HUMAIN activates in‑house B300 compute stack

HUMAIN said it has switched on a B300 artificial intelligence compute stack that its own team assembled and activated without outside integration partners. (x.com) HUMAIN is a Saudi artificial intelligence company owned by the Public Investment Fund, and it was launched on May 12, 2025 to build across the full stack: data centers, cloud infrastructure, models, and applications. (pif.gov.sa) A compute stack is the hardware and software layer that runs model training and inference, and HUMAIN markets its own platform as infrastructure for high performance computing, training, inference, and edge workloads. (humain.com) The B300 name points to Nvidia’s Blackwell generation, and Nvidia said on May 13, 2025 that HUMAIN’s first phase would use an 18,000-chip Nvidia Grace Blackwell GB300 supercomputer with Nvidia InfiniBand networking. (nvidianews.nvidia.com) HUMAIN’s claim here is not that it built the chips itself. The claim is that its in-house team put together and brought online the cluster that turns those chips, networks, and management software into usable computing capacity. (x.com) That distinction has become central in the artificial intelligence infrastructure race, where companies often buy processors from Nvidia or Advanced Micro Devices but still try to control the system design, networking, power, and deployment schedule themselves. (nvidianews.nvidia.com) (newsroom.cisco.com) HUMAIN has been building toward that position since 2025, when it announced partnerships with Nvidia, Amazon Web Services, Advanced Micro Devices, Cisco, and Groq as it pursued large-scale data center and cloud capacity in Saudi Arabia. (nvidianews.nvidia.com) (datacenterdynamics.com) In August 2025, Tareq Amin said HUMAIN had broken ground on data centers in Riyadh and Dammam, with initial capacity of 100 megawatts at each site and a target to bring them online in the second quarter of 2026. (datacenterdynamics.com) Amin has also argued that ownership of the operating model matters as much as buying hardware. At the Public Investment Fund Private Sector Forum in February 2026, he said HUMAIN had built a centralized artificial intelligence agent orchestration platform and was running more than 150 artificial intelligence agents across corporate functions. (saudigazette.com.sa) The immediate test is whether HUMAIN can turn that in-house assembly milestone into sustained capacity at scale. Its earlier public roadmap called for hundreds of megawatts of artificial intelligence infrastructure and several hundred thousand Nvidia graphics processors over five years. (nvidianews.nvidia.com)