Microsoft is said to be negotiating with Chinese artificial intelligence firm Moonshot AI on a revenue-sharing deal to have its Kimi K3 model run on Microsoft Azure. The negotiations occur at a delicate moment for the worldwide AI business, as China’s technology companies expand overseas, and the U.S. government has tightened limitations on such highly sophisticated technologies and semiconductors for China.
The Beijing-based startup Moonshot AI, which created the Kimi AI platform, is looking to share in the revenue from Kimi K3 services offered by the top U.S. cloud platforms, people familiar with the talks say. Amazon Web Services, Google Cloud and Microsoft are among the companies attempting to talk the deal. The talks are on a preliminary level and it remains uncertain whether they will lead to formal deals.
Such a deal would have massive ramifications, potentially forming a crucial commercial connection between the biggest Chinese AI developer and the largest American cloud firm. It’s possible that, if Microsoft and Moonshot agree to the deal, Azure might run Kimi K3 and offer infrastructure to customers that run the model. This would also illustrate the growing global and regulatory footprint of the demand for more advanced AI models.
Moonshot AI claims that it is aiming to earn up to 30% of its revenues from Kimi K3 services on Azure, and other cloud platforms. The share is believed to be similar to the terms that the Chinese startup has discussed with its key customers who are contemplating using its open-weight AI model. The revenue-sharing models can differ according to the infrastructure expenses, customer demand, licensing conditions, and how AI usage is tracked.
The talks also underscore the rising significance of cloud computing platforms in the AI company business. Building a strong AI model is merely part of the business equation. Large-scale computing infrastructure is also needed to train, deploy and serve those models to customers. Cloud providers like Microsoft Azure provide the data centres, computing power, and technical frameworks to enable AI systems to be deployed at scale.
Moonshot is a deal with Microsoft that would give Kimi K3 access to a large, global cloud ecosystem, and perhaps even make it more available to foreign gamers. While hosting an advanced Chinese AI model could present a new business opportunity for Microsoft, it might also spark concerns regarding regulatory compliance, data management, and the overall geopolitical landscape of AI.
Kimi K3 has been drawing interest due to its alleged features and affordability, which is key for gamers.Kimi K3 has been gaining popularity because of its claimed features and price range, particularly for gamers. Chinese AI developers have been increasingly targeting the development of models capable of competing with top western AI systems, with relatively low development and usage costs. This has led to increased competition in the global AI landscape, where companies are competing not just based on model performance, but also affordability, availability, and flexibility for developers.
The business talks are occurring amid a complex technology relationship between the United States and China. Washington has placed new limits on the sale of advanced artificial intelligence chips and technology to China based on national security concerns. This limitation has fueled a debate around the availability of high-end computing hardware to Chinese AI firms as they work to create ever increasingly powerful models.
Moonshot AI has also been criticized by U.S. officials. The US Treasury Secretary Scott Bessent earlier said the company may be placed on a list of companies and people barred from conducting business in the US. American authorities have also claimed that Moonshot has been exploiting technology linked to Anthropic’s higher-end AI model and that it has stolen Nvidia chips. Kimi K3 is accused of achieving its performance through model distillation and Moonshot has denied any part in the process.
The company has stated that refurbishment is the result of changes made to the company’s original architecture in Kimi K3. The feud is part of the larger debate on the future of the development of AI models, as governments and tech firms try to figure out how to regulate the process of building IP, computing power, and model outcomes.
In spite of those worries, however, Microsoft and Moonshot talks indicate that business interests are still significant in the quickly advancing field of AI. Cloud companies are in a global market where customers are increasingly demanding access to a multitude of AI models, not just one. By supporting different models, cloud platforms can be able to attract developers and businesses with different sets of requirements, whether it is for pricing, deployment or the availability of specific capabilities.
Meanwhile, any deal between a Chinese artificial intelligence firm and a top American tech firm would likely have to solve a number of outstanding problems. One key issue is revenue sharing, or how it would be split between the companies. Financial models would have to take into consideration the expenses of cloud infrastructure, customer acquisition, licensing and continuous technical support for models.
Another important question is how to get to the data. The amount of data AI models can handle in the cloud varies based on customer usage. Data handling, security and access security would therefore have to be clearly defined and incorporated into any commercial agreement. The picture is made even trickier when companies are based in different national regulatory jurisdictions.
Another challenge yet to be addressed is the measurement of AI usage. A lot of AI services sell their services based on the number of tokens a model processes. Tokens are units used to represent text and other information processed by an AI system. It is therefore critical to be able to accurately measure the use of the token when calculating usage based revenue and who will be entitled to what.
The proposed setup also mirrors the evolving economics of AI. Nowadays, AI firms are more interested than ever in moving beyond building models and considering how to make the models into a viable business. Revenue-sharing models can offer a direct monetary benefit to model developers and help cloud service providers make their AI service offerings more comprehensive.
