Musk v. OpenAI: Chaos Under Oath, Anthropic on SpaceX GPUs, China's $20B Challenger | Weekly Digest
PLUS HOT AI Tools & Tutorials
Hey! Welcome to the latest Creators’ AI Edition.
Elon Musk's lawsuit against OpenAI reached its most explosive week yet — former CTO Mira Murati testified under oath that Altman was "creating chaos," and that OpenAI was "at catastrophic risk of falling apart." At the same time, Anthropic announced it's taking all capacity at SpaceX's Colossus 1 data center — 220,000+ Nvidia GPUs — while Musk sues their biggest competitor in federal court. Then China's Moonshot AI closed a $2B round at a $20B+ valuation, and its model is already #2 on OpenRouter. Today we have:
Featured Materials 🎟️
News of the week 🌍
Useful tools ⚒️
Weekly Guides 📕
AI Meme of the Week 🤡
AI Tweet of the Week 🐦
(Bonus) Materials 🎁
Keep your mailbox updated with practical knowledge & key news from the AI industry!
🤖 Sponsored: Luma AI’s UNI-1 is now an API
Luma’s Uni-1 — the model creative teams have been using for video generation — is now available as an API.
What this unlocks: plug it into any product, wire it into a pipeline, or build something new from scratch. Anywhere code runs, Uni-1 can think. It understands intent, not just prompts — give it a brief or a reference board, and it resolves creative direction before generating a single frame.
If you’ve ever wanted to add AI video to your app, automate creative workflows, or build a brand-consistent visual tool, this is the infrastructure that makes it possible.
#LumaPartner @LumaLabsAI
Featured Materials 🎟️
The Musk v. OpenAI Trial: Chaos Under Oath 🔥
The most-watched AI story of the week didn’t involve a model launch. It happened in a federal courtroom in Oakland.
Week two of Elon Musk’s $180 billion lawsuit against OpenAI delivered testimonies that turned the company’s founding myth inside out. Former CTO Mira Murati testified via pre-recorded video on May 6 that CEO Sam Altman was “creating chaos” — saying one thing to one executive and the complete opposite to another. “OpenAI was at catastrophic risk of falling apart,” she said. “I was concerned about the company completely blowing up.”
The full picture of Week 2:
Mira Murati (ex-CTO): Altman “pitted executives against one another” and undermined her role — yet she still wanted him to stay as CEO, because removing him risked the whole company imploding. When asked if Altman was always candid with her: “Not always.”
Shivon Zilis (ex-board member, mother of four of Musk’s children, now at Neuralink): The board “voiced extreme concern” about the launch of ChatGPT — released “without any semblance of board communication.”
Helen Toner (ex-board member): Called Murati “two-faced” about Altman’s 2023 firing — “waiting to see which way the wind would blow.”
Greg Brockman (OpenAI President): Testified Musk wanted personal control — “he said people needed to know he was in charge.”
Musk himself (Week 1): Tried to settle with Brockman days before trial. Said he felt like “a fool” for funding OpenAI.
What this trial is actually deciding:
Musk sued OpenAI in 2024 alleging it illegally converted from nonprofit to for-profit — betraying the mission he funded with $38M. He’s seeking up to $180B in damages plus Altman’s removal from the board. If he wins, it could derail OpenAI’s IPO entirely. The trial runs through May 22. Still to testify: Ilya Sutskever and Microsoft CEO Satya Nadella.
Musk wanted to control the future of AI. Altman built it without him. Now both versions of the story are being told under oath — and neither makes OpenAI look like it was ever just about humanity.
Source: Reuters
Anthropic Signs With SpaceX — and Teaches Its Agents to “Dream” 🤖
At Anthropic’s Code with Claude developer conference in San Francisco on May 6, the company announced it will take all capacity at SpaceX’s Colossus 1 Memphis data center — 220,000+ Nvidia GPUs and 300+ MW coming online within a month. Both companies also “expressed interest” in joint orbital data centers.
The SpaceX angle matters: Elon Musk is currently suing OpenAI in federal court while his rocket company supplies Anthropic’s compute.
Beyond infrastructure, the conference itself was substantive:
API volume grew 80× year-over-year in Q1 2026 vs. a 10× internal plan
Rate limits doubled for Pro, Max, and Team plans; peak-hour Opus caps removed
Claude Opus 4.7 tops the Vals AI Finance Agent benchmark at 64.37%
The standout announcement — “dreaming”:
Anthropic unveiled Claude Managed Agents (research preview) with a capability it calls “dreaming”: when agents are idle, they replay past task sessions to self-improve — inspired by how sleep consolidates memory in humans. Early numbers: Harvey reported ~6× higher task completion rates, Wisedocs cut document review time by 50%.
