Goldman Sachs dropped a note this morning. The headline: Chinese AI model companies are tightening open-source licenses. The case study: Moonshot AI's Kimi K3 now requires a commercial agreement for any MaaS provider pulling in over $20 million in annual revenue. The market yawned. BTC barely flinched. But if you’re holding AI-related crypto tokens — Bittensor, Render, Akash — you should be paying close attention. Because this isn’t just a licensing story. It’s a structural shift in how AI value flows. And it points directly to what I’ve been saying: centralised AI is hitting a wall, and decentralised infrastructure is the escape hatch.
Smile while the liquidity drains.
Let’s break down the hook. Goldman’s analyst Ronald Keung noted that after Kimi K2’s “attribution-only” license, K3 introduced a revenue threshold. Above $20M? You negotiate. Below? Free. On the surface, it’s a modest change. But this is the first time a leading Chinese AI model has explicitly said: “If you’re making money off our weights, we want a cut.” That’s a departure from the all-you-can-eat open-source buffet that powered the 2023-2024 boom. And it signals a broader trend: every Chinese model lab — Zhipu, Baichuan, DeepSeek — is watching. They’ll follow. Because they’re all burning cash on GPU clusters and need to show revenue to investors.
Context: why now?
I’ve been in this industry since the ICO sprint of 2017. I watched Ethereum-based DEX protocols get front-run by latency arbitrage. I saw DeFi summer’s yield narratives drown out code audits. And I remember the NFT mania where a Hollywood studio’s secret backing moved more volume than any smart contract upgrade. The through line? Humans drive markets, not code. The Kimi K3 license change is a human decision — a pricing signal from a company that realises its open-source strategy was leaking value to cloud giants. Alibaba, Tencent, ByteDance — they were deploying Kimi K2 on their own MaaS platforms, charging customers, and not sharing a dime with Moonshot. That’s unsustainable. The $20M threshold is a smart filter: it targets only the whales, leaving the long tail of developers free to experiment. But the message is clear: the open-source honeymoon is over.
Now the core: what this means for crypto AI.
Immediate impact on the AI token market: short-term confusion, mid-term bullish for decentralised compute and inference networks. Here’s my original analysis, built on 23 years of watching market mechanics.
First, let’s map the value chain. Centralised AI labs (Moonshot, OpenAI, Anthropic) produce frontier models. Cloud providers (AWS, Azure, Alibaba Cloud) host and serve them. Developers and enterprises build applications on top. The labs bear the R&D cost. The cloud providers capture the recurring API revenue. The license tightening is an attempt by labs to claw back margin. But the problem is structural: centralised clouds have the distribution and the customer relationships. They’ll negotiate hard. Moonshot may win some concessions, but the cloud giants will eventually pivot to their own models or to more compliant open-source alternatives. The real winner? Networks that don’t have a single rent-collecting intermediary.
Enter decentralised AI. Projects like Bittensor (TAO) create a subnet marketplace where multiple models compete, and stakers earn rewards for validating quality. Render (RNDR) and Akash (AKT) provide decentralised GPU compute, bypassing the cloud duopoly. The Kimi K3 license change accelerates the thesis: if using a centralised model increasingly comes with licensing friction and hidden costs, rational developers will look for alternatives. Not out of ideology — out of economics. A decentralised network that charges no per-inference tax and only requires holding a token to access compute is suddenly attractive.
The chart lies. The crowd feels.
Let’s get contrarian. The obvious read is that this is bad for open-source AI. But the contrarian angle is that it’s a massive validation for crypto-native AI infrastructure. Here’s why.
Most market participants assume that “open-source” means free forever. That’s naive. The Kimi K3 move proves that commercial open-source is a spectrum, not a binary. As soon as a model becomes good enough to generate real revenue, the lab will try to capture it. This creates a trust gap. Developers building on a centralised model face the risk of sudden license changes, price hikes, or even complete withdrawal. Smart developers will hedge by diversifying onto decentralised models that cannot unilaterally change the rules because they’re governed by token holders or code.
Moreover, the $20M threshold is a clear signal of traction. Moonshot only set that bar because they believe their model is worth that much. That means the AI arms race is real, and the winning models will be those that can sustain funding. Decentralised models, funded by token emissions and community staking, have a different burn structure — no venture capital pressure to exit in 5 years. They can afford to stay open and cheap for longer.
Unreported angle: the regulatory arbitrage.
Goldman didn’t mention this, but I will. Chinese AI labs operate under content compliance rules. The Kimi K3 license change also helps Moonshot track who is using their model, which is a prerequisite for regulatory audits. Decentralised AI, with its pseudonymous validators and permissionless inference, is harder to regulate. This could drive a wedge: centralised AI will be increasingly tied to jurisdictional compliance, while decentralised AI will remain a gritty, global, low-friction alternative. For crypto investors, that’s a long-term tailwind.
Takeaway: what to watch next.
Over the next six months, watch for two things. First, which Chinese labs follow Moonshot. If DeepSeek or Zhipu announce similar thresholds, the trend is confirmed. Second, watch the TVL and inference volume on Bittensor subnets. If they spike as developers seek alternatives, the market will price it in. I’m not giving investment advice, but I am saying: the Kimi K3 license change is a landmark moment. It’s the moment the centralised AI model realised it can’t be everything to everyone — and the moment decentralised AI started to look like the smartest seat at the table.
Smile while the liquidity drains. The smile is on the face of the decentralised network. The liquidity is draining from the centralised API resellers. The clock never blinks.