May 26, 2026
There's a pattern emerging among AI companies that are serious about scaling: they're moving away from shared cloud environments and toward infrastructure they control. Not because the hyperscalers aren't capable, but because when your product depends on compute performance, unpredictability becomes a liability.
That's the context behind our new partnership with Classover Holdings (NASDAQ: KIDZ).
Classover is an AI-driven education technology company building next-generation learning systems for K-12 students, and one of a growing number of AI-native companies making a direct bet on dedicated infrastructure. Together, we've formed Ousia Compute LLC, a joint venture built to acquire, deploy, and commercialize GPU infrastructure for AI compute workloads, with a first-phase investment capacity of up to $50 million.
Classover holds majority ownership of the SPV and acts as managing member. 1Legion is the exclusive infrastructure operator. That separation of roles is deliberate: Classover brings the capital and strategic direction; 1Legion handles the full operational layer, hardware provisioning, deployment, data center management, and ongoing support.
The result is a dedicated compute environment built from the ground up for production AI workloads. No shared resources, no virtualization overhead, no surprise egress costs.
Ousia Compute's clusters will be built on NVIDIA GPUs, including H100, H200, B300, and Blackwell architectures, deployed in Tier 3+ data centers across the United States and globally, interconnected via InfiniBand networking. By eliminating the hypervisor layer entirely, the platform delivers direct hardware access and full memory bandwidth: what you provision is what you get, every time.
Pricing follows a flat-rate model. No data egress fees, no variable transfer costs. For teams running continuous AI workloads, that predictability matters as much as raw performance.
Shared cloud infrastructure is built around utilization efficiency, resources are pooled, allocated dynamically, and priced accordingly. That model works for many use cases. It doesn't work well when you need guaranteed throughput, sustained inference performance, or hardware that behaves consistently across long-running jobs.
Bare-metal changes the equation. The hardware is dedicated, the performance is predictable, and the cost structure holds at scale. Ousia Compute is already in active commercial discussions with enterprise customers looking for exactly that, long-term GPU capacity they can rely on.
The convergence of AI education, AI-native applications, and AI infrastructure is creating a long-term market opportunity that Classover is positioning itself to participate in directly. For 1Legion, Ousia Compute is a natural extension of the infrastructure model we've built, dedicated hardware, managed properly, for teams that need compute they can count on.
