TL;DR

Nvidia’s December acquisition of SchedMD has handed it control of Slurm, the open-source scheduling software that runs around 60% of the world’s supercomputers and underpins large-language-model training workloads at labs including Anthropic, Meta and Mistral. AI specialists and HPC engineers interviewed by Reuters worry the deal could see Nvidia subtly favouring its own chips, with some fearing AMD and Intel hardware will get slower software support over time.

Slurm Sits at a Strategic Choke Point

Slurm is the connective tissue that turns a cluster of GPUs into a usable training and supercomputing system. It is critical for forecasting weather, designing nuclear weapons, and training the largest frontier models. That makes it precisely the kind of software where vendor neutrality matters most — and where sources cited by Reuters are now watching closely.

The benchmark test, according to one Slurm engineer quoted in the piece, will be how quickly Nvidia integrates AMD’s upcoming chips into Slurm versus how quickly it ships support for its own InfiniBand networking and other Nvidia-specific kit. “Could take what’s a common open-source tool and make it so that it works better or exclusively for its own parts,” Intersect360 Research CEO Addison Snell warned.

A Familiar Pattern

The concerns are amplified by Nvidia’s 2022 acquisition of Bright Computing, a cluster-management vendor whose software AI industry sources told Reuters became “optimised for Nvidia, creating a performance penalty for users of other chips without additional work”. Nvidia disputes the characterisation, saying Bright supports “nearly any” CPU or GPU cluster.

Notably, OpenAI does not use Slurm — it relies on Google-derived scheduling — meaning the leverage Nvidia gains here is concentrated on the rest of the frontier-lab and HPC universe rather than being industry-wide.

Looking Forward

For UK universities, national supercomputing facilities and any enterprise running large mixed-vendor GPU clusters, the practical question is contingency planning. Slurm is open-source, so a fork is technically possible — but, as Reuters notes, “it takes effort to produce fully working software”. British research institutions that depend on Slurm for AMD or Intel deployments would do well to start tracking commit velocity and merge patterns now, rather than waiting to discover the lock-in after the fact.