ZML targets AI chip lock-in with cross-platform inference software
ZML’s free inference server is designed to run open-source models across multiple chip platforms, targeting hardware lock-in and rising inference costs.
The AI bottleneck is shifting from training models toward running them efficiently across a fragmented hardware market.
What happened
ZML launched ZML/LLMD, a free LLM inference server designed to run open-source models across multiple hardware platforms.
The software supports systems including Nvidia and AMD GPUs, Google TPUs, Apple Metal and Intel Arc.
The company says the goal is to reduce hardware silos and let organisations mix chip architectures while optimising inference performance.
Why it matters
AI deployment is becoming increasingly expensive, and companies do not want to rebuild the software stack every time they change hardware.
A cross-platform inference layer can make alternative chips more viable and give buyers more leverage against vendor lock-in.
That matters because hardware concentration is reinforced not only by chip performance, but by software ecosystems that make switching difficult.
The bigger picture
Inference infrastructure is becoming one of AI’s most strategic software layers.
As companies use more models and more hardware types, the ability to move workloads across chips could become a major source of flexibility and cost control. ZML is targeting that transition directly.