TL;DR — Key Takeaways
- Chinese AI models are moving into Western enterprise channels. Moonshot AI is reportedly negotiating with Microsoft, AWS and Google Cloud to host and sell access to its Kimi K3 model.
- Cloud distribution could remove a major barrier to adoption. Managed inference, billing, compliance and infrastructure would make Chinese open-weight models easier for enterprises to deploy at scale.
- Low-cost models are becoming commercially significant. Chinese providers are looking for revenue-sharing arrangements that let them monetize open-weight models even when the weights themselves can be freely downloaded.
The Chinese AI models are gaining popularity with more than just people with small budgets.
As we know, China’s low-priced, open-weight AI models have more fans than American models as their quality approaches the top frontier-level models. We’ve been thinking that most of this rise in popularity came from individual developers and companies with tight budgets. We were wrong. Now, Chinese AI models are gaining a more direct path into Western corporate systems. Moonshot AI is in talks with Microsoft, Amazon and Google over arrangements to host and sell access to its Kimi K3 model through their cloud platforms.
Specifically, Moonshot is seeking revenue-sharing deals with Microsoft, Amazon Web Services and Google Cloud that would enable the three hypercloud giants to host Kimi K3. The Beijing-based startup is reportedly asking for as much as a 30% revenue share. That’s serious money even by AI standards.
If completed, an agreement could give Kimi K3 access to the sales, compliance, billing, and infrastructure machinery enterprises demand before using a model in production. That matters because even open-weight models can be prohibitively difficult and expensive to operate at scale. Besides, most companies using AI, for now anyway, would rather pay for an AI service than take a DIY approach.
For example, while you can certainly download and run Kimi K3 yourself, how many of you can self-host a Large Language Model (LLM) with 2.8 trillion parameters? That’s enough to give even the most hard-nosed self-hoster pause. Major cloud marketplaces and managed inference services could become the crucial bridge between Chinese AI companies and corporate buyers.
Chinese models have built momentum this year by combining increasingly competitive performance with much lower prices than top proprietary U.S. systems. That calculus is especially attractive in “agentic” workflows, where a strong, lower-cost model can dramatically reduce operating expenses even with a cloud bill attached.
It’s not just Moonshot. Reuters also reported that Alibaba is seeking revenue-sharing arrangements with major users of its open-source models. Chinese AI companies are seeking ways to retain commercial value from models whose weights can be freely downloaded or modified. Hey, it worked for open-source Software-as-a-Service; it can work for AI.
In the meantime, NVIDIA is expanding technical support for leading Chinese open models even as it warns that U.S. restrictions could limit that work. That hasn’t stopped them, though.
This week, the chipmaker powerhouse highlighted a “local AI initiative” that includes optimizations for DeepSeek’s V4 Flash and Alibaba’s Qwen 3.8, as well as systems from Google and NVIDIA. It also announced “day-zero support” for Qwen 3.8-27B on its RTX GPU systems and software updates intended to make clusters of DGX Spark systems easier to use for models such as Z.ai’s GLM 5.2 and DeepSeek V4 Flash.
NVIDIA’s position is straightforward. Model makers can compete in many arenas, but so long as their AI systems run best on NVIDIA hardware, the company will be happy. NVIDIA wants to be the Intel that dominated the PC market in the days of WinTel.
As Forrester analyst Charlie Dai told CNBC, NVIDIA’s moves reflected the growing influence of Chinese models in the global AI ecosystem and reinforced Nvidia’s goal of remaining the preferred infrastructure layer regardless of where a leading model originates.
Washington, with its anti-Chinese AI stance, can’t be happy about this. Additionally, the Trump administration is considering measures that could restrict NVIDIA’s ability to support Chinese-based third-party applications and models. The company warned that restrictions affecting models such as DeepSeek, Qwen, or Kimi could materially affect its business.
Of course, Moonshot itself has come under government scrutiny. Treasury Secretary Scott Bessent has said he might add the company to a trade blacklist. U.S. officials have also accused Moonshot of using distillation from Anthropic’s Fable model and obtaining NVIDIA chips illegally.
The debate is complicated by the nature of open-weight AI. Unlike API-only systems, anyone can tune and deploy DeepSeek, Qwen and Kimi, and not just their China-based developers. Microsoft, Meta, Palantir, NVIDIA and more than 20 other companies urged policymakers in July not to impose “premature restrictions” on open-weight models,
For now, the Moonshot negotiations remain unresolved. But their existence suggests Chinese model makers are moving beyond open-source visibility and low-cost API competition toward the more consequential contest for enterprise distribution, recurring revenue and control of the global AI platform layer. Pay attention, Anthropic, OpenAI, Perplexity, and the rest of the U.S.-based AI crew: Chinese competition is getting real.
