Flatkey Compute Bets on Single-Tenant B300 Clusters to Win Production AI Inference Buyers

What happened

On November 4, 2026, San Jose-based Flatkey Inc. launched Flatkey Compute, a cloud service that builds and runs dedicated NVIDIA B300 clusters for AI inference. Each cluster is reserved for one customer, and Flatkey operates it as a compliant inference cloud focused only on AI workloads.

The pitch targets teams moving inference into production, where shared general-purpose clouds can bring unpredictable latency, noisy neighbors and harder compliance reviews. Clusters are sized per customer by GPU count, region and network topology, with reserved terms of one, three or five years and quotes promised within 24 hours. Flatkey says it has already delivered capacity to dozens of major customers, is GDPR compliant, monitors security through Vanta and does not retain request content. It also offers hourly on-demand GPUs for smaller teams and a public reference GPU price index. The company only launched in July 2026.

Overview

Flatkey is a NeoCloud provider that builds and operates dedicated NVIDIA B300 infrastructure for AI inference, giving each customer its own hardware sized to its workload. It sells reserved multi-year cluster capacity to AI teams, including regulated customers, and rents GPUs by the hour to smaller teams through its console. The company also gives developers access to more than 100 AI models through a single API key. Flatkey launched in July 2026 and is based in San Jose, California.