// NODE-02 · NEURAL FEED · DAILY TRANSMISSION //
May 9, 2026 · WINDYVIEWS.COM
Anthropic has reportedly hit a $30 billion annualized revenue run rate, overtaking OpenAI's $24 billion, marking the first time the Claude maker leads its rival in ARR. Enterprise demand for Claude — especially for coding and long-context workloads — is credited as the main driver. The shift comes alongside Anthropic's growing involvement in U.S. AI policy via the new CAISI safety institute.
Meta, Amazon, Microsoft and Alphabet have collectively signaled roughly $725 billion in capital expenditures for 2026, almost entirely earmarked for data centers, custom chips, GPUs and frontier models. At the same time, Meta is cutting 8,000 roles in May, Amazon has trimmed ~30,000 in recent months, and Microsoft has offered voluntary buyouts to about 125,000 staff. The bifurcation underscores a violent reallocation of headcount budgets toward GPUs and inference capacity.
Fortune reports the White House has begun adopting AI oversight ideas it previously rejected, working with Anthropic, the Mythos Project and the newly rebranded CAISI (Center for AI Standards & Innovation, formerly the AI Safety Institute) on safety evaluations and risk thresholds. The pivot follows the National AI Policy Framework released March 20 and a fresh $10M boost to CAISI on top of $55M for NIST AI work.
Danish pharma giant Novo Nordisk announced a strategic partnership with OpenAI to embed AI across drug discovery, clinical trials, manufacturing, supply chain and commercial operations, with full deployment targeted by year-end 2026. The deal is one of the largest single-company AI integrations in the pharmaceutical sector and signals that domain-specific deployment — not horizontal chat — is where the enterprise dollars are flowing.
JPMorgan Chase has formally reclassified its AI investments from experimental R&D to core infrastructure inside a $19.8 billion 2026 technology budget, with 2,000 staff dedicated to AI. The bank projects $2.5 billion in annual value from AI agents focused on internal productivity, cybersecurity hardening and personalized retail banking — a template several Tier-1 banks are now expected to copy.
Stanford HAI's 2026 AI Index reports performance on SWE-bench Verified — a real-world coding benchmark — climbed from 60% to near-100% in a single year, with global git pushes up 78% YoY. The U.S.–China model gap has narrowed to roughly 2.7% at the top, and global AI usage among working-age adults hit 17.8% in Q1 2026, up 1.5 points in a single quarter.
Microsoft has unveiled Majorana 1, the first quantum chip built on topological qubits and the only architecture engineered to natively detect and correct errors. The chip is positioned as a long-horizon bet to eventually accelerate AI workloads via quantum-classical hybrid pipelines. Independent verification of the underlying physics is still in progress.
Boston Dynamics released new footage of its Atlas humanoid demonstrating a step-change in balance and dynamic mobility. In parallel, Amazon launched commercial drone delivery in the UK with its MK30, carrying 5-pound packages within a 12-mile radius using onboard lidar, cameras and AI navigation. Embodied AI is moving from demo videos to deployed services within weeks of each other.
Domain-specific models continued to dominate this week's research feed. MatterChat, a multimodal LLM published in Nature Machine Intelligence, fuses crystal-structure data with language to deliver interpretable property predictions. Stripe disclosed Minions, autonomous coding agents shipping over 1,300 PRs per week internally, and AWS open-sourced Strands Labs as a sandbox for agent-based AI tooling.
Two themes dominate the week: capital and consolidation. Hyperscalers are pouring three-quarters of a trillion dollars into AI infrastructure while shedding tens of thousands of jobs, and Anthropic just lapped OpenAI on revenue — proving the enterprise buyer has chosen sides. On the policy front, Washington is quietly building safety guardrails (CAISI, NIST funding, the National AI Framework) even as it preempts state law, while deployment leaves the lab: humanoids walking, drones delivering, vertical LLMs writing code, drug pipelines and bank risk models. Generic chatbots are no longer the story — domain integration is.