Philippine Partners Plan 10,000 Nvidia Blackwell Ultra GPUs
YCO Cloud and Singapore-based Aolani plan a Philippine AI cluster with more than 10,000 Nvidia Blackwell Ultra GPUs in its first phase. They have chosen the GB300 NVL72 architecture and target a launch in the first quarter of 2027. The partners intend to sell computing capacity to government bodies, domestic businesses, startups and AI companies operating across borders. The proposed scale gives a concrete form to a Philippine sovereign AI project: a named system, local infrastructure partners and a launch period rather than a general statement of ambition. Such a first phase would create a local option for organizations that depend on foreign clouds. It also tests whether an operator can gather smaller workloads into a cluster large enough to justify an advanced rack-scale design. The announcement is a plan rather than an installed system.
Aolani is providing GPU financing, deployment and operating capabilities. YCO Cloud brings the site and local infrastructure experience, and says its partnership with Science Park of the Philippines Inc. covers power, water and connectivity. YCO chief executive Nik de Ynchausti pointed to the country's submarine cable connections and growing renewable energy resources. The partners also describe water-efficiency standards and dedicated renewable power. These details matter for a large rack-scale system because the GPU choice and the facility's ability to operate it are being assembled by different organizations. Their division of labor gives a regional cloud project both the capital structure and the physical operating base required to serve customers. In a tropical market, cooling, water and reliable electricity can become as important to the economics of utilization as the accelerator specification itself.
The target market is broader than one anchor tenant. Government agencies may value domestic processing for public services and sensitive data, while local enterprises and startups could rent advanced computing instead of financing their own clusters. Aolani's role as operator makes the project a potential commercial cloud platform, with utilization drawn from many customers. That differs from an installation built around one hyperscaler's internal workload: the partners must assemble a market for capacity across public and private buyers. That mix may include model development, enterprise inference and public-service applications with different usage patterns. Filling the system across those patterns could improve asset utilization, while domestic hosting gives the operator a commercial reason to differentiate itself from general-purpose capacity sold from abroad.
Philippine officials welcomed the project. Leandro Angelo Y. Aguirre, an undersecretary in the Department of Information and Communications Technology, said it could help Filipinos become “builders and developers of AI.” Rosemarie Edillon, an undersecretary in the economic planning department, described the government as a partner in a trusted and inclusive AI ecosystem. Their comments indicate policy support for domestic computing capability and a potential public-sector customer base. The officials did not announce a government purchase, but their emphasis on local development helps explain why a Philippine facility might attract support from agencies and research institutions. The platform’s commercial reach will depend on converting that policy interest and the private sector’s demand into sustained computing use.
For Nvidia, the project is a prospective sale of integrated Blackwell Ultra systems through regional infrastructure specialists. GB300 NVL72 connects accelerators, CPUs and networking at rack scale, so the architectural choice can carry more of Nvidia's stack into a market where local ownership and customer relationships are central. A working Philippine cluster would also give other Southeast Asian operators a regional example of how to package that stack as sovereign capacity. The size is notable; the partners' ability to assemble financing, facilities and paying workloads will determine its commercial reach. At 10,000 GPUs, the first phase would be substantial enough to make component supply, installation, networking and customer acquisition central to the outcome. Aolani’s financing model is therefore part of Nvidia’s route into this market, not merely an ancillary detail of the site announcement.
Analysis
The proposed 10,000-GPU first phase needs both capital formation and enough customers to keep a large cluster utilized. Aolani supplies financing and operation, YCO supplies the local site and access to infrastructure, and the pair would aggregate demand from government and private users who may be too small to buy rack-scale systems individually. Nvidia’s GB300 NVL72 choice can carry CPUs and networking alongside accelerators, increasing the content of a successful installation. The commercial test for the partners is whether domestic demand supports attractive utilization and pricing; for Nvidia, a working site could establish a repeatable channel through regional operators.