// NODE-02 · NEURAL FEED · DAILY TRANSMISSION //
June 02, 2026 · WINDYVIEWS.COM
Anthropic raised $65 billion on May 29 at a $965 billion post-money valuation, overtaking OpenAI's $852 billion to become the world's most valuable private AI company. The round was led by Sequoia, Dragoneer, Altimeter and Greenoaks, with participation from Microsoft and NVIDIA, and is earmarked for compute infrastructure and safety research.
Anthropic released Claude Opus 4.8 on May 28, its newest flagship model. As of March 2026 the top Anthropic model led the frontier by only about 2.7% over rivals — a thin margin that underscores how tightly bunched the leading labs have become.
Google made Gemini 3.5 Flash generally available, pitching frontier-level intelligence at roughly 4x the speed of comparable models with a 1M-token context window. Pricing lands near $1.50 / $9 per million input/output tokens, aimed squarely at high-throughput, cost-sensitive workloads.
On June 1 GitHub Copilot moved from flat-rate subscriptions to token-based billing across all plans, and the developer backlash was immediate. Some users projected monthly bills jumping from $29 to as much as ~$750; Microsoft's competing in-house MAI coding model, expected at Build 2026, is positioned to blunt the cost concerns.
SoftBank announced a €75 billion (~$87.5B) investment on May 30 to build 5 gigawatts of AI data-center capacity in France, with a first phase delivering 3.1 GW by 2031 across three Hauts-de-France sites. It is Europe's largest single AI infrastructure commitment, leaning heavily on France's nuclear-powered grid.
Cognition raised $1 billion on May 28 at a $26 billion post-money valuation, led by Lux Capital and General Catalyst. The company says Devin's annualized revenue grew 1,230% in a year — from $37M to $492M — with enterprise customers including Goldman Sachs and NASA.
Security firm Sysdig reported the first confirmed in-the-wild attack driven autonomously by an LLM agent on June 1. The agent exploited a Starlette flaw (CVE-2026-48710, dubbed "BadHost") to exfiltrate an AWS database in under an hour, affecting FastAPI, vLLM and LiteLLM deployments.
OpenAI unveiled the Rosalind Biodefense Program on May 29, giving the US government and Five Eyes allies access to a GPT-Rosalind model for pandemic modeling, variant identification and vaccine prioritization. The program is named after chemist Rosalind Franklin.
Google DeepMind CEO Demis Hassabis publicly placed AGI — broadly human-level performance across cognitive tasks — as a real possibility by 2029. The estimate reinforces an industry-wide convergence toward a 2028–2030 arrival window for human-level systems.
Capital and compute defined the week: Anthropic's $965B valuation, SoftBank's €75B France buildout, and a $1B raise for Devin show the frontier race is now as much about data centers and balance sheets as about models. Fresh releases keep pace — Claude Opus 4.8 and a generally-available Gemini 3.5 Flash — even as the gap between top labs narrows to a few percent. And the edge is getting sharper: agentic coding tools are scaling explosively, the first autonomous LLM-agent cyberattack hit the wild, and AGI timelines are tightening toward the end of the decade.