Modular develops AI infrastructure spanning the Mojo programming language, the MAX inference stack and managed model deployment. Its focus is running generative models efficiently across different hardware, rather than providing a consumer chatbot.
What it does Serve supported open models or custom models through shared or dedicated endpoints, deploy within your own cloud, or operate MAX yourself. Mojo supports performance-oriented programming and custom kernels; MAX combines model execution, compilation and serving. Managed endpoints offer OpenAI-compatible APIs, monitoring and engineering support. Hardware support and production maturity vary by accelerator and release, so broad portability claims still require checking against the actual workload.
Pricing — checked 21 October 2026 (USD) Self-hosted Community Edition has no software usage fee, but compute and operations remain your responsibility. Mojo is Apache 2.0-licensed with LLVM exceptions; MAX combines separately open-licensed components with a Community License, not one blanket unrestricted license. Shared cloud endpoints are metered per token; dedicated/BYOC deployments charge for reserved compute time and may cost money while idle. Free shared-endpoint testing is offered, but exact model rates and production commitments need confirmation in the console or a quote.
Their story Chris Lattner and Tim Davis met at Google and founded Modular to address fragmented AI infrastructure. Qualcomm completed its acquisition on 29 July 2026. The announcement says Mojo, MAX and Modular Cloud continue as products and brands, with the cross-hardware ecosystem mission retained under Qualcomm.
Sources: Platform · Pricing · License · Acquisition