Welcome back! Anthropic bought the company that ships OpenAI's SDK yesterday (yes, that's exactly as awkward as it sounds). A federal jury said Musk waited too long to sue OpenAI and tossed the case in under two hours. Cursor's new coding model undercut Claude and GPT on price. And Google teamed up with Blackstone on a $5 billion AI cloud built around its TPUs.
In today's Generative AI Newsletter:
Stainless deal: Why did Anthropic buy the company that ships OpenAI's SDKs?
Musk vs Altman: What's left of Musk's case after a two-hour jury verdict?
Composer 2: Can Cursor's own model undercut Anthropic and OpenAI?
Google + Blackstone: What does a $5 billion bet on TPUs do to Nvidia?
Anthropic paid a reported $300 million plus for Stainless, the company that turns an API spec into clean SDKs across TypeScript, Python, Go, Java, Kotlin and more. It also powers the MCP server tooling Anthropic built to give Claude agents real access to outside systems.
Here's why that matters:
When developers build apps that talk to AI they use a ready-made toolkit called an SDK.
Stainless is the company that builds those toolkits for every major AI lab, OpenAI included.
Stainless still ships OpenAI's SDKs for now.
Google, Cloudflare and Replicate also depend on Stainless.
Anthropic isn't shutting OpenAI's SDK down tomorrow. It now sits underneath every major AI developer's stack and decides who gets the next feature first when there's a tie. Founder Alex Rattray said the team gets to keep doing the work it loves. They keep doing it for Anthropic's competitors too, until Anthropic decides otherwise.
Many of you have been part of GenAI Works from the beginning. What started as AI content is now a $2M+ revenue business with strong growth:
2.4x annual revenue growth
300+ paying customers (NVIDIA, Google, IBM, ElevenLabs)
14M+ audience driving distribution
Now we’re entering the next stage, which is building our own AI products. GenAI Academy already has 30K users, our 9 hackathons drew 28K participants and led us to build the Hackathon App, and ToneUp powers our 14M+ content machine.
We’ve now opened a crowdfunding round to scale product development and expansion.
CTA: Invest until July 1 and earn up to 25% extra shares.
In making an investment decision, investors must rely on their own examination of the issuer and the terms of the offering, including the merits and risks involved. Genai Works, Inc. has filed a Form C with the Securities and Exchange Commission in connection with its offering, a copy of which may be obtained here: https://bit.ly/3APlUkJ
Special highlight from our network
Most AI chatbots invent answers. Trained on the entire internet, they sound confident even when they get it wrong. Can your business afford to give wrong answers?
What if your AI only knew what you taught it?
That's CustomGPT.ai.
Every response comes from your content, with citations. No source? It simply says so.
MIT, Dropbox, UN and Adobe all chose it for exactly that reason.
We've been covering this story for a while now. The trial started in late April in Oakland and ran three weeks. On Monday, a federal jury took under two hours to side with Sam Altman.
Musk's argument that Altman "stole a charity" by turning OpenAI for-profit never reached the jury at all. They decided he waited too long to file, outside the three years statute of limitations, and never voted on the substance.
Musk called the result a "calendar technicality" on X and promised an appeal. No reason to get excited here, though. The judge said she was ready to dismiss that appeal "on the spot."
Musk's team had wanted $180 billion clawed back, Altman and Brockman pushed out of leadership and the 2025 for-profit restructure unwound. They got nothing, and Microsoft, also named in the suit, walked away clean.
Btw, Greg Brockman (OpenAI's president, equity now worth around $30 billion) had one of the cleaner lines of the trial about Musk: "He knows rockets, he knows electric cars. He did not, and I believe does not, know AI."
What do you think? Does the drama end here, or are we seeing more of Musk and Altman in court?
Cursor shipped Composer 2, the coding model that powers its AI editor. It hits frontier level on Terminal-Bench 2.01 and SWE-bench Multilingual. And it costs $0.50 per million input tokens and $2.50 per million output tokens, well below Claude or GPT for similar work.
The details:
Continued pretraining: Cursor's first "continued pretraining" run gave the model a stronger base before reinforcement learning kicked in.
Long-horizon RL: They trained Composer 2 with reinforcement learning on coding tasks that need hundreds of actions in sequence.
Fast variant: A faster version with the same intelligence costs $1.50 in and $7.50 out, and is now the default.
Cursor used to live on top of someone else's model. With Composer 2 it has a competitive model of its own at a price well below Anthropic and OpenAI.
If you're already paying Cursor for the editor, you have less reason to pay one of the labs on top. Whether the model feels as sharp as Claude Code in daily work is the test this week.
Google is teaming up with Blackstone on a new AI cloud company. Blackstone is putting in $5 billion as initial equity, and the venture will sell data center capacity and Google's custom Tensor Processing Units on a pay-as-you-go basis.
The first 500 megawatts of capacity come online in 2027. Benjamin Treynor Sloss, a 20-year veteran of Google's infrastructure team, runs it as CEO.
Google has been selling TPUs through Google Cloud for years. Standing up a separate company with Blackstone capital lets it scale chip rentals fast without dragging Google Cloud's balance sheet through every new data center build.
More compute and less mess for Google. Let’s see how that plays out.
On a side note, are you as excited as us about Google I/O 2026? Stay tuned for our coverage!
Rox is an AI revenue platform that gives every account executive their own swarm of agents. The agents pull together CRM, finance, support, product telemetry and web data, then research accounts, summarize meetings, draft personalized outreach and auto-fill proposals on demand. They run asynchronously while reps are offline and show up as co-pilots inside Slack, web, iOS and macOS during the day.
Try this yourself:
Sign up at rox.com and connect your CRM, calendar and email so the knowledge graph has something to chew on.
Drop in a list of your target accounts and let Rox spin up one agent per account.
Open the agent dashboard in the morning to see what each one drafted overnight before your first call.
Use the Slack integration to ping any agent for a fast account brief without leaving the channel.
GitHub changes Copilot's default model: GPT-5.3-Codex replaced GPT-4.1 as the base model for Copilot Business and Enterprise.
Anthropic gets a papal cosign: Co-founder Christopher Olah is on the launch lineup for Pope Leo XIV's first encyclical, "Magnifica Humanitas," a Vatican document on AI publishing May 25.
The Meta cuts are here: Meta starts cutting roughly 8,000 jobs this week even as AI spending keeps climbing.
Alexa+ does podcasts now: Amazon's Alexa+ can generate AI podcasts hosted by two robot co-hosts on demand
Learn more about AI from the experts building it
📸 Follow us on Instagram for fast, visual AI updates in 30 seconds.
🐝 Subscribe to our Atlas newsletter — trusted by 3M+ subscribers — to stay ahead of AI news across tech, education, and business.
📺 Watch us on YouTube to hear insights directly from leading AI voices, builders, and innovators.
🐦 Follow us on X for breaking AI news and real-time industry updates.
🧠 Learn how to build your next AI application with practical resources and expert guidance.
🚀 Explore investment opportunities in the future of AI and join our community-backed growth journey.
