Welcome back! The AI infrastructure race entered a completely new phase. Elon Musk unveiled plans for a chip factory that would produce 50 times the world's current compute output per year, with most of it deployed on satellites in orbit. The ambition is staggering. But is it the fantasy of a lunatic?
In today’s Generative AI Newsletter:
Musk: Can a $25B chip factory really put AI compute in space?
Anthropic: What does messaging Claude Code from your phone change?
Meta: Why is Zuckerberg building an AI agent to help him be CEO?
Snowflake: What happens when a company records its writers for 8 months then lays them off?
Latest Developments
Musk Unveils $25B 'Terafab' AI Chip Factory
Elon Musk just announced Terafab, a joint Tesla-SpaceX-xAI chip facility in Austin that aims to produce a terawatt of AI compute per year. That is roughly 50 times what the entire world outputs today.
Vertical integration: The facility would handle logic, memory, packaging and testing under one roof. Musk says that level of integration exists nowhere else.
Two chip types: One designed for Tesla vehicles and Optimus robots. A second space-grade chip built for solar-powered AI satellites launched via Starship.
The space pitch: Musk argued that nobody wants AI data centres in their backyard and predicted space-based compute will undercut ground costs within two to three years.
The vision: Musk framed the project as the first step toward a "galactic civilisation" with a post-scarcity economy.
Building a chip fab from scratch is an enormous bet. The facilities are expensive, slow to develop and notoriously difficult to operate. But the demand for AI compute is real and accelerating. Musk has built a career on ignoring what the industry says cannot be done.
Special highlight from our network
LLMs are powerful, but they only know what they’ve been trained on.
Retrieval-Augmented Generation (RAG) helps them go further by pulling in fresh, relevant data at the moment a response is generated.
That means more accurate responses and outputs grounded in your/ your company’s knowledge.
MongoDB Atlas makes it possible with semantic search built into your data.
Start building with MongoDB
Special highlight from our network
Your best context already exists: it’s in your calls.
Most AI tools start empty. You write the prompt, fill in the gaps, and provide the context. The output is only as good as what you put in.
Supernormal captures every meeting automatically. No bots joining. No setup. No awkward intros with clients.
As soon as your call ends, your work is already in motion:
Recap decks and slide presentations
Follow-up emails and client briefs
Strategy docs and project summaries
Budget spreadsheets
Turn every meeting into completed work right away.
Anthropic Ships Claude Code Channels
Anthropic released Claude Code Channels, a new feature that lets users message existing Claude Code sessions from Telegram or Discord. The company also rolled out support for recurring tasks, letting users automate routine coding workflows.
Remote access: Channels lets developers send instructions to active Code sessions from their phone without needing a terminal open.
Recurring tasks: Users can now schedule and automate repetitive workflows directly inside Claude Code.
Research preview: The feature is live now as a research preview, with plans to expand it further.
This directly mirrors functionality that OpenClaw introduced last month. OpenAI hired OpenClaw's creator shortly after. Anthropic clearly decided it could not leave that gap open.
Zuckerberg Is Building an AI Agent to Help Him Be CEO
Mark Zuckerberg confirmed he is building a personal AI agent to help him run Meta. He described a future where everyone has their own agent handling day-to-day tasks.
Current use: The agent helps Zuckerberg retrieve information faster and manage his workflow.
Company-wide push: AI tool usage is now a factor in employee performance reviews at Meta. The company runs AI tutorial meetings several times a week and hosts regular hackathons.
The bigger picture: Zuckerberg framed this as the starting point for a world where personal AI agents are universal.
The performance review detail is the most telling part of this story. Meta is not just encouraging AI adoption internally. It is tying it to career progression. That is a fundamentally different signal to the rest of the industry.
Snowflake Recorded Its Tech Writers for 8 Months, Then Laid Them Off
Snowflake is reportedly cutting around 400 jobs. The company had been screen-recording every documentation session for eight months, building training datasets from its senior technical writers' workflows.
Knowledge transfer: Senior writers spent their final six weeks transferring knowledge directly to the AI system that would replace them.
The claim: Snowflake says documentation quality has not dropped. Management is celebrating 300% efficiency gains from the new AI documentation pipeline.
The scale: The reduction is part of a broader workforce cut targeting technical writing and documentation teams specifically.
This is one of the most concrete examples yet of a company systematically using its own employees to train their replacements. The 300% efficiency claim will get the headlines. The eight months of screen recording will stick in people's minds longer.
Tool of the Day: Google Stitch 2.0
Google's updated Stitch tool lets you vibe-design production-ready UI in seconds. Upload a screenshot of any existing page, describe what you want improved and Stitch generates multiple layout variations you can export to Figma or download as code.
Try this yourself:
Take a screenshot of a page on your site that needs work.
Go to Google Stitch, upload it and prompt: "Improve the layout so the user sees more content and perceives more value. Reduce the dead space."
Generate variations, pick the best one and export.
You can also click Export > AI Studio to build a live prototype with Google's vibe coding tool.