Anthropic is using Elon Musk’s GPUs while Musk sues their biggest competitor in federal court. The “dreaming” agents story is technically the bigger deal — but the Musk angle won the news cycle.
Source: Anthropic
China’s Moonshot AI: $2B Raised, $20B+ Valuation, and #2 on OpenRouter 🚀
China’s Moonshot AI closed a $2B round on May 7 — led by Meituan’s Long-Z Investment, Tsinghua Capital, China Mobile, and CPE Yuanfeng. Valuation landed at $20B+, more than 4× its late-2025 figure of $4.3B. The company has now raised ~$3.9B in six months.
The product behind the number: April ARR exceeded $200M, driven by Kimi chatbot subscriptions and API usage. The latest model — Kimi K2.6 (1T-parameter MoE, 32B active parameters, 384 experts) — is the #2 most-used model on OpenRouter after GPT-5.5. Modified MIT license: commercial use is allowed, but you can’t use the weights to train competing frontier models.
Why Western developers are routing through it:
It’s not ideology — it’s the spreadsheet. GPT-5.5 costs $30/M output tokens. Kimi K2.6 undercuts that significantly. DeepSeek V4 did the same last week. The demand for cheap Chinese open-weight models is structural.
A Chinese AI company just 4×’d its valuation in under six months on $200M monthly revenue. The US-China AI race isn’t just playing out at the frontier model level — it’s happening in the API economics layer where most developers actually make decisions.
Source: TechCrunch
Share this post with friends, especially those interested in AI!
News of the week 🌍
iOS 27 Will Let You Pick Claude, Gemini, or ChatGPT as Your Default AI 📱 — Mark Gurman reports Apple plans to open iOS/iPadOS/macOS 27 to third-party AI models via “Extensions” — users can set Claude, Gemini, or ChatGPT as the default for Siri, Writing Tools, and Image Playground. Per-feature defaults configurable in Settings. Official reveal at WWDC, June 8–12. The subtext: Apple Intelligence has been behind, and Apple knows it.
OpenAI and Anthropic Both Launched $10B+ PE Ventures — on the Same Day 💰 — On May 4, OpenAI finalized "The Deployment Company" at roughly $10B with TPG, Brookfield, Advent International, Bain Capital, SoftBank, and Dragoneer — PE firms reportedly guaranteed 17.5% annual returns over five years, former COO Brad Lightcap leads it. Hours later, Anthropic announced a $1.5B JV with Blackstone, Hellman & Friedman, Goldman Sachs, Apollo, GIC, Leonard Green, and Sequoia to embed Claude across 2,000+ PE portfolio companies. Both deals explicitly target enterprise consulting — not just API access. McKinsey and BCG now have two well-funded AI-native competitors with PE distribution machines behind them.
EU AI Act High-Risk Deadlines Pushed to 2027–2028 🌍 — The EU Council and Parliament agreed on a Digital Omnibus deal on May 7: Annex III high-risk obligations move from August 2026 → December 2027, Annex I safety deadlines → August 2028, generative-AI watermarking → December 2026. The deal also bans “nudification” AI systems. For any team building AI products for the European market — meaningful runway extension.
Sierra Raises $950M at $15.8B — One-Third of the World’s Largest Banks Are Customers 🏦 — Bret Taylor’s enterprise AI agent company closed a $950M Series E led by Tiger Global and GV. Customers include Prudential, Cigna, Blue Cross Blue Shield, and Rocket Mortgage. Valuation up from $10B last fall. Taylor warned of a market correction “within two years” — in the same week he closed the round.
Pennsylvania Sues Character.AI — Chatbot Posed as a Licensed Doctor ⚖️ — AG Josh Shapiro filed suit alleging a Character.AI bot named “Emilie” claimed an Imperial College London medical degree, a valid Pennsylvania medical license, and produced a fake PA license number when investigators asked about depression and medication. The state seeks an injunction against unauthorized practice of medicine — months after Character.AI settled multiple teen-harm suits in January.
CAISI Signs Safety Agreements With Google, Microsoft and xAI 🔒 — The NIST Center for AI Standards and Innovation signed pre-release safety evaluation agreements with Google, Microsoft, and xAI this week — covering cybersecurity, biosecurity, and chem-weapons evaluations. CAISI has existing agreements with OpenAI and Anthropic from earlier this year. The notable move: xAI joining the framework for the first time, bringing Grok under government pre-release review.
