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 TytanInferenceMicrosoft Foundry adds Baseten, DeepSeek AI inference (MSFT:NASDAQ)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 inferenceDealsOpenAI Signs 457 MW AI Data Center Deal as AMD Expands Its Challenge to NvidiaClustersBehind Dell AI Factory鈥檚 NVIDIA & AMD Enterprise ExpansionDealsCorvex 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 TytanInferenceMicrosoft Foundry adds Baseten, DeepSeek AI inference (MSFT:NASDAQ)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 inferenceDealsOpenAI Signs 457 MW AI Data Center Deal as AMD Expands Its Challenge to NvidiaClustersBehind Dell AI Factory鈥檚 NVIDIA & AMD Enterprise Expansion
EveryNeocloud

Training clusters

Reserved clusters of thousands of GPUs on one fabric. 105 providers on EveryNeocloud so far.

A training cluster is a block of hundreds to hundreds of thousands of GPUs joined on one fast network, reserved by one customer for months or years. This is how frontier models are trained, and it is the core business of the largest neoclouds. Deals are signed as contracts worth billions of dollars, often before the data centre is built.

What makes a cluster

Inside each server, GPUs talk over NVLink; between servers they use InfiniBand or fast Ethernet in a layout called a fat tree, so any GPU can reach any other at full speed. Shared storage feeds training data, and software such as Slurm or Kubernetes schedules jobs and restarts them when a GPU fails, which in a cluster of 100,000 GPUs happens many times a day.

How it is sold

Customers sign take-or-pay contracts: they pay for the capacity whether or not they use it. Microsoft, OpenAI, Meta and Google have signed such deals with CoreWeave, Nebius, Nscale, IREN and others. The contracts let providers borrow against them to buy GPUs. Critics say this ties a small number of buyers and lenders together; providers say long contracts make the business stable.

Famous training clusters providers

Further reading

馃嚭馃嚫 United States and 4 moreGoogle data centers Hyperscale campuses running Search, YouTube, Gemini and Google Cloud AI workloadsHyperscaler 路 Google TPU 馃嚜馃嚫 SpainBarcelona Supercomputing Center Spanish national supercomputing centre running the MareNostrum machine in BarcelonaResearch cluster 路 Other Nvidia GPUs 路 2005 Centre for Development of Adva 馃嚠馃嚦 IndiaCentre for Development of Advanced Computing Indian government society building supercomputers and advanced computing systemsSupercomputer 路 Other and custom chips 路 1988 馃嚘馃嚭 AustraliaNational Computational Infrastructure Australia's national high-performance computing and data facility at ANU in CanberraSupercomputer 路 Other Nvidia GPUs 路 2007 Humain 馃嚫馃嚘 Saudi ArabiaHumain Saudi AI company building data centres, cloud services and Arabic modelsNeocloud 路 Other Nvidia GPUs 路 2025 馃嚚馃嚦 ChinaBaidu Chinese search and AI group with its own cloud, chips and foundation modelsHyperscaler 路 Other and custom chips 路 2000 The Stargate Project 馃嚭馃嚫 United States and 4 moreThe Stargate Project US AI infrastructure venture building data centres for OpenAISupercomputer 路 AMD Instinct 路 2025 馃嚛馃嚜 GermanyJUWELS Modular supercomputer at J眉lich with an AMD and Nvidia booster moduleSupercomputer 路 Nvidia Ampere 路 2020 Selene 馃嚭馃嚫 United StatesSelene Nvidia's own DGX SuperPOD supercomputer, built in three months in 2020Supercomputer 路 Nvidia Ampere 路 2020 馃嚠馃嚬 Italy and 1 moreGalileo Italian national Tier-1 supercomputer for scientific and technical computingSupercomputer 路 Other Nvidia GPUs 路 2015 馃嚭馃嚫 United States and 2 moreAmazon Web Services Amazon's on-demand cloud platform, offering compute, storage and AI infrastructure worldwideHyperscaler 路 Other Nvidia GPUs 路 2006 PARAM 馃嚠馃嚦 India and 4 morePARAM India's home-grown supercomputer line, built by C-DAC in Pune since the early 1990sSupercomputer 路 Other and custom chips 路 1991 馃嚭馃嚫 United StatesTexas Advanced Computing Center University of Texas centre running supercomputers and storage for U.S. researchResearch cluster 路 Other Nvidia GPUs 路 2001 Oak Ridge Leadership Computing 馃嚭馃嚫 United StatesOak Ridge Leadership Computing Facility US Department of Energy supercomputing user facility at Oak Ridge National LaboratorySupercomputer 路 Other and custom chips 路 1992 Voyager-EUS2 馃嚭馃嚫 United StatesVoyager-EUS2 Microsoft Azure supercomputer in the East US 2 regionSupercomputer 路 Nvidia Ampere EuroHPC Joint Undertaking 馃嚤馃嚭 Luxembourg and 4 moreEuroHPC Joint Undertaking EU public-private partnership building a pan-European supercomputing and AI infrastructureSupercomputer 路 Other and custom chips 路 2018 Fluidstack WorldFluidstack Compute infrastructure company building large-scale AI data centresNeocloud 路 Other and custom chips Ubitus 馃嚡馃嚨 JapanUbitus GPU virtualisation and cloud gaming company running AI compute centres across AsiaNeocloud 路 Other Nvidia GPUs 路 2012 Core42 馃嚘馃嚜 United Arab EmiratesCore42 Sovereign AI cloud and infrastructure provider based in the UAE.Sovereign cloud 路 Other Nvidia GPUs

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.