Modular MAX
Unified AI compute layer: one stack to serve models across GPUs, TPUs, and edge accelerators.
Acquired — still operating
acquired by Qualcomm mid-2026
About
MAX is Modular's unified AI compute and inference platform, built around the Mojo language and the MLIR-based MAX Engine. It targets heterogeneous hardware — NVIDIA GPUs, AMD GPUs, Apple Silicon, Google TPUs, AWS Trainium, and Qualcomm Cloud AI100/Dragonfly accelerators — from the same modeling APIs, with industry-leading throughput claims on Blackwell B200 and AMD MI355X.
2025-2026 was a landmark stretch: Modular raised $250M at a $1.6B valuation, open-sourced the MAX kernels (2025), acquired BentoML (Feb 2026), brought Modular Cloud (hosted inference, OpenAI-compatible endpoints, GA), and was acquired by Qualcomm mid-2026. At ModCon 2026, MAX removed device usage restrictions, becoming source-available under the Modular Community License with an open alliance program for hardware and cloud partners.
What it can do
- Serve models across NVIDIA, AMD, Apple, TPU, Trainium, and Qualcomm accelerators
- MAX Engine graph-level optimization with Mojo kernels
- Modular Cloud: OpenAI-compatible shared and dedicated inference endpoints
- PyTorch-like MAX Python API with model.compile() for production
Setup
Install the Modular platform via pip; Modular Cloud offers hosted endpoints with signup.
Prerequisites
Target accelerator hardware or a Modular Cloud account; Python familiarity.
Use cases
Works with
Trust score — 8/10
Scored 8/10 (verified tier, metrics-v1). Evidence: Verified via vendor site/docs/GitHub and reputable coverage, wave 73; Trending score 84/100 from researcher assessment.
Sources
Every fact above was checked against these sources. Where a claim comes from the provider and couldn't be independently verified, the text says so.