Segment
The AI software stack: what makes the silicon usable
The gap between an accelerator's specification and what a real workload achieves is closed, or not closed, by software. This branch covers the three layers that do it: how kernels are written, how models are distributed across machines, and how a cluster is shared.
In one sentence
The AI software stack is the layer between models and hardware: the kernel programming model and compiler, the frameworks that distribute training across many devices, and the schedulers that allocate a shared cluster.
Every claim about accelerator performance is really a claim about a hardware and software pair. The same silicon delivers very different results depending on whether its kernels are well tuned, whether the framework overlaps communication with computation, and whether the scheduler keeps the machine full.
This is also the layer where competitive position is most durable. Silicon advantages last a generation; a mature, trusted software stack that thousands of engineers already know how to use lasts much longer, and it is the reason a faster chip does not automatically win.
How this breaks down
Split by the layer of abstraction — kernels, distributed training, and cluster scheduling.
- Kernel librariesThe hand-tuned and generated code that turns a chip's specification into achieved performance.Definition page
- Container platformsThe general-purpose cluster manager that accelerator scheduling is built as an extension of.Definition page
- Programming modelsHow code reaches the arithmetic units, and why porting between vendors is hard.Definition pageChokepoint
- Training frameworksData, tensor, pipeline and expert parallelism — and the communication each one costs.Definition page
- Cluster orchestrationScheduling, gang allocation and failure handling across a shared accelerator fleet.Definition page
What this depends on
Technology dependencies are solved by engineering; supply dependencies are solved by building something, which takes years.
- Technology
Accelerator hardware
The stack exists to expose specific hardware; each generation requires new tuning.
AI accelerators - Technology
The interconnect the collectives run on
Everything above the kernel layer assumes gradients and activations can be exchanged between devices quickly. The libraries are written against the fabric's topology, so a change in one forces a change in the other.
AI networking
What depends on this
Other pages in this map that name Software stack as something they cannot do without.
- Artificial intelligenceAI computeCompute nobody can program is inventory. Every part of this layer is chosen for the compilers, kernels and schedulers written against it as much as for its specification.
- Artificial intelligenceModel architectureA model is expressed in a framework and executed through tuned kernels and a scheduler. None of the architectural work below can be run at scale without that layer underneath it.
Companies across Software stack
Every company named on a step below this page, ordered by how many of those steps it appears at. Compiled from the pages themselves rather than written separately, so the two cannot disagree. Not a ranking and not a recommendation.
- Huawei TechnologiesPrivate3 steps
- AnyscalePrivate2 steps
18 more companies appear at a single step each; they are named on the pages for those steps.
How these pages are written
Each page explains one technology in plain language, states what it depends on, and names companies by what they supply at that step. Company roles are described qualitatively and deliberately carry no market shares, revenue figures or rankings — those change faster than an explainer can, and a stale number is worse than none. Ticker links point at company pages on this site and are provided for reference only.
Nothing here is investment advice, a recommendation, or a forecast. A company named on a page about a technology is not thereby a good investment, and the chokepoints described are structural facts about supply chains rather than predictions about prices. Technology moves; where a page describes something as unresolved or in development, that was true when it was written.
Plutux is not an investment adviser. Market data and AI-generated analysis are for information and education only, not investment advice. Disclaimer