Nvidia's AI Cloud partner network now spans six continents, with six providers — CoreWeave, Crusoe, Lambda, Nebius, Vultr and YTL — qualifying for Exemplar Cloud status since the tier launched last year. The expansion, detailed in a May 31, 2026 post from Nvidia, added Cassava in Africa and Claro in South America to a roster already anchored by CoreWeave, Firmus, IREN and Nscale. Nvidia is also rolling out DSX MaxLPS, a configuration that fits up to 40% more GPUs into a fixed power budget — the kind of number that matters when grid capacity is the binding constraint on every new AI buildout.
The pitch to partners is economic. Nvidia is selling AI Clouds on lowest cost per token and best throughput per watt, the two metrics that now dictate whether a frontier-model training run or a high-volume inference workload pencils out. The company says full-stack codesign across compute, networking, memory and storage is what delivers that cost-per-token edge versus alternatives.
“Every company and every country needs AI factory infrastructure to turn data into intelligence.”— Jensen Huang, Founder and CEO of Nvidia
Regional growth is concentrated in Southeast Asia, Australia and the Americas. Firebird, GMI Cloud, Indosat Ooredoo Hutchison, Lambda, Naver Cloud, Sharon AI, Yotta and YTL are building capacity targeted at national AI programs, financial services, telecommunications, manufacturing, education and healthcare. For governments, the regional footprint is a sovereignty play; for developers, it's latency and data residency.
Key facts
- 01Nvidia's AI Cloud partner network now reaches six continents after adding Cassava in Africa and Claro in South America.
- 02Six partners have hit Exemplar Cloud status since the tier launched last year: CoreWeave, Crusoe, Lambda, Nebius, Vultr and YTL.
- 03Nvidia's DSX MaxLPS lets power-constrained AI factories fit up to 40% more GPUs within a fixed power budget.
- 04CoreWeave is among the first adopters of Spectrum-X Ethernet Photonics, targeting million-GPU AI factory networking.
- 05Jensen Huang delivers the GTC Taipei keynote on June 1 at 11 a.m. Taipei time.
Firmus Technologies is one of the more aggressive builders. Through Project Southgate, the company is putting AI factories in Tasmania, Melbourne, South Australia and New South Wales, and has deployed infrastructure in Singapore via ST Telemedia Global Data Centres. The Firmus HyperCube, a liquid-cooled modular design built around Nvidia DSX, is engineered for fast deployment and low cost per token.
Liquid cooling is the throughline. Asia-Pacific operators are skipping air-cooled retrofits and going straight to liquid at gigawatt scale, betting that the cooling efficiency translates directly into token economics. The DSX reference architecture is meant to compress the design-to-deployment cycle for those builds.
Firmus is framing this as industrial demand, not experimentation.
“AI agents are creating a new class of industrial-scale demand for tokens, and Asia-Pacific needs AI factories that can be built faster, liquid-cooled more efficiently and operated at gigawatt scale.”— Tim Rosenfield, Co-CEO of Firmus
CoreWeave is taking a different lane: frontier infrastructure for the largest labs. The company is an early adopter of Nvidia Vera Rubin and the Vera CPU, and among the first deploying Spectrum-X Ethernet Photonics — the networking layer Nvidia says is required for million-GPU AI factories. Anthropic is named as one of the labs running on CoreWeave's stack.
CoreWeave is also pushing into physical AI, integrating Nvidia Cosmos 3 — the latest world foundation model — to help robotics teams generate synthetic data, fine-tune models and run training flywheels. We covered the Cosmos 3 launch earlier this month; CoreWeave's adoption is the first sign of how that model moves from announcement to production workload.
“AI factories are becoming the foundation for the agentic era.”— Michael Intrator, Cofounder, chairman and CEO of CoreWeave
Nebius is building in the same direction with a different emphasis. The company is layering its Nebius AI Cloud, a Token Factory inference layer, and a new Physical AI Workbench that composes Cosmos 3, Isaac Sim and Isaac GR00T into workflows AI agents can assemble themselves. Founder and CEO Arkady Volozh's framing is that developers shouldn't spend weeks wiring infrastructure before they can train a model.
The DSX platform underneath all of this is Nvidia's bid to standardize how AI factories get built. DSX Sim models deployments before steel goes in the ground. DSX Flex adapts workloads to grid conditions. DSX MaxLPS handles the power-constrained case with the 40% GPU-density gain. DSX OS automates lifecycle management once a site is live.
The skeptic's case is that announced capacity and delivered capacity remain two different things. Million-GPU AI factories require not just chips and cooling but power interconnects, water rights and substation timelines that don't move on Nvidia's roadmap. Several of the named partners — Cassava, Claro, Firebird — are early enough in their buildouts that token-economics claims are forward-looking rather than measured.
For Nvidia, the AI Cloud ecosystem is a hedge against hyperscaler concentration. Every Exemplar Cloud and sovereign partner is a customer for Vera Rubin, Spectrum-X and the DSX platform that doesn't route through AWS, Azure or Google Cloud — and a buyer Nvidia can sell directly to as the agentic-AI inference market scales. Jensen Huang's GTC Taipei keynote on June 1 at 11 a.m. local time is where the next tranche of partners and silicon details will land.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
