Here's what I've got for you today:

Musk testified for three days in federal court that OpenAI's Sam Altman betrayed the founding mission of the company they started together. Five days after his testimony wrapped, his SpaceX leased 220,000 GPUs to Anthropic, OpenAI's biggest rival.

What's happening

The trial opened on 28/04/2026. Musk dropped his fraud claims days before trial and is suing on breach of charitable trust and unjust enrichment. He's chasing more than $130 billion USD in damages. The standout exhibit is a late-2022 email exchange. Musk: "This is a bait and switch." Altman: "I agree it feels bad."

On 06/05/2026, Anthropic doubled Claude Code rate limits for Pro, Max, Team and Enterprise users, and lifted peak-hour throttling for Pro and Max. The reason: it had taken over the entire Colossus 1 data centre in Memphis, originally built for Musk's xAI. 220,000 Nvidia GPUs, available within the month.

For Anthropic the timing isn't coincidence. Anthropic's CEO Dario Amodei said in early May the company grew 80 times in a quarter against a planned 10x. Compute was the bottleneck.

For SpaceX, the deal lands ahead of an IPO. New Street Research estimates it will bring SpaceX $3 to $4 billion USD a year.

Read the full recap →

What it means

OpenAI has restructured three times in a decade. Nonprofit in 2015, capped-profit (investor returns capped at a multiple) in 2019, and now a Public Benefit Corporation. The OpenAI Foundation, the original nonprofit, holds about $130 billion USD in equity. Musk's case rests on the original charitable mission, which keeps shifting beneath the litigation.

His own posts make the contradiction obvious. Three months ago he called Anthropic "misanthropic and evil" and accused it of hating "Western Civilization." The day he announced the deal: "No one set off my evil detector."

Anthropic was founded in 2021 by OpenAI alumni who left over the same commercialisation pivot Musk is now litigating. The compute deal accelerates the alternative his lawsuit gestures at.

The quieter story is what xAI couldn't do. Grok, xAI's chatbot, had already moved its training to Colossus 2, leaving the original GPUs idle. xAI moved too slowly to use them, so SpaceX rented them to the company that could.

Why you should care

If you're not already using Claude in your business, your competitors might be. JPMorgan, Goldman Sachs, and Citi just signed up for Anthropic's new finance agents. Harvey, the legal-AI platform used by A&O Shearman and PwC, added Claude to its model lineup last year. And Apple's Support app shipped on 30/04/2026 with internal Claude config files inside, pulled within 24 hours.

OpenAI is reportedly missing the revenue and user targets it needs for its $1 trillion IPO. The lawsuit could be the trigger, not the cause. The Anthropic deal makes that worse.

The trial argues over who broke the founding promise; the deal decides who writes the next chapter. The next six months show what an unconstrained Anthropic looks like. The man suing OpenAI is the one who set it up.

Last Tuesday, the company released 10 Claude agent templates for financial work. They handle pitchbooks, KYC screening, month-end close, earnings reviews, and more. Plus integrations with Excel, PowerPoint, and Word, and a Moody's app embedded inside Claude.

Claude's been moving through industries like a grim reaper. Financial services is the latest. FactSet shares dropped around 8% on the news; Moody's, Morningstar and Thomson Reuters all sold off too.

If your finance team or accountants pay for any of the financial-data incumbents, the bill just started looking different.

Read more →

→ Here's the BYS pack: 01: Obsidian setup. The skill, the one-prompt master, the conversation-sync prompt, the vault starter. Bookmark it, it'll wait.

I came into this cynical. You have to be when you test AI products for a living, and most of what I see is gimmicks dressed up as breakthroughs. Going in, I assumed this'd be another one. Two weeks later though, this one's actually useful, just not in the way I expected.

What I wanted was a neural network in the background, supercharging any AI tool I used because it knew everything about my business. And that's what I got... kinda.

There's a lot of standout stuff people show off online: pattern recognition, clever queries, daily reports landing in your vault. Ask it things like "what have I brought up repeatedly in the past month?" or "what's the obvious thing I keep missing across all this context?", and what comes back is a real synthesis. No other tool gives you that, especially in the first week.

But the novelty dies off pretty quickly, and none of that is really what earns its keep.

The honest catch is you really need Claude Code to make this worth it. Claude Desktop and Claude.ai can technically call the Obsidian vault, but it's slow, and they can't sync your conversations into the vault automatically. If you do most of your AI work in the web app like most people do, your actual back-and-forth with Claude is the most interesting context you generate. And it sits outside the system unless you do something about it. The conversation-sync prompt in the hub fixes that for Claude Code, but the web app is a dead end for now.

Which brings me to the part I didn't expect. This project genuinely pulled me into using Claude Code as my main AI for business work, and I'm grateful for the push. Two weeks ago I was barely using it. Now it's where I do most of my AI work, because once everything's wired up, going back to Claude Desktop just feels slow. The vault, the syncs, the conversation history all sit in one place. Claude Code runs locally so everything's instant, it can edit files directly, and every conversation I have ends up in my vault automatically as part of the same memory.

Verdict: TRY IT, especially if you're non-technical. Sounds counterintuitive, but hear me out. The master prompt in the hub does most of the heavy lifting for you, and the BYS skill researches the right setup for your specific situation. You walk away with a real source of truth for all your information, and just as importantly, you start understanding how to talk to a computer directly instead of relying on packaged apps to translate for you. That second part compounds. Every AI thing you try after this, you'll get more out of, because you've stopped treating AI as a magic black box.

If you do run a business or agency, the daily payoff is even more obvious. But the setup works for anyone who's tired of starting from scratch every conversation, regardless of how technical you are.

One non-negotiable rec (not sponsored): use Warp instead of your standard terminal. It has an AI built in, so if you get stuck halfway through setup, you can just query the agent inline and it'll explain what's happening. For non-coders, that's the difference between giving up and finishing.

→ Full how-to, downloads, scoring detail: 01: Obsidian setup

That's a wrap for this week. If something sparked a thought, shoot me a DM. I read every one.

Back Sunday. Maran