We spent 2 years calling Chinese AI models "cheap copycats." Last week something changed the space forever, and almost nobody in crypto is connecting it to DeAI yet. What happened? Kimi K3. Moonshot released Kimi K3 for anyone to download and run. It's the biggest open model ever released, and independent rankings put only three models on the planet ahead of it (all closed American models such as Claude and GPT). The difference is money. Chinese open models are running 60 to 90% cheaper than Claude and GPT. And developers have already switched. On OpenRouter (the platform devs use to access AI models), Chinese models now handle 3x more traffic than American ones. A year ago it was the reverse. How does this affect DeAI? Open models are the backbone of DeAI. Training new models in a decentralized way is still too hard, so most DeAI projects today are really doing one thing: running open models and selling access to them. So the bet looks like this: Better open models → more demand to run them → a bigger market for DeAI to compete for. But let's be real: better open models don't mean DeAI automatically wins. Devs can just buy access from big centralized companies instead. K3 shifts this in DeAI's favor though. The model is so massive that Moonshot recommends 64+ GPUs just to run it. Nobody's running this at home, which means everyone who wants K3 has to pay someone to run it for them. That's the market. And DeAI providers have an angle the centralized guys can't copy: private access with no signups and no KYC. @AskVenice is already offering K3 on exactly that pitch. There's also a catch in the fine print that most people haven't read. K3 isn't fully free to use. Any company selling access to K3 needs a separate deal with Moonshot once they make over $20m a year. And that $20m counts everything the company makes, not just what they make from K3. Small networks are fine today. The catch only matters once a provider gets big, and how decentralized networks handle that is an open question worth watching. Which brings me to the two areas I'm watching. These are two different trades. Inference layers (the platforms selling access to models): @AskVenice (already serving K3) @dphnAI @chutes_ai @OpenGradient GPU networks (the raw computing power): @AethirCloud @akashnet @ionet One nuance on the second group: a model this big needs serious data center hardware in one place. A network stitched together from random gaming PCs around the world can't run it. So not every GPU network benefits equally from this era. What I'm watching to confirm or kill the thesis: • Usage numbers on the inference platforms • GPU demand and pricing on the compute networks • Open model share on OpenRouter (already the majority, does it keep climbing) The thesis isn't that DeAI wins because open models got good. It's that the market DeAI competes for just got permanently bigger, and the networks are priced like nothing changed. I'm probably missing some good ones. Who else is on your radar?(Edgy - The DeFi Edge 🗡️)