Top Stories
Anthropic Overtakes OpenAI in Business Adoption — First Time Ever. Ramp's May AI Index shows Anthropic at 34.4% business adoption vs OpenAI's 32.3% — a historic crossover. Anthropic quadrupled enterprise adoption in 12 months while OpenAI grew 0.3%. The engine: Claude Code, now authoring 4% of all public GitHub commits (doubled in one month). Separately, Anthropic is in talks to raise $30-50B at up to $950B valuation, which would surpass OpenAI's $852B. Revenue run-rate now exceeds $30B, up from ~$9B at end of 2025. Three threats remain: model commoditization, token-pricing incentive misalignment, and Google/Amazon compute dependency. Developing — builds on Anthropic's Stainless acquisition talks from May 13. (TechCrunch — "Anthropic now has more business customers than OpenAI"; Ramp AI Index May 2026; VentureBeat — "3 big threats could erase its lead")
Google I/O Monday: Gemini 4, Gemini Spark Agent, and Android XR Glasses. The May 19 keynote (10am PT) is shaping up as Google's biggest AI moment of 2026. Expected headlines: a new Gemini model (leaked ARC-AGI-2 scores: 84.6% vs Opus 4.6's 68.8%), plus "Gemini Spark" — a leaked 24/7 autonomous agent that manages email, creates reusable "skills," pulls context from linked apps, and runs in the background. Google warns it's experimental and may "make purchases without asking." Also confirmed: Android XR glasses preview and Android 17 merging Android, Chrome OS, and XR. Developing — extends pre-I/O leaks from May 13. Monday is confirmation day. (9to5Google — "Gemini Spark is Google's upcoming AI agent"; Android Authority — "What to Expect from Google I/O 2026")
Musk v. Altman: Jury Deliberations Open Monday in Oakland. The advisory jury of nine begins deliberating May 18 on breach of charitable trust and unjust enrichment. Musk seeks $134B in remedies, Altman/Brockman removal, and unwinding of OpenAI's for-profit conversion. Brockman's 2017 journal entry — "Cannot say we are committed to the non-profit... if three months later we're doing b-corp then it was a lie" — emerged as the trial's most damaging evidence. Judge Gonzalez Rogers delivers the binding ruling, likely by end of May. Stakes remain existential for OpenAI's planned IPO. Developing — Week 3 complete, courtroom phase over. (CNBC — "Closing arguments conclude"; TechTimes — "OpenAI Jury Weighs Removal of Altman"; MIT Tech Review — "Musk v. Altman week 3")
Cerebras Day 3: CBRS Slides to ~$280, Down 25% From Peak. After debuting at $350, surging to $380+, and closing Day 1 at $311, Cerebras fell to ~$280 by Day 3 — a 25% decline from intraday highs. The $185 IPO price still gives early investors a 50%+ gain, but the post-IPO volatility pattern mirrors classic tech IPO pullbacks. The bull case (OpenAI compute deal, $510M revenue, 47% margins) hasn't changed; neither has the bear case (customer concentration, UAE scrutiny). Developing — watch whether $275 holds as support. (Yahoo Finance; TECHi — "CBRS now has to prove the $100B pop")
Meta's 8,000 Layoffs Begin May 20. Cuts are four days away. Teams affected: Reality Labs, Facebook app, recruiting, sales. The $115-135B AI infrastructure pivot continues — Zuckerberg has now cut ~25,000 since 2022. Broader context: 95,000+ tech layoffs in 2026 across 247 events. Oracle cut 30,000, Amazon 16,000, Microsoft offering voluntary buyouts to 8,750. The industry has shed nearly 900,000 since 2020. Developing — extends May 14 coverage. Execution begins next week. (TheNextWeb; CNN — "Meta to cut 10% of staff")
Model Updates
- Microsoft DELEGATE-52 benchmark exposes agent fragility. Frontier models (Gemini 3.1 Pro, Claude 4.6 Opus, GPT-5.4) lose 25% of document content over 20 delegated interactions. Of 52 professional domains tested, only Python programming cleared the 98% "ready" bar. Implication: agentic workflows still need close human supervision. (The Register — "Microsoft researchers find AI agents can't handle long-running tasks")
- SubQ: First commercial non-transformer LLM. Subquadratic launched May 5 with $29M seed, shipping a 12M-token native context window at 1/5th frontier cost and 52x faster attention. Architecture is not a transformer — the first serious challenger to the dominant paradigm. (WhatLLM — "New AI Models May 2026")
- No new frontier model drops this week. The April ceiling (GPT-5.5 xhigh at 60.24 on benchmarks) has held. Google I/O Monday is the next likely frontier event.
Open Source
- ZAYA1-8B from Zyphra: Apache 2.0 MoE reasoning model, 8B total params, ~760M active per token. Trained end-to-end on AMD Instinct hardware. Optimized for intelligence density per active parameter. (WhatLLM — "New AI Models May 2026")
Industry Moves
- OpenAI launches $4B Deployment Company, acquires consulting firm Tomoro (~150 engineers). Backed by TPG, Advent, Bain Capital, Brookfield. OpenAI's answer to Anthropic and Accenture's enterprise consulting push. (Bloomberg; OpenAI blog)
- Claude Code reaches 4% of all GitHub public commits, doubled in one month. SemiAnalysis projects 20%+ by year-end. Average Claude Code developer spends 20 hrs/week with the tool. (SemiAnalysis; Ramp AI Index)
- Trump-Xi summit aftermath: H200 deliveries remain stalled. Despite US Commerce Dept approval for ~10 Chinese firms at 75K chips each with 25% revenue-sharing, Beijing is discouraging purchases to push domestic self-sufficiency. No bilateral AI governance agreement was signed. (Time — "AI Was the Elephant in the Room"; CNBC — "Trump-Xi talks left chip export issue unresolved")
Safety & Policy
- EU AI Omnibus deal reached May 7 — simplifies high-risk compliance, extends deadlines to Dec 2027/Aug 2028, and bans nudification apps and AI-generated CSAM (compliance by Dec 2026). Formal adoption expected before Aug 2, 2026. (European Parliament; Euronews — "EU reaches tentative deal to simplify AI rules")
- Microsoft warns on agentic security: Separate from the DELEGATE-52 paper, Microsoft Security Blog flagged RCE vulnerabilities in AI agent frameworks where "prompts become shells." (Microsoft Security Blog, May 7)
Worth Watching
- "Elon Musk Vs. OpenAI (ft. Zac)" — Deep dive on the trial's closing arguments and what the jury faces Monday.
- "AI NEWS | OpenAI Lawsuit, Google Hacks, Grok Build Beta" — Roundup covering the trial, pre-I/O leaks, and Grok's new coding mode.
- "Exciting AI Updates Weekly - May 15, 2026" — Comprehensive weekly digest covering all major stories.