Top Stories
Cerebras Debuts at ~$50B — 2026's Biggest IPO Opens Hot. CBRS priced last night and opened trading today on Nasdaq. After being 20x oversubscribed, shares opened around $350-370 (above the raised $150-160 guidance range), giving Cerebras a ~$49B fully diluted valuation. CNBC's Dan Primack called it a significant moment — $9B in IPO value this week alone, compared to $18B total before it. The 750MW OpenAI compute deal and $510M revenue anchor the bull case, but risks remain: UAE investor scrutiny delayed this IPO over a year, and Chinese portfolio divestiture concerns linger. Altman disclosed during trial testimony he's personally invested in Cerebras. Developing — priced yesterday, now trading. Watch first-week price action. (CNBC — "Cerebras IPO is a huge success"; Meet Kevin — "WARNING The Cerebras IPO")
Musk v. Altman Enters Final Day — Closing Arguments Begin, Jury Deliberations Monday. The last full day of testimony wrapped today. Key recap: Altman testified Musk wanted 90% of OpenAI and repeatedly proposed merging it into Tesla. The "pass it to my children" quote became the trial's defining moment. Greg Brockman's personal journal was read aloud in court. Helen Toner's deposition accused Mira Murati of being "remarkably passive" during the board crisis. Siobhan Zillis (mother of four Musk children) was revealed as his inside liaison on OpenAI's board. Musk seeks $130B+ in remedies and Altman's removal. Advisory jury verdict expected week of May 18; Judge Gonzalez Rogers' ruling is final. Stakes are existential for OpenAI's upcoming IPO. Developing — Week 3 complete. Closing arguments today, deliberations Monday. (CNBC — "Final full day of testimony"; Hard Fork — "Top 5 Moments From the Musk v. OpenAI Trial")
Trump-Xi Summit Day 2: State Banquet, Boeing Deal, AI Export Controls in Play. Trump and Xi held a state banquet at Beijing's Golden Hall following Day 1 arrivals with Musk, Jensen Huang, and Tim Cook. Boeing's CEO is part of the delegation hoping to close a 200+ jet deal. The real negotiation: China wants loosened semiconductor export controls; the US wants IP theft protections. Taiwan ($11B weapons package just approved), Iran (Strait of Hormuz), and trade truce terms round out the agenda. Multiple channels describe this as "transactional diplomacy" — the CEOs are there because the deals are about compute access, not soybeans. Developing — from Day 1 arrivals yesterday, now in substantive talks. (The Hill — "Trump's visit to China"; NBC News — "Trump's Transactional Diplomacy"; Bloomberg — "Nvidia's CEO Joins Trump in China")
AI Layoffs at Record Earnings: Cloudflare Cuts 1,100 After Best Quarter, Cisco Cuts 4,000. Mo Bitar's viral analysis (185K views) crystallized the pattern: Cloudflare CEO Matthew Prince announced his company's best quarter ever ($639M revenue) and 1,100 layoffs in the same breath. His justification — employees are "100x more productive" and "support roles won't drive companies forward" — drew intense backlash. Cisco simultaneously cut 4,000 as AI orders surge. Bitar's thesis: companies are eliminating "bench players" because AI serves as institutional knowledge backup, but Jevons' paradox means demand will eventually rebound. The gap between AI-driven productivity claims and actual ROI (per last week's Gartner data: no correlation) remains the unresolved tension. Developing — extends the Gartner "AI layoffs don't deliver ROI" story from May 11. (Mo Bitar — "The AI layoffs end in 12 months"; Ox Talks — "Cisco Cuts 4000 Jobs")
Model Updates
- GPT-5.6 in active development. Internal checkpoints "Ember Alpha" and "Beacon Alpha" reportedly in testing. Mid-June release window looking likely. OpenAI's red-teaming process described as "far more dialed in" than 2025. (WorldofAI — "GPT-5.6, Gemini 3.2, Robots Running A Full 8-hr Shift")
- Gemini 3.2 Pro/Flash variants leaking ahead of I/O (May 19). Multiple checkpoints surfacing — "Sprite" and "Cola" variants tested. Early results: Flash variant impressive for its tier, but Pro variant "doesn't feel like a major leap." Gemini Omni (video generation) also leaked with decent editing consistency. Community consensus: Google may be saving the strong variant for I/O launch. (WorldofAI; Universe of AI — "Gemini 3.2 PRO? Is Google Back?")
