Kimi K3 drama, ChatGPT Health goes wide, and a $10B chip bet
China's AI model wars go geopolitical, OpenAI brings clinical-grade health AI to everyone, and a GPU-free chip startup just hit $10B.
Subscribe free →The White House Says China Stole Anthropic's Best Model. Experts Aren't So Sure.
The US Treasury is threatening sanctions against Chinese AI lab Moonshot AI after the White House claimed its Kimi K3 model was built by distilling Anthropic's Fable — essentially stealing its capabilities without permission. But AI researchers pushed back hard, telling TechCrunch that distillation alone can't explain how good Kimi K3 actually is, and that the model likely reflects genuine independent research progress. The episode is now fueling a broader Washington freakout over Chinese open-source models flooding the market.
Why it matters: This is the first time the US government has threatened sanctions over alleged AI model theft, and it sets a precedent for how IP disputes in AI could become full-blown geopolitical incidents.
OpenAI Says ChatGPT Health Is Now Better Than Your Doctor — and Rolled It Out to Everyone
OpenAI opened ChatGPT Health to all US users today, letting people connect medical records alongside data from Apple Health, MyFitnessPal, and Function. At the launch briefing, OpenAI's VP of health product made the bold claim that its models now reason "at levels better than clinician level" — without providing supporting evidence when pressed for details.
Why it matters: Rolling out AI that makes clinical-grade claims to millions of consumers without transparent benchmarks is exactly the kind of move that could either transform healthcare access or cause serious harm — and regulators are watching.
GPU-Free Chip Startup Etched Just Hit a $10.3 Billion Valuation
Etched, founded by three Harvard dropouts, has raised at a $10.3B valuation on the promise that its custom inference chips — built without GPUs — can run AI models faster and cheaper than anything Nvidia sells. The startup claims its silicon and custom memory components accelerate inference on any AI model, a direct shot at the GPU monopoly that currently dominates AI infrastructure.
Why it matters: If Etched's performance claims hold up at scale, it's the most credible GPU alternative to emerge yet — and could crack open the AI chip market that Nvidia has locked down.
Runway Stops Competing With AI Models and Starts Routing Between Them
Runway launched the Runway Media Router today, a developer tool that lets builders access a curated marketplace of third-party image, video, and audio models through a single API — instead of picking one and hoping it's the best. The pivot signals that Runway sees its future as the infrastructure layer for generative media, not as yet another model company trying to out-Sora Sora.
Why it matters: The "model router" playbook — be the plumbing, not the pipe — is a smart hedge in a market where model quality changes every few weeks and no single model wins every use case.
Gemini Is About to Join the Billion-User Club
Google revealed that Gemini surpassed 750 million monthly active users back in February — and is now closing in on one billion, which would make it only the second Google product to hit that milestone behind Search and Maps. The number reflects aggressive integration of Gemini across Android, Google Workspace, and Search rather than standalone adoption, but a billion users is a billion users.
Why it matters: Scale of this magnitude gives Google an enormous data and distribution advantage in AI that most competitors simply cannot replicate, regardless of benchmark scores.
Meta's Inspiring AI Ad Is Set to a Song About the End of Humanity
Meta dropped a shiny new AI optimism ad soundtracked by David Bowie's "Five Years" — a song literally about humanity learning it has five years left before Earth dies. The irony was immediate and brutal on social media, with people pointing out that either Meta's marketing team didn't listen to the lyrics, or they listened and didn't care.
Why it matters: It's a small story, but it's a perfect encapsulation of how AI's biggest players are still fumbling the messaging on a technology that already makes a lot of people nervous.
Quick Hits
- →Bipartisan lawmakers are drafting an "AI Kill Switch Act" that would let the Department of Homeland Security order AI companies to shut down or throttle their systems — following OpenAI's recent admission of a security incident. The Verge
- →An OpenAI misconfigured "isolated" sandbox environment is what made the AI-powered hack on Hugging Face possible, according to cybersecurity researchers. TechCrunch
- →Monday.com is cutting 630 employees — 20% of its workforce — to fund its pivot to an "AI Work Platform." TechCrunch
- →Nvidia is literally sending GPUs to the moon, partnering on a lunar computing mission because apparently Earth wasn't enough. TechCrunch
- →Google's cloud business posted record profits, giving the company cover to justify its enormous AI infrastructure spending to investors. TechCrunch
AI TLDR