ELON BUYS THE CODING TOOL EVERYONE USES — $60 BILLION IN COLD HARD SPACEX STOCK
So SpaceX just bought Cursor for sixty billion dollars. Let that number roll around in your head for a second. Sixty. Billion. For a four-year-old startup built by four MIT kids that makes a code editor. Just days after SpaceX went public in the biggest IPO the market has ever seen, Elon Musk’s rocket company turned around and spent its shiny new public currency on an AI coding tool that half the developers in Silicon Valley are already hooked on.
Here is what Cursor actually is, for anyone who hasn’t used it. It is basically an IDE, a place where programmers write code, except it has a very capable AI baked in that watches what you’re doing and helps you write, fix, and understand code in real time. Think of it like autocomplete but one that actually understands what you’re trying to build rather than just guessing the next word. The thing genuinely changed how a lot of developers work, and it grew fast. Explosive growth is the polite way to put it. It went from a curiosity to a daily driver for enormous numbers of professional coders in a very short window.
Now Musk owns it. And that is where things get interesting, because SpaceX doesn’t really have an obvious reason to own a coding assistant. They build rockets. They run Starlink. They are not, historically, an enterprise software company. But look at the bigger picture and it starts to make sense. xAI, Musk’s AI company, has been building coding tools as well. Cursor gives SpaceX an instant enormous user base in exactly the developer community that every AI company in the world is trying to capture. Developers are the tastemakers of the AI era. They decide which tools their companies adopt. They build on top of platforms. If you own the thing developers use to write code all day, you have your hands on something strategically important.
The deal also makes Cursor’s four co-founders billionaires, which is a nice outcome for a team that, not so long ago, was probably arguing over pizza toppings in a dorm room. The acquisition is expected to close in the third quarter of 2026, pending the usual regulatory review that nobody expects to be a real obstacle.
The number that keeps coming back to me is that this is the largest acquisition of a venture-backed startup in history, outside of Musk buying xAI for himself, which was its own kind of wild. Sixty billion dollars, paid in SpaceX stock that was trading at peak value right after the IPO. Whether that stock holds its value is a separate question. But the move signals something: the AI coding tools race is now a full-contact sport, and the prize money is not small.
Source: TechCrunch | CNBC
WASHINGTON PULLS THE PLUG ON ANTHROPIC’S MOST POWERFUL MODELS — AND NOBODY REALLY KNOWS WHY
The US government walked up to Anthropic last week and said, essentially, your two best AI models are now national security assets, and we are cutting off everyone who isn’t American. Fable 5 and Mythos 5, Anthropic’s most capable models, were hit with an export control directive from the Commerce Department, and Anthropic had no practical choice but to comply. So they shut off access for everyone. Not just foreign users. Everyone, at least temporarily, while they figured out how to implement the restrictions.
The justification the government gave, at least in verbal form, was a claimed jailbreak of Fable 5. Someone apparently showed officials a method to prompt the model in a way that could help identify vulnerabilities in software code. A cybersecurity trick. But here is the part that made a lot of people in the security community furious: the jailbreak was described by Anthropic’s own people as narrow and non-universal. It wasn’t some skeleton key that unlocked chaos. It was a specific, limited technique, and Anthropic said the government only provided verbal evidence of it, not a written technical report, not a reproducible demonstration they could actually study and respond to.
There is a darker reading of this whole situation that some people in the industry have floated. Anthropic, more than any other frontier AI lab, has been loudest about warning that advanced AI could be genuinely dangerous. They published safety papers, they testified to Congress, they built their entire brand around the idea of responsible development of powerful systems. And now the government cited those very warnings as part of the justification for treating their models as too dangerous for the world to have. The company that spent years telling everyone to take AI risk seriously may have inadvertently handed the government the argument it needed to regulate them first and hardest.
This sent shockwaves to the G7 summit in France, where world leaders are meeting right now and where, as it turns out, American AI models are a very hot topic. Emmanuel Macron was apparently looking for ways around the ban. Other governments are alarmed. The pattern here is not subtle: Washington is treating frontier AI like nuclear technology, with export controls and national security reviews. That is a very different world than the one the AI industry thought it was operating in a year ago.
