Santa Clara, California, United States
NVIDIA
The company whose chips sit under almost all of this stack. NVIDIA does not operate UAE infrastructure itself, but its GPUs are what every cloud and sovereign-AI provider in this catalog is actually renting out.
Where it sits in the stack
- User - NVIDIA does not work at this layer
- Application - NVIDIA does not work at this layer
- Model - NVIDIA does not work at this layer
- Inference - NVIDIA does not work at this layer
- ComputeThe processing power - NVIDIA works at this layer
- GPUThe physical chips - NVIDIA works at this layer
- Data Centre - NVIDIA does not work at this layer
- Power & Network - NVIDIA does not work at this layer
Not in the UAE
NVIDIA does not operate UAE infrastructure directly; its GPUs reach the UAE through cloud and sovereign-AI partners - Core42's AI cloud, for instance, is built on NVIDIA H100 GPUs.
What it offers
H100 / H200 Tensor Core GPU
Hopper-architecture data-centre GPUs widely used for AI training and inference; the H200 adds 141GB of HBM3e memory.
Blackwell (B200 / GB200)
NVIDIA's newest data-centre GPU architecture, positioned for large-scale LLM training and inference.
DGX Systems
Purpose-built AI servers combining eight data-centre GPUs with NVLink interconnect.
What it can do
- Designs the GPUs underneath most large-scale AI training and inference
- DGX/HGX reference systems adopted by cloud and sovereign-AI providers
- NVLink and InfiniBand networking for multi-GPU clusters
What its 2 layers mean
- GPU
The specialised chips - mostly made by NVIDIA - that do the actual arithmetic behind AI, built to run millions of small calculations in parallel far faster than a general-purpose computer chip can.
- Compute
The raw computing capacity - rented by the hour, the month or the year - that trains and runs AI models. This is what "the cloud" mostly means in practice.
Something out of date?
Tell us and we will check it against the official source.