Azure is expanding its support for open-source ecosystems by integrating Fireworks AI into Microsoft Foundry. This move allows developers to access high-performance inference—the process where a trained AI generates responses—directly through a unified Azure endpoint. By removing the need to build custom infrastructure (bespoke serving stacks), the platform enables organizations to scale open models with the security and compliance expected of enterprise software.

The integration focuses heavily on speed, leveraging an engine capable of processing 13 trillion tokens daily. For developers utilizing models like DeepSeek V3.2 or the new MiniMax M2.5, this translates to reduced lag and higher throughput. This level of performance is essential for agentic AI, where systems must perform rapid, multi-step reasoning to solve complex problems without constant human intervention.

A standout feature of this launch is Bring-Your-Own-Weights (BYOW), which lets teams upload models they have specialized or compressed elsewhere. These custom models can then be managed through a single control plane—a centralized system for monitoring and deployment. Whether choosing serverless pay-as-you-go pricing or dedicated capacity for heavy workloads, the integration provides a flexible foundation for the entire AI lifecycle.

Azure is expanding its support for open-source ecosystems by integrating Fireworks AI into Microsoft Foundry. This move allows developers to access high-performance inference—the process where a trained AI generates responses—directly through a unified Azure endpoint. By removing the need to build custom infrastructure (bespoke serving stacks), the platform enables organizations to scale open models with the security and compliance expected of enterprise software.

The integration focuses heavily on speed, leveraging an engine capable of processing 13 trillion tokens daily. For developers utilizing models like DeepSeek V3.2 or the new MiniMax M2.5, this translates to reduced lag and higher throughput. This level of performance is essential for agentic AI, where systems must perform rapid, multi-step reasoning to solve complex problems without constant human intervention.

A standout feature of this launch is Bring-Your-Own-Weights (BYOW), which lets teams upload models they have specialized or compressed elsewhere. These custom models can then be managed through a single control plane—a centralized system for monitoring and deployment. Whether choosing serverless pay-as-you-go pricing or dedicated capacity for heavy workloads, the integration provides a flexible foundation for the entire AI lifecycle.