Welcome back to the Technology newsletter 🤖
Anthropic and xAI spent the last two years calling each other names, but this week Anthropic essentially took over the entire supercomputer xAI built to train Grok.
Also, Cloudflare cut 1,100 jobs and pointed the finger straight at AI, Apple's reportedly testing AirPods with cameras inside them, and a three-hour indie game about high school is getting some of the year’s best reviews.
Let's get into it 👇
☁ Cloudflare (the company that protects roughly a fifth of all websites) announced Thursday it's laying off about 1,100 employees, around 20% of its workforce, as it shifts to what CEO Matthew Prince called an "AI-first operating model." The kicker: Cloudflare just reported revenue up 34% year over year. The business is doing fine; the company just decided it can run with fewer humans. The stock fell a whopping 18% after-hours. With revenue up, the business is thriving… even if the humans aren’t.
🫀 Google announced Thursday it's renaming the Fitbit app to "Google Health" starting May 19 and adding an AI Health Coach powered by Gemini that summarizes your sleep and fitness data, builds personalized weekly workout plans, and answers questions about your medical records. It's bundled with a $9.99/month Google Health Premium subscription (free if you already pay for Google AI Pro or Ultra). Google also unveiled a tiny new tracker called the Fitbit Air, which goes on sale May 26. Google already knows what you search. Now it wants to know how you sleep.
📤 As of today, May 8, Meta is killing the end-to-end encryption option for Instagram DMs that it rolled out back in 2023. The company says barely anyone was using it. The practical effect: from Friday onward, anything you send in an Instagram DM can be read by Meta, handed over to law enforcement with a warrant, and stored on Meta's servers like a normal Facebook message. Anyone who wants encrypted messaging is changing lanes over to WhatsApp, where it's still on by default.
🧑⚖ The Musk vs. Altman trial got juicy on Thursday. Former OpenAI board member and the mother of four of Musk's children, Shivon Zilis, testified that during 2017–2018 negotiations to fold OpenAI into Tesla, Musk personally offered Sam Altman a seat on the Tesla board to sweeten the deal. Altman ultimately rejected the merger. Internal text messages produced as evidence also showed Musk was actively trying to poach OpenAI talent at the same time he was publicly fighting with the company about its direction. Musk was fighting OpenAI in public and raiding it in private. The texts don't lie.
🚀 SpaceX rolled out the next-generation V3 version of its Super Heavy booster (the bottom half of Starship) and lit up all 33 of its Raptor engines simultaneously in a static fire test in Texas on Thursday. It's the most powerful version of the booster ever built, and a critical milestone before the rocket's first V3 launch, which SpaceX is now targeting for mid-May. Starship V3 is what NASA is counting on to carry astronauts back to the moon under Artemis III. 33 engines and one shot. The moon is waiting.
🍏 Bloomberg reported Thursday that Apple's work on AirPods Pro with built-in cameras has reached the "advanced testing" stage, with prototypes featuring near-final designs. The cameras in each earbud aren't there to take selfies. Their job is to feed visual data to Siri so it can answer questions about whatever you're looking at: "What flower is that?" — "What's that book on the shelf?" There'll be an LED that lights up whenever the cameras upload anything. Launch is now expected alongside the long-delayed new Siri in iOS 27 this fall.
🎮 Mixtape, an indie game from Annapurna Interactive about three high-school best friends on the night before they graduate, came out Thursday on PS5, Xbox, PC, and Switch 2. Already, it's racking up some of the year's best reviews (94 on OpenCritic, 85 on Metacritic). The whole thing is set to a curated late-'80s and '90s soundtrack and unfolds as a series of playable music videos. It costs $20, runs about three hours, and it's free on Xbox Game Pass if you already subscribe.
If you'd told someone six months ago that Anthropic would soon be paying Elon Musk for computing power, they'd have laughed at you. Looks like Musk is laughing now.
Anthropic spent the last two years in the public crosshairs of Musk's xAI, the AI company Musk founded specifically to challenge it and OpenAI.
Musk personally went after Anthropic on X earlier this year, calling them "safety theater."
So it was a genuinely jarring moment on Wednesday when both companies announced that Anthropic would be taking over essentially all the computing capacity at Colossus 1, the enormous Memphis data center that xAI built to train its own Grok models.
The numbers are eye-popping even by AI-boom standards.
The deal gives Anthropic immediate access to more than 300 megawatts of new computing capacity (roughly enough power to run 250,000 homes) and the use of over 220,000 of Nvidia's most expensive GPUs, the same chips that have made Nvidia the most valuable company on Earth.
xAI is essentially handing Anthropic about half of its total GPU fleet.
Anthropic gets the breathing room it badly needed: Claude usage has grown 80x in the last year, and paying Pro and Max subscribers had been hitting frustrating rate limits that the company could not engineer its way out of.
As of Thursday, Anthropic doubled the daily limits for those subscribers and removed peak-hour throttling.
The other half of the announcement is, frankly, in science-fiction territory. Strap in.
Both companies said they're now exploring building gigawatts of orbital data centers. We’re talking actual computing hardware launched into space on SpaceX rockets, powered by the sun and connected back to Earth through Starlink.
The pitch is that data centers in space wouldn't compete for Earth's increasingly strained power grid or scarce groundwater for cooling.
Cathie Wood's ARK Invest argued this week that orbital data centers could eventually dwarf Starlink itself as a SpaceX business.
It is, to put it mildly, very early days.
The deal also has to be read against the backdrop of last week, when the Pentagon awarded classified-network AI contracts to seven companies (including SpaceX, OpenAI, Google, Microsoft, Amazon, and Nvidia) and pointedly excluded Anthropic, after the company refused to let the military use Claude for "all lawful purposes," including autonomous weapons and mass surveillance.
Eight days later, Anthropic is cutting a multi-billion-dollar deal with one of those seven companies to keep Claude running.
The AI race in 2026 is so absurdly compute-hungry that even the most public corporate feud in tech can be quietly set aside in less than 96 hours when both sides need each other.
What did you think of today's newsletter?