This website uses cookies

Read our Privacy policy and Terms of use for more information.

Welcome back! SpaceX has been a public company for about a week and already dropped $60 billion on the coding app a million developers live in, and the reason it wants Cursor points straight up at space. Cursor, for its part, is done renting other people's AI and is building its own from scratch on Elon's machines. The money stories are wild too. OpenAI's secret books leaked and the loss is enormous, while DeepSeek pulled in $7.4 billion and handed its backers almost no say in return.

In today's Generative AI Newsletter:

  • SpaceX: Why does a rocket company need the coding app on your laptop?

  • Cursor: What's Cursor building now that Elon's paying the compute bill?

  • OpenAI: How deep is the hole under OpenAI right before it goes public?

  • DeepSeek: What did DeepSeek's backers give up to get in?

SpaceX has been a public company for four days, and it just agreed to spend $60 billion of fresh stock on Cursor, the AI coding app more than a million developers open every day. The deal closes by the end of September, and then a rocket company owns your code editor. Grok's behind on code, so "they wanted a better coding tool" is the easy read. That's also way too small to explain a $60 billion check.

Back in January, SpaceX asked the FCC for permission to put up to a million solar-powered satellites in orbit and run them as data centers, 500 to 2,000 kilometers up. The FCC took the filing and opened it for public comment. So yeah, SpaceX wants to move computing off the planet.

It gets weirder. In February, Musk told his xAI team he wants a factory on the Moon that builds those satellites out of Moon rock and flings them into orbit with a giant electromagnetic catapult. Two of Cursor's top engineers had already jumped to SpaceX in March to work on xAI and the lunar stuff.

Connect the dots. You can't send someone up to reboot a server floating in space, so running data centers up there means AI that can write, watch and fix the software on its own. That's the thing Cursor does. SpaceX is buying the robot engineers before it's built the thing they'd run.

They first need to solve a few physical problems before anything. The AI part is there, but a lot of work needs to be done. We’ll see how this plays out in the future. 

Special highlight from our network

Anthropic ships new Claude capabilities almost every week. Skills, Connectors, Cowork, vibe coding. Most professionals are still using Claude for one-off chats, missing the modes that actually change how work gets done.

Outskill's 2-Day Claude AI Mastery Workshop is a live, end-to-end walkthrough taught by mentors who condensed 800+ hours of research and real-world testing into one weekend.

You'll walk away knowing:

  • How to use Claude's three modes: Chat, Cowork, and Code

  • How to set up Skills and Connectors that automate your Notion, files, and desktop

  • How to vibe code working apps and dashboards alongside 10+ tools that pair with Claude

Saturday and Sunday, 10 AM to 7 PM EST. Free for the next 48 hours.

AI is reshaping how we work. Discover these GenAI Academy courses curated by the world’s top AI experts to help you work faster and achieve more with less effort.

Cursor previewed its next model on Tuesday, and this one's built from scratch, their first ever. Until now, Cursor's coding AI was basically Moonshot's Kimi, a Chinese open model, with Cursor's polish on top. 

It was the quick way to get something good out the door, and building from zero means the whole thing is finally theirs.

So what's the new one? 

Cursor says it's in the same league as Claude Opus and GPT-5.5, runs on 10 to 20 times the compute of their current model and is smart across the board, not just at coding. 

It lands in the next couple of weeks. The only reason they can train something this big is SpaceX, which handed over its Colossus supercomputer.

There's no benchmark or demo yet, so for now it's a promise. But if it shows up in two weeks and actually hangs with Opus, Cursor stops being a skin over someone else's model and becomes a real lab. Not bad for a company that just sold itself to a rocket company.

Somebody leaked OpenAI's audited 2025 books right before its IPO, and the Financial Times says they're real. The big number is a $38.5 billion loss, up from $5 billion the year before. That's the bleeding getting almost eight times worse in twelve months.

Most of that is an accounting mirage. When OpenAI turned from a nonprofit into a for-profit last year, it had to book a one-time $41.55 billion charge for the ballooning value of the ownership stakes it owes investors. No money actually left the building.

It's like owing a friend a slice of your startup. The startup's value explodes, so the slice you owe them is suddenly worth a fortune. You owe more now, and on the books, that bigger debt counts as a loss, even though you never hand over a dime.

Strip that charge out and the actual business still lost $20.9 billion, on $34 billion spent against just $13 billion of revenue. Roughly half its costs, about $17 billion, went straight to Microsoft for Azure.

Now look at Anthropic, same race, same kind of spotlight. It just told investors it expects its first profitable quarter right now, about $559 million in operating profit on $10.9 billion of revenue. So while OpenAI heads to its IPO trying to explain a $20 billion hole, its closest rival is about to turn a profit. The wild part is they're selling shares to the same investors months apart.

DeepSeek took outside money for the first time ever, $7.4 billion at a valuation north of $50 billion, per The Information. Tencent, CATL, JD, NetEase and IDG Capital all wrote checks. Then they found out how little that actually buys you.

To get in, investors had to park their cash in a fund run by founder Liang Wenfeng instead of buying into DeepSeek directly. No voting rights, and they're locked in for five years. The only backer that got real ownership and an actual vote is China's state AI fund. Read that twice.

A year and a half ago, DeepSeek's dirt-cheap models knocked about a trillion dollars off US tech stocks in a single day. Now it's pulling in billions on its own terms, with a Chinese state fund as the only outside investor that gets a real say. For a company Washington's already jumpy about, that's the part to sit with.

Your AI writes code that looks perfect and breaks the moment it hits production. AppMap is built to catch that before it ships. It watches your app actually run, recording the real function calls, database queries and requests, then checks your new code against how it behaves instead of how it reads. Think of it as a second set of eyes that has seen your app in motion, not just the diff.

Try this yourself:

  • Install the AppMap plugin in VS Code or a JetBrains IDE. The free tier covers solo work.

  • Run your app or test suite once so it records how the code really behaves, no code changes needed.

  • Open the diagrams it builds to follow one request through your functions, SQL and auth in a single view.

  • Ask Navie, the built-in assistant, why something broke and get an answer grounded in the actual run, not a guess.

  • Who it's for: anyone shipping AI-written code who wants a second set of eyes that knows how it runs.

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.

Reply

Avatar

or to participate

Keep Reading