pull requests, code browsing, and storage — without leaving the editor where their AI agents already work.

The beta, rolling out to paid plans since August 17, ships four core pieces: repositories, pull requests, code browsing, and two-way GitHub sync that pulls upstream changes in and pushes local work back out. Enterprise admins can opt their organizations out entirely.

The timing is pointed. Origin launches into a wave of developer frustration with GitHub, and Cursor is explicitly positioning it as a low-risk layer on top of GitHub today — and a replacement tomorrow, with agent-native features still on the roadmap. GitHub now has its first serious challenger from inside the AI coding wave. (Sources: TechCrunch, SiliconANGLE, DigitalOcean)

3. 🏗️ Groq Raises $350M to Complete Its Pivot From Chips to Neocloud

Groq raised $350 million at a $3.5 billion valuation to fuel its transformation from an AI chipmaker into a neocloud — an inference-focused cloud provider now expanding an Nvidia-powered data center footprint, TechCrunch reported.

The round puts Groq directly inside Nvidia’s AI infrastructure ecosystem, joining CoreWeave, Lambda, and Nebius — clouds Nvidia supplies and, in several cases, invests billions in as capacity races demand. A company spokesperson framed the valuation as “the post-Nvidia-licensing-deal version of Groq” rather than a down round.

For enterprise operators, the signal matters more than the headline: inference capacity is consolidating into a handful of vendor-run neoclouds that own scheduling, capacity planning, and reliability decisions that used to live inside enterprise platform teams. (Source: TechCrunch)

4. 🤖 Grok 4.6 Ships With Always-On Agents — and Lands on Bedrock

SpaceXAI (formerly xAI) released Grok 4.6 on August 12, building on Grok 4.5 with a sharp focus on long-running agents and “more ambitious interactive and visual work” — sticking with complex tasks across many steps, from research and codebase analysis to turning ideas into polished applications. It keeps the 500K-token context window and the $2/$6 per-million-token pricing of its predecessor.

The bigger story this week is distribution. Grok Bot — teams of always-on agents that have their own computer, work inside tools and apps, and keep running 24/7 — expanded to SuperGrok Plus, Cursor Pro+, and all Cursor Teams plans. Grok 4.6 also landed on Amazon Bedrock and, in a striking crossover, on Gemini Enterprise Agent Platform — an OpenAI-rival model running on Google’s enterprise stack. Build Mode, the drag-and-drop app builder previously limited to SuperGrok Heavy, opened to all users. (Sources: SpaceXAI, orcarouter)

5. 🧠 Zhipu’s GLM-5.3 Claims Open-Source Coding Crown — Weights Land This Week

Zhipu AI (Z.ai) released GLM-5.3 on August 14, claiming the strongest open-weights coding system available. The remarkable part: it’s the same base model as GLM-5.2 (~743B-parameter MoE, 1M-token context) — every gain came from roughly one extra month of reinforcement learning on more executable environments, longer tasks, and stronger verifiers.

The results: a 50% coding improvement on Z.ai’s in-house Code Bench, Terminal-Bench 3.0 jumping from 4.6 to 28.3 (a 6.2x gain, first among open models), first place among open models on Agents’ Last Exam, and 84.5% on CyberGym — slightly above Claude Mythos 5 and GPT-5.6 Sol. The API went live August 18 through the GLM Coding Plan, and after what Zhipu calls its most extensive risk review yet, open weights hit Hugging Face around August 28. (Sources: kie.ai, mlq.ai, apidog)

🧭 The Thread Connecting Today’s Stories

Three undercurrents run through today’s news:

1. The revenue race has a new leader. Anthropic’s enterprise-first, API-driven strategy just overtook OpenAI on run rate — and the IPO window is open.
2. Agents are getting infrastructure. Always-on Grok Bots with their own computers, Cursor’s editor-native hosting, and inference consolidating into neoclouds all point the same direction: the agent economy is building its own rails.
3. Post-training is the new scaling. GLM-5.3’s 6.2x benchmark jump without a new pre-training run is the clearest evidence yet that RL on executable environments — not raw compute — is where frontier gains live now.

Cover image generated with Cloudflare Workers AI (SDXL). Tips or corrections? Reach out via the about page.