SpaceX Closes in on Outlaunching the Rest of the World Combined as Anthropic Reportedly Raises at $950B

SpaceX is now about 200 satellites away from having launched more orbital payloads than every other country and company in history combined, which is wild considering everyone else got a 61-year head start. Google is reportedly in talks with SpaceX to launch its own data centers into orbit, which is part of a broader push to put compute infrastructure in space. Starship Flight 12, which debuts the new V3 version of the rocket, is targeted for as early as May 19. Musk also confirmed that SpaceX is scouting new launch sites both in the US and abroad to keep increasing how often it can fly. Ron Baron is now predicting SpaceX hits a $30 trillion valuation in the next 10 to 15 years. And a startup called Star Catcher just raised $65M to build what is effectively a power grid in orbit, beaming optical energy down to existing satellites to give them 2 to 10x more power on demand.

The sky is also starting to give up some of its secrets. Japan's government said it is analyzing the UAP files that the Pentagon recently released under the PURSUE Act, including videos shot near Japan, and will begin doing its own case-by-case disclosure. Congressman Tim Burchett, who pushed PURSUE through, responded with a single word on social media: "Dominoes."

All of this still takes a staggering amount of electricity. xAI added 19 gas turbines to its second data center campus, Colossus 2, in Southaven, Mississippi, in just the last two months, basically routing around the grid by generating its own power on-site. Ames National Lab also rolled out DuctGPT, an AI system designed to find new materials for fusion reactors. The goal is to cut materials research from months down to hours, which would help bring fusion energy online and eventually replace those gas turbines with something cleaner. While the digital side keeps scaling, the physical side is starting to scale walls too. A Chinese company called RobotPlusPlus debuted a humanoid robot that uses magnetic wheels to climb vertical steel surfaces in chemical plants, shipyards, and energy facilities. It can switch tools at the wrist to do welding, inspections, rust removal, and spraying, all in places that are too dangerous for human workers.

The same kind of AI capability is also being packaged for specific industries. Anthropic just launched Claude for the legal industry, which connects Claude to more than 20 of the software tools that law firms actually use, along with 12 specialty plugins for different practice areas like litigation, contracts, and IP. They are also partnering with the Free Law Project and the Justice Technology Association to make legal help available to people who currently cannot afford it. Google is going deeper into the operating system layer with Gemini Intelligence, which lets users build their own Android widgets just by describing what they want, and a new Gemini-powered mouse pointer that understands what you are pointing at on screen. Google also announced the Googlebook, a new laptop launching this fall that combines ChromeOS and Android into one Gemini-optimized operating system. It is Google's direct shot at Apple's MacBook Neo.

Intelligence is also moving into the human body. Columbia researchers built the first real-time brain-controlled hearing system. It reads high-resolution brain signals from electrodes inside the skull to figure out which voice in a noisy room you are actually trying to listen to, then amplifies that voice and quiets the others. This is the "cocktail party problem" that regular hearing aids have never been able to solve. Isomorphic Labs, which is Alphabet's drug discovery company, also raised $2.1B in a round led by Thrive to scale up AI-driven drug development.

The economy is repricing all of this in real time. Anthropic warned investors this week to stay away from eight unauthorized secondary marketplaces selling its stock, just as the company is reportedly in talks to raise up to $50B at a $950B valuation. Princeton, meanwhile, just voted to end its 1893 honor code, which has let students take exams without a proctor for over 130 years. Starting this summer, all in-person exams will be proctored, because AI has made cheating easier and harder to detect at the same time. And in Hollywood, struggling screenwriters are now calling AI training gig work "the new waiting tables", signing up with platforms like Mercor to train the same models that will eventually replace their work.

The models themselves keep getting smarter on the benchmarks. A new evaluation called ProgramBench, which tests whether AI can rebuild programs from scratch, just had its first task solved by both GPT-5.5 high and xhigh (xhigh is the higher-reasoning version), with xhigh dominating the rest of the benchmark. A separate evaluation called the AI IQ meta-eval, which takes 12 existing benchmarks and maps them onto implied IQ scores, gave GPT-5.5 a score of 136, which is past the Mensa cutoff. AI agents are also starting to write their own instructions. Users are now prompting OpenAI's Codex to draft its own "/goal" file, and one of them described the resulting setup as the most powerful AI agent configuration currently available.

That's today. More tomorrow.