China's military AI, space data centers, and Anthropic's $350B flex

Nvidia helped DeepSeek optimize their models in 2024 with "co-designed algorithms" and training tricks. Those models later ended up helping China's military. There were zero public indications of military use when Nvidia was helping. Sure Jan.
Why it matters: Export controls are about as effective as a "Please Don't Copy" sticker on a USB drive. The U.S. just played itself by teaching China how to train models better. Not great!! (source)

Claude's parents are DOUBLING their fundraising target after demand hit 6x expectations. New valuation?? $350 BILLION. That's more than Disney, Nike, and McDonald's combined btw.
Why it matters: VCs are throwing money at AI like it's 2021 crypto all over again. Nobody knows if this stuff will make money but hey - line go up 🚀 (source)

China's space contractor CASC plans gigawatt-class orbital AI infrastructure with solar panels that generate 5x more power than Earth-based ones. Because apparently regular data centers aren't sci-fi enough anymore??
Why it matters: The AI race just went LITERALLY INTERSTELLAR. We're solving energy constraints by... checks notes... launching GPUs into orbit. This timeline is unhinged. (source)

ServiceNow made Claude their "preferred AI model" exactly ONE WEEK after announcing the same thing with OpenAI. The audacity I-
Why it matters: Enterprises are done picking sides in the AI wars. They want ALL the models and they want them NOW. Model monogamy is dead 💀 (source)

Google Chrome now has Auto Browse that controls your browser autonomously (!!!) and Gemini living permanently in your sidebar like that friend who won't leave your apartment.
Why it matters: Google's using its browser monopoly to steamroll AI startups. OpenAI's browser never stood a chance lmaooo (source)

Google released AlphaGenome which predicts how DNA typos cause disease across 11 biological processes. Full weights included. For free. Wild.
Why it matters: After AlphaFold's Nobel Prize, Google's on a "let's cure diseases" speedrun. We love to see it :) (source)

Arcee AI trained Trinity - one of the largest U.S. open source models EVER - with just $20M and 6 months. Meta's Llama team in shambles rn.
Why it matters: Proof that you don't need $100B to compete. Just cracked engineers and pure spite apparently?? (source)

Apple Creator Studio Pro ($12.99/month) adds AI that assists creation instead of doing it for you. Features include transcript search and chord extraction. How... reasonable??
Why it matters: Apple's taking the "AI as copilot" approach while everyone else speedruns toward replacing humans. Different strategy or just late to the party? Yes. (source)

Meta will charge $0.0691 per AI message in Italy after regulators forced them to allow third-party chatbots. The pettiness!!!
Why it matters: This is what happens when you force monopolies to play nice. They'll open up but make you PAY for every byte 💸 (source)

Deezer's tool catches AI tracks from Suno and Udio with 99.8% accuracy. They're getting 60,000 AI tracks DAILY and 85% are fraud. Yikes.
Why it matters: The AI slop apocalypse hit music streaming and now we need AI to detect AI. It's AIs all the way down... (source)

Flapping Airplanes raised $180M at $1.5B to train AI without "ingesting half the internet" (shade detected)
Ex-OpenAI researcher wants $1 BILLION for continuous learning AI
Google's Gemini 3 can now zoom and annotate images like it's CSI: Miami
Why it matters: Scaling might be hitting a wall so everyone's throwing billions at literally any other approach. The cope is real. (source)