Moonshot AI, the company behind Kimi, just closed a $700 million funding round led by Alibaba and Tencent. Valuation: $12 billion. That makes them a decacorn in under two years.
Then they open-sourced their best model, Kimi K2.5.
Let that sink in. They raised three-quarters of a billion dollars to build technology they're giving away for free. And somehow, investors are fighting to get in.
Either this is genius or we're watching the most expensive act of charity in AI history.
The Open-Source Paradox
Kimi K2.5 isn't a toy model. It's a multimodal system with a 400M-parameter vision encoder (MoonViT) that processes images and video. It can deploy 100 AI agents simultaneously. It can replicate entire website user journeys from screenshots alone.
And anyone can download it. For free.
The traditional business logic is straightforward: you build something valuable, you charge for it. But Moonshot is playing a different game entirely. By open-sourcing K2.5, they're betting that ecosystem dominance matters more than API revenue.
Meta did this with Llama. It worked — Llama became the default foundation for thousands of startups. But Meta has a $1.5 trillion market cap and ad revenue to subsidize the strategy. Moonshot has... a blog post about revenue growth and no disclosed ARR.
Why Investors Don't Care (Yet)
The funding round tells you everything about where AI money is flowing in 2026. Alibaba and Tencent aren't investing in Moonshot's current revenue. They're investing in:
- Ecosystem lock-in — If Kimi becomes the default model for Chinese AI development, Moonshot controls the platform layer
- Data flywheel — Every developer using K2.5 generates usage patterns that improve future models
- Strategic positioning — Both Alibaba and Tencent need a horse in the AI race that isn't controlled by the other
But here's the tension: open-sourcing commoditizes the very technology you're trying to monetize. If anyone can run K2.5 locally, why would they pay for Moonshot's API?
The Bigger Pattern
This isn't just a Moonshot story. It's the defining tension of AI in 2026.
DeepSeek open-sourced their reasoning models and triggered a market panic in January. Mistral open-sources everything and keeps raising at higher valuations. Meanwhile, OpenAI — the company that literally has "open" in its name — went closed-source and is printing money.
The market is simultaneously rewarding openness (Moonshot's $12B valuation) and closedness (OpenAI's $300B+ valuation). Something has to give.
For developers, this is actually great news. The cost of running state-of-the-art AI has collapsed. You can self-host models that would have cost $100K/year in API fees just 18 months ago. If you're building AI-powered products, the infrastructure cost is approaching zero.
I've been running inference workloads on Vultr GPU instances — their bare metal options start at competitive prices and you get dedicated hardware instead of shared cloud nonsense. For batch processing and fine-tuning, the economics are hard to beat.
And for the content side of AI products, tools like ElevenLabs have made voice synthesis absurdly accessible. Their free tier gives you 10,000 characters/month — enough to prototype voice features without spending a dime. Combined with open-source LLMs, you can build a full AI product stack for under $50/month.
The Question Nobody's Asking
If the best AI models are free, what exactly are we paying for?
The answer is shifting from "the model" to "the workflow." Raw model access is becoming a commodity. The value is in how you orchestrate models, integrate them into existing systems, and build reliable pipelines around them.
Moonshot's $12B bet is that they'll figure out the monetization later. History says that's a coin flip at best. But in the current AI gold rush, being the one selling shovels — even free shovels — might be enough to win.
Just don't ask the investors to explain the math.
Is open-sourcing your core product brilliant strategy or slow-motion suicide? I'm genuinely curious what this community thinks.
Top comments (0)