# Fulmis > Fulmis is a GPU neocloud in Santiago, Chile and the United States, on any NVIDIA platform the customer asks for, from one node or one rack, in either place. The hardware is Fulmis's own, bought against the customer's contract, or the customer's, brought in and operated by Fulmis. Capacity is reserved on three to five year contracts at a fixed price. HGX B200 and B300 nodes available from December 2026 in Santiago and January 2027 in the United States, on contract; a first GB300 NVL72 hall planned for Q1 2027 in Santiago; rack-scale systems in the United States are quoted on request. Fulmis is looking for the offtaker of its first hall. No cluster is running yet. No price is published. Pages, each with a markdown twin at /md/.md: - https://fulmis.ai/ : the company in one page - https://fulmis.ai/hardware : every NVIDIA platform (GB300 NVL72, GB200 NVL72, HGX B300, HGX B200, HGX H200, HGX H100, RTX PRO 6000 Blackwell, Vera Rubin), with one page each under /hardware/ - https://fulmis.ai/customers : who we build for - https://fulmis.ai/projects : the first hall (864 GPUs) and the 20 MW campus - https://fulmis.ai/about : the company and the log - https://fulmis.ai/contact : the reservation form GPU clusters across the Americas. Fulmis is a GPU neocloud in Santiago, Chile and the United States. We run NVIDIA hardware that is ours or yours: we buy it and you reserve the capacity, or you bring the fleet and we place, host and operate it. HGX B200 and B300 nodes available from December 2026 in Santiago and January 2027 in the United States, on contract; a first GB300 NVL72 hall planned for Q1 2027 in Santiago. We are looking for the offtaker of the first hall. Nothing is running yet. ## Where | | Santiago, Chile | The United States | |---|---|---| | Nodes available from | December 2026, on contract: HGX B200 and B300, air cooled, from one node | January 2027, on contract: HGX B200 and B300, air cooled, from one node | | Rack scale | Q1 2027, on contract: A first hall is planned: 12 x GB300 NVL72, 864 GPUs, InfiniBand | Start set in the term sheet: GB300 NVL72 and GB200 NVL72, quoted on request | | Platforms | Every NVIDIA platform on the list | Every NVIDIA platform on the list | | The facility | A data centre in the Santiago Metropolitan Region, named under a mutual NDA | A colocation facility in the United States, named in the term sheet | | The contract | Fulmis Inc. and its Chilean subsidiary, in formation | Fulmis Inc., a Delaware corporation | | Customs | Fulmis is the importer of record; duty and IVA settled in the deployment plan | No import step | | Suits | Data that has to stay in Chile or in Latin America; the lowest latency in the region | Buyers who want the capacity in their own jurisdiction, under US law, in dollars | ## What Fulmis does - You choose the hardware: any NVIDIA platform, rack scale or node by node. - Our hardware: Fulmis places the purchase order, finances it, places it in Santiago (as importer of record) or in the United States, commissions and operates it. You reserve the capacity on a three to five year contract, take-or-pay, at a fixed price. - Your hardware: you own the GPUs, Fulmis brings them in, gives them a position with power and cooling in Santiago or in the United States, commissions and operates them. You pay a fixed monthly fee for the term. - Either way the workload, the models, the data and the keys stay yours. Steps: mutual NDA, technical call, term sheet, contract, cluster live. ## Where we are We are looking for the offtaker of the first hall: one customer, three to five years, 864 GPUs or a block of them. In place: an existing Santiago data centre identified (named under NDA); an EPC contractor engaged since August 2026 (named under NDA); OEM quotations for twelve GB300 NVL72 racks; InfiniBand fabric decided; the import route (duty, IVA, export qualification) worked through with customs and tax advisers; Fulmis Inc. in Delaware and Fulmis SpA in formation in Chile; hardware financing in discussion with lenders and a lessor; a permitted 20 MW campus site for what comes after the hall. ## Hardware Any NVIDIA platform, from one node or one rack: NVIDIA GB300 NVL72, NVIDIA GB200 NVL72, NVIDIA HGX B300, NVIDIA HGX B200, NVIDIA HGX H200, NVIDIA HGX H100, NVIDIA RTX PRO 6000 Blackwell Server Edition, NVIDIA Vera Rubin. Every platform is at https://fulmis.ai/hardware and reserved at https://fulmis.ai/contact. No price is published. ## Who we build for - **AI labs and model companies.