Cognition hits $40B, Nvidia bets $500B, and Claude users caught cheating
AI valuation madness, Nvidia's GPU insurance scheme, watermark backlash, and Amazon's opt-out data grab — a lot happened today.
Subscribe free →Cognition Is Already Back for More Money — $40B This Time
The AI coding startup raised $1B at a $26B valuation just a few months ago, and is reportedly already in talks to raise again at a $40B valuation. That's a 54% jump in valuation in under a year, for a company most people outside AI circles still can't describe in a sentence.
Why it matters: Either AI coding agents are genuinely worth this much, or we're deep in a bubble that makes 2021 look measured.
Nvidia's $500B Plan Is Basically Insurance for Its Own GPUs
Nvidia wants to keep financiers lending for AI infrastructure buildouts — even as GPUs age and depreciate. The plan would essentially create a financial ecosystem that preserves the value of existing hardware by keeping demand artificially high through structured lending.
Why it matters: If it works, Nvidia doesn't just sell chips — it controls the economic conditions that make those chips worth buying.
Claude Now Watermarks Its Output, and Cheaters Are Very Upset
Anthropic rolled out watermarking on Claude-generated content, and some users hit social media to complain — because it will expose them using AI at work or in school without disclosure. The backlash is less about privacy and more about losing plausible deniability.
Why it matters: Anthropic is drawing a line between legitimate AI use and covert AI use, and the anger proves the latter is way more common than anyone admits.
Amazon Is Training on Twitch Streamers' Content — Unless They Say No
Amazon quietly made AI training on Twitch content opt-out by default. When the Twitch CPO was asked why not opt-in, he answered honestly: "If this was opt-in, nobody would opt in. That's honestly the answer."
Why it matters: That candid admission is a perfect encapsulation of how the entire AI training data industry operates — extract first, apologize if caught.
Microsoft Kills Copilot Features Nobody Was Using
Microsoft is merging its consumer and business Copilot apps into one and axing AI-generated podcasts, Group Chats, Deep Research, and its "Mico" character. The product consolidation is essentially a quiet admission that the original Copilot launch threw too many things at the wall.
Why it matters: Microsoft's Copilot strategy is maturing from "ship everything" to "ship what works" — a necessary correction, but it raises questions about what's actually driving enterprise adoption.
Hinton, Li, and Ng Walk Into a Conference and Agree on Open Source
At the Ai4 conference, three of the most respected names in AI — Geoffrey Hinton, Fei-Fei Li, and Andrew Ng — made a joint case for keeping AI open, even as safety concerns grow. The rare alignment between a safety hawk (Hinton) and an open-source advocate (Ng) is notable.
Why it matters: When the godfather of deep learning and the loudest open-source advocate in AI are on the same page, the "open vs. closed" debate may be shifting more than the headlines suggest.
Quick Hits
- →OpenAI-backed **Thrive Holdings** raised $2B at a $12B valuation from SoftBank and others to push AI into the enterprise. TechCrunch
- →**Apple** is reportedly considering a nine-figure deal to pay publishers for real-time news to power Siri. TechCrunch
- →**Suno** launched Studio 2.0 with MIDI support, pushing its AI music tool closer to a real DAW. The Verge
- →A **supply-chain attack** on a compromised AI package leaked terabytes of credentials from 2,500 users. Ars Technica
- →**Meta** is testing on-device AI scam detection for WhatsApp messages in a limited beta rollout. The Verge
AI TLDR