China's $2B Open-Source Bet Is Working

Ninety percent of the performance. One-tenth of the cost. That math just raised $2 billion.

Moonshot AI, the Beijing-based lab behind the Kimi series of large language models, has closed a funding round of approximately $2 billion at a valuation of $20 billion, according to Huafeng Capital, which advised investors in the deal. The round was led by Meituan's venture arm, Long-Z Investment — yes, the food delivery giant is now backing frontier AI. That detail alone says something about where Chinese capital is flowing.

For context: Moonshot was valued at $4.3 billion at the end of 2025. By early 2026, it had hit $10 billion after a $700 million raise. Now, five months later, it's at $20 billion. The company has raised $3.9 billion in the past six months alone.

What Kimi Actually Is — And Why It Matters

Moonshot AI was founded in 2023 by Yang Zhilin, a researcher who previously worked at Meta AI and Google Brain. The company builds open-weight models — meaning the model weights are publicly released, allowing anyone to run, fine-tune, or deploy them without paying per API call.

Earlier this year, Kimi K2.5 made waves in the developer community by posting benchmark scores close to OpenAI and Anthropic's flagship models at the time — at a fraction of the inference cost. The latest iteration, Kimi K2.6, is currently the second-most used LLM on OpenRouter, a platform that aggregates model usage across providers. That's not a marketing claim. That's a usage signal from actual developers making real choices.

Revenue is following. The company's annualized recurring revenue crossed $200 million in April 2025, driven by paid subscriptions and API usage — a figure that gives the $20 billion valuation a concrete, if still aggressive, anchor.

Sponsored

Advertise with Us

[email protected]

Sponsored

The Bigger Pattern: China's Open-Source Surge

Moonshot isn't an outlier. It's a data point in a broader shift.

DeepSeek, arguably China's most recognized AI lab globally, is reportedly in talks to raise outside capital for the first time at a valuation of approximately $45 billion. Meanwhile, two other Chinese AI companies have gone public in Hong Kong: Zhipu AI (trading as Knowledge Atlas Technology) closed Thursday with a market cap of HK$434.7 billion (~$55.9 billion), while MiniMax ended the day at HK$257.3 billion (~$33 billion) — both rallying on new model releases.

The pattern is consistent: Chinese AI labs, constrained by U.S. export controls on advanced semiconductors, have leaned into efficiency. They build competitive models with less compute, then release the weights openly to capture global developer mindshare. It's a strategy that sidesteps the hardware bottleneck and turns a geopolitical disadvantage into a distribution advantage.

Three Ways to Read This

For enterprise buyers and developers, the calculus is straightforward: if an open-weight model delivers 90% of the performance at 10% of the cost, the procurement decision writes itself — at least for cost-sensitive workloads. Kimi's OpenRouter ranking suggests this is already happening at scale, not just in theory.

For Western AI incumbents — OpenAI, Anthropic, Google — this is a slow-moving margin problem. Their moat has always been model quality. As that gap narrows, the premium they charge for API access faces structural pressure. The response so far has been to accelerate model releases and expand into enterprise services, but neither fully addresses the cost differential that open-weight models exploit.

For policymakers and regulators, the picture is more complicated. U.S. export controls were designed to limit China's access to the compute needed to train frontier models. But once model weights are released publicly, they propagate across global servers regardless of origin. The regulatory lever that works on hardware doesn't translate cleanly to software weights distributed over the internet. Moonshot's backers — which include Alibaba, Tencent, HongShan (formerly Sequoia China), and IDG Capital — are sophisticated enough to understand this asymmetry.