- Claude Opus 4.7 fast mode in research preview — 2.5x faster responses, same model quality, higher token cost. (WorldofAI)
- Anthropic rate limit controversy. 50% weekly limit increase announced, but third-party agents/SDKs pushed to separate paid API credits — effectively a 10-40x usage cut for heavy agent workflows. Developer backlash intense, tied to ongoing compute shortage. (WorldofAI — "Anthropic's Downfall?")
- OpenAI Codex now in preview on the ChatGPT mobile app. (OpenAI official channel)
- Mythos V2 reportedly dropped (Claude security model). Governments described as "panicking." EU still denied access. (Universe of AI; Your AI Guy — "Claude Mythos Did It Again")
Open Source
- Qwen 3.6 Plus now free through Hermes Agent's news portal for a limited time. Hermes Agent v2.0 shipped with WebUI, computer use, and expanded provider integrations. (WorldofAI — "Hermes Agent v2.0")
- Natural language autoencoders — Anthropic research turning Claude's internal representations into readable text for interpretability. Described as extending sparse autoencoder work (Golden Gate neuron era) into natural language. Significant for alignment. (Dylan Curious — "Experts Just Broke the AI Black Box")
- Open Code gaining traction as open-source alternative to Claude Code/Codex, with multiple videos discussing the shift. (Various channels)
Industry Moves
- Anthropic-Stainless acquisition ($300M+) still in advanced talks. Would give Anthropic control of SDK generation for OpenAI, Google, Cloudflare, and Cerebras. 2x last valuation. (VARINDIA; AI Reset To Zero — "Anthropic Buys Stainless for $300M")
- Figure AI humanoid robots ran a full autonomous 8-hour warehouse shift using Helix O2 neural network. Fleet coordination, autonomous battery swaps, self-diagnosed hardware issues. Live-streamed. Near human-level speed for packaging/sorting. (WorldofAI; Benzinga — "Humanoid Robots Pulling 24 Hour Shift")
- Apple $250M settlement for over-promising AI capabilities on iPhones. iOS 27 / WWDC (June 8) leaks show major Siri overhaul: dedicated Siri app, dynamic island integration, proper chatbot interface to compete with ChatGPT/Claude/Gemini. (MacRumors — "iOS 27 Siri LEAKED"; MacBreak Weekly)
- NVIDIA hit another all-time high ($232+), market cap above $5.2T. Cerebras IPO and AI infrastructure momentum driving the sector. (TraderTV Live; Fox Business)
- DeepSeek CEO Liang Wenfeng hasn't been seen publicly since February 2025 — over a year. Last seen meeting Xi Jinping after R1 launch. Absent for V4 launch in April. Jack Ma disappearance parallels being drawn. (Daily Mail — "The mystery of the vanishing AI CEO")
Safety & Policy
- Missouri rejected an AI safety bill — state-level AI regulation continues to struggle. (KSDK News)
- "How Money Works" viral video (140K views) argues the "can't let China win the AI arms race" framing is being used to justify deregulation, drawing significant engagement. (How Money Works — "bUt wE cAn'T lEt cHinA WiN")
- AI-generated zero-day exploits continue reverberating from Google's disclosure. Matthew Berman's deep dive (52K views) warned this is just the beginning — malicious actors "don't even have the latest tools yet." IBM Technology covered LLMjacking (stealing API keys). Multiple channels covered the implications. Developing — from Google's disclosure May 12. (Matthew Berman — "Everyone's getting hacked"; IBM Technology — "LLMjacking")
Worth Watching
"The AI layoffs end in 12 months and I know why" — Mo Bitar (185K views, 10 min). Sharp analysis of Cloudflare/Cisco layoff logic, Jevons' paradox applied to software labor, and practical advice for developers. Best single video on the AI employment question right now.
"Everyone's getting hacked" — Matthew Berman (52K views, 42 min). Comprehensive breakdown of the AI cybersecurity landscape: Google's first AI-generated zero-day detection, OpenAI Daybreak, threat actor capabilities. Genuinely concerning analysis.
"The Top 5 Moments From the Musk v. OpenAI Trial" — Hard Fork (2.8K views, 10 min). Kevin Roose and Casey Newton's entertaining recap of the trial's most revealing moments — the "jackass trophy," Brockman's journal, Siobhan Zillis as Musk's mole. Essential context before the verdict.