Mountain View, California, United States
Groq
Groq builds LPUs - chips designed specifically for inference rather than training - and sells access to them through GroqCloud. Its major Gulf build-out is in Saudi Arabia, not the UAE.
Where it sits in the stack
- User - Groq does not work at this layer
- Application - Groq does not work at this layer
- Model - Groq does not work at this layer
- InferenceRunning the model, live - Groq works at this layer
- Compute - Groq does not work at this layer
- GPUThe physical chips - Groq works at this layer
- Data Centre - Groq does not work at this layer
- Power & Network - Groq does not work at this layer
Not in the UAE
Groq's Gulf build-out is in Saudi Arabia, not the UAE: a Dammam data centre built with Aramco Digital is GroqCloud's second global region, backed by a $1.5B Saudi commitment announced at LEAP 2025.
What it offers
LPU Inference Engine
Chips designed specifically for low-latency AI inference rather than model training.
GroqCloud
Hosted API access to open models, run on LPU hardware.
What it can do
- Inference-only chip architecture built for low latency
- GroqCloud API access to open models
- A 19,000-LPU regional data centre built with Aramco Digital
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.
- Inference
The moment a trained model is actually used to produce an answer, in about the time it takes to read this sentence. Training a model happens once; inference happens every single time someone uses it.
Something out of date?
Tell us and we will check it against the official source.