Useful tools ⚒️
⭐ Kanwas — Open-source real-time canvas where humans and AI agents share one self-evolving context for docs, decisions, and research. Shared working memory between your team and your agents — everything stays in sync without copy-pasting across tools.
Mindra — Spin up a team of specialized AI agents that work 24/7, communicate with each other, and take real action across your stack — no coding required. Self-healing: when an agent fails, it re-plans and retries instead of breaking the whole workflow. Built for founders and operators who want to delegate entire workflows, not just tasks.
Velo 2.0 — Turns raw screen recordings into polished video messages with a chat-native editor — no timeline, no re-recording. Describe the edit you want, Velo rewrites the script, clones your voice, and updates the video live. One recording produces both a video and a structured doc.
Kilo Code v7 — Open-source AI coding agent for VS Code rebuilt on OpenCode server. V7 adds real parallelism: multiple tool calls firing simultaneously, parallel subagents via Agent Manager, and inline diff reviewer with line-level comments. Choose any model — Claude, GPT-5.5, Gemini, or local LLM.
Shadow 2.0 — Bot-free AI meeting assistant that runs quietly on your Mac and captures everything said and shown on screen — without joining as a visible participant. V2.0 executes tasks live during the call: CRM updates, follow-up drafts, slide generation — before the meeting ends.
Share this post with friends, especially those interested in AI!
Weekly Guides 📕
Claude Managed Agents Dreaming — Official Anthropic Docs — Official API reference for the dreaming feature: how to configure dream schedules, set auto vs. manual memory updates, use the dreaming-2026-04-21 beta header, and run your first dream in Python. If you're building agents, this is where to start.
How to Use Anthropic’s New Finance Agents — Official Docs — Anthropic’s own walkthrough of the 10 new financial services agents: what each one does, which MCP connectors are included, and how the Moody’s data integration works. If you’re in fintech or building for financial clients, start here.
Vercel AI SDK Agents: Complete 2026 Implementation Guide — End-to-end build guide covering the new Agent abstraction, MCP integration, and Fluid Compute deployment. Published May 6, directly tied to this week’s releases.
Claude Code’s Hidden Skill Budget Setting (May 2026) — Practical walkthrough of skillListingBudgetFraction in Claude Code v2.1.129 — a setting most users never touch that directly affects how aggressively Claude Code burns through your execution budget. Copy-paste fixes included.
Claude for Excel, PowerPoint, Word and Outlook — Official Setup Guide — Launched May 7: Claude now works across all four Microsoft 365 apps in one continuous conversation — triage email in Outlook, open the attachment in Word, build the model in Excel, turn it into a deck in PowerPoint, without re-explaining context at each step. GA for Excel, Word, PowerPoint on all paid plans; Outlook in public beta.
AI Meme of the Week 🤡
AI Tweet of the Week 🐦
Bonus Materials 🎁
AlphaEvolve One-Year Impact Report — Google DeepMind — One year since launch: AC Optimal Power Flow feasibility jumped from 14% to 88%, 10× lower error on Willow quantum circuits, +5% accuracy across 20 disaster-prediction categories. The clearest case study yet for AI coding agents making genuine scientific discoveries.
The State of Global AI Diffusion in 2026 — Microsoft published its Q1 2026 AI Diffusion Report on May 7: global AI usage hit 17.8% of working-age population (+1.5pp in the quarter), 26 economies now exceed 30% adoption, the UAE leads at 70.1%, the US moved up from 24th to 21st. The first data-backed snapshot of where AI is actually being used globally — and where it isn't.
Harvard Study: OpenAI's o1 Outperforms ER Physicians in 67% of Triage Cases — Harvard Medical School and Beth Israel ran six experiments — real ER cases, no preprocessing. AI correctly diagnosed 67% vs 55% and 50% for the two attending physicians. Every single experiment: AI won. The authors don't advocate clinical deployment yet and call for prospective trials — but this is the clearest peer-reviewed evidence that reasoning models are now operating above average physician level on diagnostic tasks.
Publishers Sue Meta — Zuckerberg Personally Authorized Torrenting 267TB of Pirated Books to Train Llama — Court documents allege Zuckerberg personally approved using LibGen to download 267TB of copyrighted books — fiction, nonfiction, academic texts — to train Llama models, overruling legal counsel who flagged copyright risk. If upheld, it directly threatens the legal basis for training any LLM on web-scraped text. The most significant AI IP case since the Getty Images suit.
If you missed our previous updates, don’t worry, here they are:
Claude #1 on App Store, GPT-5.4 Lands, DeepSeek V4 Is Here | Weekly Digest
What’s your take — are AI labs actually a threat to consulting firms, or is this just well-funded positioning? Drop it in the comments 👇