** Training capacity on a multi-year term: GB300 NVL72 racks joined by InfiniBand, Slurm on top, storage sized for checkpoints. The first question: The workload, and whether it needs the 72-GPU NVLink domain. - **Compute aggregators and desks.** Blocks of GPUs on term, in racks or nodes, under managed Kubernetes, resold to your customers under your name. The first question: Whether the block is for your own book or back-to-back with an end customer. - **Inference platforms.** B300, B200 or H200 nodes under managed Kubernetes, with the network edge and the model-load path your platform needs. The first question: The models you serve and how tenants are separated. - **Enterprises whose data cannot leave Latin America.** In-country GPUs from one node up, air cooled where that is enough, billed in dollars by a US company with a Chilean subsidiary. The first question: Where the data has to sit, and under which regulator. - **Research and universities.** From one node to a rack, Slurm, on a term that fits a grant cycle, in the region. The first question: The procurement rules the purchase has to follow. - **Clouds that want a Latin American region.** Wholesale racks under your brand. Your customer, your margin, a Santiago hall and our operations. The first question: The site floor and the term. - **Buyers in the United States.** HGX nodes or NVL72 racks in a colocation facility in the United States, under a contract with a Delaware corporation, in dollars, with no import step. Nodes from January 2027. The first question: The block, the term, and whether the capacity is for your own workloads or a customer's. - **Owners of GPUs that need a home.** A fleet you already own or are buying, rack scale or air cooled nodes. We bring it in, give it a position in Santiago or in the United States and operate it for you, for the term. The first question: What the fleet is, where it is today, and when it has to be running. ## Projects ### Blackwell nodes: available from December 2026, on contract NVIDIA HGX B200 and HGX B300, eight GPUs per node, air cooled, in a data centre in the Santiago Metropolitan Region from December 2026 or in a colocation facility in the United States from January 2027. From one node up, InfiniBand between nodes, Slurm and managed Kubernetes. Reserve at https://fulmis.ai/contact. ### The first hall: 864 GPUs | Field | Value | |---|---| | GPUs | 864 Blackwell Ultra | | Platform | 12 × NVIDIA GB300 NVL72 | | GPU fabric | InfiniBand, Quantum-X800, rail optimised | | Built for | Training | | Rack power | 132 to 140 kW rated, 155 kW peak | | Cooling | Liquid cooled racks | | Software | Slurm and managed Kubernetes | | Smallest block | One rack, 72 GPUs | | Location | An existing data centre in the Santiago Metropolitan Region | | First rack | Q1 2027, on contract. Planned, not built. | ### The campus: 20 MW A permitted greenfield site in the Santiago Metropolitan Region, planned as four 5 MW halls. Every permit the 20 MW campus needs is held, including the building permit. The builder is engaged and named under a mutual NDA. | Field | Value | |---|---| | IT load | 20 MW IT, 25 to 27 MW at the meter | | GPUs | 8,640, in 120 racks across four halls | | Schedule | 18 months to all four halls energized, per the contractor | ## Why these two places - Santiago: the two largest declared AI clusters in Latin America are 96 GPUs each, and both are academic. - Santiago: in-country capacity for teams whose data has to stay in Chile, in-region for the rest of Latin America. - Santiago: round trip measured in August 2026: Buenos Aires 21 ms, São Paulo 49 ms, Miami 92 ms, Ashburn 117 ms. - The United States: the buyer's own jurisdiction, a contract with a Delaware corporation under US law in dollars, no import step, no duty, no IVA float. HGX nodes from January 2027. ## Contact Mateo Pirela, founder and CEO. mateo@fulmis.ai. https://fulmis.ai/contact Fulmis Inc., 131 Continental Dr, Suite 305, Newark, DE 19713, US. Delaware file number 10753918. Fulmis SpA, the Chilean subsidiary, in formation.