Microsoft could use Kimi K3 to cut Copilot AI costs by 60%

Cheaper AI without sacrificing quality is the dream. Microsoft may be closer than expected. According to reports, Satya Nadella's firm is testing Moonshot AI's Kimi K3 for use within Copilot and Azure AI.

The move could give customers access to a broader range of AI models while significantly lowering the cost of serving everyday AI requests. The shift could reduce Microsoft's AI inference costs by as much as $600 million, according to The Information.

Why Kimi K3 is on Microsoft's radar

As AI adoption grows, companies are increasingly looking beyond a single large language model. Instead, they are choosing the most suitable model for different tasks based on cost, speed and capability.

Kimi K3 is reportedly being evaluated as a cost-effective option for handling Copilot workloads such as document summarisation, coding assistance, content generation and general productivity tasks, features currently powered by models from OpenAI and Anthropic. Many of these requests do not require the most expensive frontier AI models, making a lower-cost alternative attractive.

For Microsoft, expanding its model portfolio could also give Azure customers more flexibility while reducing operating expenses across Copilot services.

Where the savings could come from

The reported savings relate to inference, the process of generating responses after an AI model has been trained.

According to reports, Kimi K3 could lower these costs through more competitive per-token pricing, efficient deployment for specific workloads and greater flexibility in scaling AI services.

While the savings for a single request may appear modest, they could translate into significant reductions when applied across millions of daily Copilot interactions. Lower operational costs may also allow Microsoft to offer more competitive AI pricing to enterprise customers using Azure.

What Kimi K3 brings to the table

Kimi K3 is designed for knowledge work, software development and reasoning-intensive tasks. Unlike many closed AI systems, it is an open-weight model, with Moonshot planning to release the full weights later this month, giving organisations greater flexibility over how and where it is deployed.

That approach can reduce dependence on a single AI provider, improve deployment options and allow businesses to optimise infrastructure based on performance, compliance and cost requirements.

Enterprise adoption still faces challenges

Any integration involving a Chinese-developed AI model is likely to receive close attention from regulators and enterprise customers, particularly around data security and governance.

Businesses will want assurances on data privacy, model transparency, reliability and long-term support before deploying such models in production environments.

Compliance requirements and regional regulations could also influence adoption.

What happens next?

The reported pilots are expected to evaluate whether Kimi K3 can deliver the right balance of performance, speed and cost for typical Copilot workloads. Microsoft has not publicly confirmed the savings estimate or which features could move to the new model.

Microsoft may ultimately adopt a model-routing strategy, directing simpler requests to lower-cost models like Kimi K3 while reserving its most powerful AI systems for complex reasoning tasks. If the testing proves successful, the company could significantly reduce Copilot's operating costs while giving Azure customers access to a more competitive and diverse AI ecosystem.