LayersLayer 1 of 15
Compute
Compute is the hardware AI runs on and the companies that rent it out: GPU clouds, AI chips, servers, data centres, and the software that schedules jobs across them. It is where the cost of training and running models is set.
The categories
14 categories and 371 products in Compute.
CategoryWhat it coversProducts
AI chips / accelerator cloudsAI chips / accelerator clouds covers AI chips other than NVIDIA's, offered through a cloud or an API.19
AI data centres & colocationAI data centres & colocation covers operators building/leasing high-density power and space for AI clusters.38
AI networking & optical interconnectAI networking & optical interconnect covers scale-up and scale-out fabrics, AI-tuned Ethernet/InfiniBand, NICs and co-packaged/photonic optics that wire accelerators inside and between AI racks.33
AI servers & on-prem systemsAI servers & on-prem systems covers vendors of GPU servers, AI workstations and turnkey AI factory racks.34
AI storage for GPU clustersAI storage for GPU clusters covers high-throughput parallel file and object storage that feeds GPUs during training, checkpointing and inference (including KV/context offload).30
Cloud GPU notebooks & dev environmentsCloud GPU notebooks & dev environments covers hosted notebooks/workspaces giving instant GPU access for development.23
Compute price indexes & GPU price trackersCompute price indexes & GPU price trackers covers benchmarks, indices and comparison sites for GPU rental prices.14
Edge AI accelerators & NPU toolchainsEdge AI accelerators & NPU toolchains covers edge inference chips, M.2/PCIe accelerator cards and NPUs, plus their vendor model-optimization toolchains and device clouds, used to run models in cameras, PCs, vehicles and IoT devices (not robot dev kits, not datacenter accelerator clouds).31
GPU cluster schedulers & orchestrationGPU cluster schedulers & orchestration covers software that schedules and orchestrates AI jobs across GPU clusters (Slurm, K8s, multi-cloud).28
GPU marketplacesGPU marketplaces covers marketplaces where owners of spare GPU capacity rent it out, directly or through a broker.20
GPU observability & efficiencyGPU observability & efficiency covers tools that monitor GPU health, utilisation, power and cost.23
Hyperscaler GPUHyperscaler GPU covers GPU and AI-accelerator machines rented from the largest cloud providers.20
Neoclouds / GPU cloudsNeoclouds / GPU clouds covers clouds built around GPUs, selling whole clusters and on-demand GPUs.38
Serverless GPUServerless GPU covers services that run code on a GPU and bill by the second, with no server to manage.20
Questions
Compute, answered.
What is the Compute layer of AI infrastructure?Compute is the hardware AI runs on and the companies that rent it out: GPU clouds, AI chips, servers, data centres, and the software that schedules jobs across them. It is where the cost of training and running models is set.
What does the Compute layer include?It has 14 categories: AI chips / accelerator clouds, AI data centres & colocation, AI networking & optical interconnect, AI servers & on-prem systems, AI storage for GPU clusters, Cloud GPU notebooks & dev environments, Compute price indexes & GPU price trackers, Edge AI accelerators & NPU toolchains, GPU cluster schedulers & orchestration, GPU marketplaces, GPU observability & efficiency, Hyperscaler GPU, Neoclouds / GPU clouds and Serverless GPU.
How many Compute products does Unfragile track?371 products across 14 categories, each category with its own page.
What works here
The three products on this layer.
SearchLive
RecommendNot live yet.
RouteNot live yet.