Agentic AI is creating a new class of industrial-scale demand for tokens, forcing a global race to build liquid-cooled, high-throughput AI factories.
NVIDIA's AI Cloud ecosystem has officially reached six continents, following the addition of Cassava in Africa and Claro in South America. This expansion targets the exploding token demand behind today's most popular AI applications, moving compute closer to regional data, developers, and sovereign AI requirements.
Scaling Infrastructure for the Agentic Era
As AI shifts from model development to high-volume reasoning and inference, the primary metric for infrastructure has moved from raw capacity to the economics of token output. Cloud partners are optimizing for the lowest cost per token and highest throughput per watt to support frontier and open-source models.
CoreWeave is leading this charge by adopting NVIDIA Vera Rubin and Vera CPU architectures, alongside NVIDIA Spectrum-X Ethernet Photonics to provide the networking foundation for million-GPU AI factories. Their platform is specifically extending into robotics and physical AI, utilizing NVIDIA Cosmos 3 to generate synthetic data and accelerate robotics data flywheels. This infrastructure already supports major AI labs, including Anthropic, at scale.
Regional AI Factories and Sovereign Control
In the Asia-Pacific region, Firmus Technologies is expanding its footprint through Project Southgate, developing AI factories in Tasmania, Melbourne, South Australia, and New South Wales. To meet the need for rapid deployment, Firmus is utilizing the NVIDIA DSX platform to design the liquid-cooled Firmus HyperCube, which optimizes for low cost per token while emphasizing renewable power and advanced cooling.
Nebius is similarly building a full-stack platform for physical AI development. Their new Physical AI Workbench integrates NVIDIA Cosmos 3, Isaac Sim, and Isaac GR00T into composable workflows. This allows robotics and autonomous systems teams to move faster from simulation and synthetic data to actual training and evaluation.
Optimizing Token Economics with DSX
To reduce deployment risk and maximize revenue, cloud providers are increasingly adopting the NVIDIA DSX platform. DSX provides tools like DSX Sim for modeling factories before deployment and DSX MaxLPS, which helps power-constrained facilities maximize compute within a fixed power budget—enabling up to 40% more GPUs.
By automating lifecycle management through DSX OS, these AI clouds are bringing capacity online faster and delivering more tokens per watt. This evolution in cloud architecture ensures that as AI agents become more pervasive, the underlying infrastructure can scale to meet their reasoning requirements efficiently.
Source: NVIDIA AI Cloud Ecosystem Expands Worldwide to Meet Global AI Compute Demand
Domain: blogs.nvidia.com
Comments load interactively on the live page.