News
DealsCorvex agrees USD 103 million take-or-pay deal to perform AI inference with Mid-Atlantic GPUsSovereignNeysa scales AI infrastructure in India with AMD Instinct MI350X GPUs on TytanSovereignNeysa Deploys AMD GPU Cluster in Mumbai for AI Training and InferenceInferenceMicrosoft Foundry adds Baseten, DeepSeek AI inference (MSFT:NASDAQ)ClustersWhat Does Hewlett Packard Enterprise (HPE) Gaining A Top 14 AI Supercomputer Mean?ClustersDataOne and Rowan University sign engineering co-op program agreement for DataOne's Vineland AI factoryClustersLook inside Cal Poly's $3 million AI Factory (Video)DealsOpenAI Taps AMD Chips for 457 MW of AI Capacity in Six-Year TensorWave DealClustersCal Poly launches an AI data center on campus. Take a look insideInferenceArista takes aim at the next AI network battleground: distributed inferenceDealsCorvex agrees USD 103 million take-or-pay deal to perform AI inference with Mid-Atlantic GPUsSovereignNeysa scales AI infrastructure in India with AMD Instinct MI350X GPUs on TytanSovereignNeysa Deploys AMD GPU Cluster in Mumbai for AI Training and InferenceInferenceMicrosoft Foundry adds Baseten, DeepSeek AI inference (MSFT:NASDAQ)ClustersWhat Does Hewlett Packard Enterprise (HPE) Gaining A Top 14 AI Supercomputer Mean?ClustersDataOne and Rowan University sign engineering co-op program agreement for DataOne's Vineland AI factoryClustersLook inside Cal Poly's $3 million AI Factory (Video)DealsOpenAI Taps AMD Chips for 457 MW of AI Capacity in Six-Year TensorWave DealClustersCal Poly launches an AI data center on campus. Take a look insideInferenceArista takes aim at the next AI network battleground: distributed inference
EveryNeocloud

Cerebras wafer-scale

The WSE-2 and WSE-3. 3 providers on EveryNeocloud so far.

Cerebras makes the largest chips ever sold: each Wafer-Scale Engine is a whole 300 mm silicon wafer, instead of the many small chips usually cut from one. The WSE-2 of 2021 had 2.6 trillion transistors; the WSE-3 of 2024 has 4 trillion and 900,000 cores. Cerebras sells them in CS-3 systems and, increasingly, as a fast inference cloud.

Why a whole wafer

Most of the time in large AI jobs goes into moving data between chips and memory. By keeping a whole model layer on one wafer, with 44 GB of fast on-chip memory on the WSE-3, Cerebras cuts those trips. The hard parts are yield, since every wafer has defects, which Cerebras handles with spare cores, and power and cooling for a chip that draws over 20 kW.

Who runs it

Cerebras runs its own data centres in North America and Europe for its inference service, launched in August 2024, which serves open models at very high speeds. G42 in Abu Dhabi has been its largest customer, and the two built the Condor Galaxy supercomputers from 2023. Research centres such as the Pittsburgh Supercomputing Center and US national labs also run Cerebras systems.

Famous cerebras wafer-scale providers

Further reading

Photo credits on this page

The small pictures above come from Wikimedia Commons under free licences. Some are cropped to fit.

The weekly digest

Every Monday: the provider of the day, new pages on EveryNeocloud, and the conferences and anniversaries of the week ahead.