Modular alliance program

Latest models on the best hardware.

Today, frontier AI workloads are too often limited to one vendor’s silicon. Enabling a full AI stack is costly and repetitive, and AI developers suffer from a fragmented software ecosystem because of a lack of standards at the bottom of the stack.

The Modular alliance program is building “LLVM, but for AI”: making it easier than ever before to enable a competitive software stack for every type of compute. It includes a conformance testing ecosystem to ensure a compatible ecosystem that unlocks choice for all developers.

AI developersFrameworkServingRuntimeKernelsHardwareGPUCPUTPUASICNPUTODAY: 5 full stacks, built 5 timesTOMORROW w/ HCA: 1 unified layer, built once

Every hardware vendor builds the full AI stack for its own chips. Some of it’s better, some worse, and none of it’s compatible with the rest.

LLVM, but for AI: one unified software layer, built once in common, with every kind of compute underneath and a compatible ecosystem on top.

The platform

A next-generation AI platform, now open.

Modular has been building a next-generation AI platform, and now it’s open. It includes the Mojo language and the MAX AI framework, which make it easier than ever before to enable new hardware and unlock the power of the chips you already have.

Modular recently joined Qualcomm, which opened up the Modular stack and wants to catalyze even more innovation across the entire domain, enabling everyone to build better products.

Accelerated computing language
Serving & modeling framework
MAXServing & modeling framework
MojoAccelerated computing language
The idea

LLVM, but for AI.

LLVM turned monolithic compiler stacks with little reusability into a modular and reusable ecosystem of technologies that the whole industry builds on. AI has nothing like it: every hardware vendor has to implement the full AI stack for its own chips.

The leading companies provide good software, some better and some worse, but none of it’s compatible with the rest, so work done in one vendor’s stack doesn’t extend to the others. The Modular alliance program applies the LLVM model to AI compute and adds conformance testing: a unified software layer, built in common, strict in the core, neutral at both ends, and compatible across every kind of compute.

A shared, reusable ecosystem like LLVM, not a standards body
Neutral by design, open to everyone, tiered by contribution
Permissive licensing to grow the ecosystem
The flywheel

Contributions that compound.

This is an industry-wide challenge, and no single company can crack it alone. We want AI labs, silicon partners, deployers and developers to build one common platform together. Every turn of the wheel enables the AI ecosystem as a whole to move even faster, particularly for innovators at the frontier.

EarlycheckpointsPlugins andextensionsa unified stackfor a heterogeneous fleet,day-zero modelsSOTA performancewith best economicsDeveloperEcosystemSilicon PartnersAI LabsMAXframeworkDeployers

Silicon partners

Bringshardware plugins & extensions

Getsfaster time to market

Deployers

Bringscapacity for development & CI

Getsmore choice, better ROI

AI labs

Bringsvisibility: early checkpoints

Getsbroader distribution

Developers

Bringsvalue-add services & apps

Getsan open platform

MAX framework

Bringsnext generation AI platform

Getsecosystem contributions

Learn more

Let’s build it together.

Any model on every hardware platform.

Learn more about Modular alliance program

How would you like to contribute?

Our team reads each request and will reach out with next steps. Your data is handled in accordance with Modular's Privacy Policy.

Thank you. Your request is in, and a member of our team will be in touch.
Oops! Something went wrong while submitting the form.