Anthropic is meeting with the Trump administration to try to sort this out. The outcome of that conversation will tell us a lot about how American AI policy actually works when it collides with American AI companies.
Source: TechCrunch | Axios
TRUMP SITS DOWN WITH THE AI BOSSES IN FRANCE — AND THE REST OF THE WORLD WATCHES NERVOUSLY
The G7 summit in Evian, France turned into something nobody quite expected: a meeting where the presidents and prime ministers of the world’s wealthiest democracies sat down to lunch with the CEOs of OpenAI, Anthropic, and Google DeepMind. Sam Altman was there. Dario Amodei was there. Demis Hassabis was there. Donald Trump was there. If you are the kind of person who thinks about power and who holds it, this is a photograph you stare at for a while.
The official topics were frontier AI risks, infrastructure, and sovereignty. The unofficial topic, hanging over everything, was the Anthropic export ban that had just dropped days before. Foreign governments, including some very close American allies, are now staring at a world where the United States can flip a switch and cut off access to the most powerful AI systems on the planet. France’s Macron was reportedly pushing hard to find some kind of workaround. Other European leaders are asking the same uncomfortable question: if we build our economies and security systems around American AI, what happens when Washington decides to act in its own interest?
Amodei and Hassabis reportedly called for a US-led global coalition on AI standards. They want America to be the country that sets the rules, rather than ceding that ground to China or a fragmented patchwork of national frameworks. It is a reasonable argument, and it also happens to align perfectly with the interests of American AI companies, which would benefit enormously from being the ones operating under the internationally endorsed standard. OpenAI’s Chris Lehane talked about creating a global forum for AI governance. Everyone is very eager to be at the table where the rules get written.
What is striking is that this is now visibly a geopolitical issue, not just a tech policy issue. AI was already being treated like that in government circles, but seeing it play out at the G7, with tech CEOs at the same table as heads of state, makes it concrete. These companies are not just startups anymore. They are being treated like strategic national assets, invited to summits where trade policy and military alliances get negotiated. That changes what they are, and it changes what the stakes are for everyone trying to navigate this space.
Source: CNBC | TechCrunch
SOMEONE HAS TO TEACH THE ROBOTS TO PICK UP A GLASS OF WATER — MEET THE STARTUP GETTING PAID TO DO IT
Here is a story that doesn’t get nearly enough attention: the robots are coming, yes, everyone knows that, but before they can do anything useful they have to learn how to do things the same way a toddler learns to walk. By watching, by trying, by failing, by getting data fed back into their systems that tells them what went wrong. And that data has to come from somewhere. It doesn’t exist yet in anywhere near the quantities needed. That is the problem XDOF is betting $70 million that they can solve.
XDOF just came out of stealth with a raise from Thrive Capital, Andreessen Horowitz, Spark Capital, and Lux Capital. Their pitch is straightforward and a little unglamorous: they collect the physical-world training data that AI labs need to teach robots how to actually interact with objects. Pick up a cup. Open a drawer. Sort objects in a bin. These sound trivially simple until you try to teach a robot to do them, at which point you realize that the robot has no intuition, no prior experience of the physical world, and needs to see those tasks performed thousands of times in structured, labeled, high-quality data before it can reliably get anywhere close.
The company’s founder described collecting robot training data as dirty and unglamorous work. That is a refreshingly honest sales pitch. They are not claiming to have cracked robot intelligence. They are claiming to be the people willing to do the grinding, methodical work of building the datasets that make robot intelligence possible. And they think that is going to be worth a lot of money, because the labs building humanoid robots, and there are quite a few of them now, all have the same problem: the data feedback loop doesn’t exist yet.
OpenAI just relaunched the robotics program they shut down back in 2021, which tells you something about how the industry reads the current moment. Everyone is racing to teach machines to operate in physical space. The bottleneck is no longer the model architecture or the compute. It is the data. That is a good time to be the company that specializes in gathering it.
XDOF has three tiers of data collection: teleoperation on the actual robot being deployed, which is the highest quality; teleoperated robots doing more general physical tasks; and human egocentric data using wearable sensors they are building themselves. None of this is glamorous. All of it might be essential. In ten years, when robots are doing useful things in warehouses and hospitals and homes, the unglamorous data company might look like one of the smarter bets made in this era of AI investment.
Source: TechCrunch | Axios