Flatkey Launches Flatkey Compute, a NeoCloud for Dedicated NVIDIA B300 Clusters

Flatkey launched Flatkey Compute, a dedicated NVIDIA B300 inference cloud available by request, with each cluster reserved for a single customer and sized to its workload. The company says its service is GDPR compliant, continuously monitors security controls through Vanta, and does not retain request content. Reserved terms of one, three and five years are available, with longer commitments priced lower; no pricing figures were disclosed.

Analysis

The key economic trade-off is dedicated capacity versus pooled utilization—not simply “better” versus “worse” cloud. Flatkey’s model could appeal to production workloads with steady demand, latency sensitivity, or procurement rules that make shared infrastructure harder to approve. But customer isolation also removes the provider’s ability to pool variable demand across tenants; profitability therefore hinges on utilization, power and networking costs, and whether contracted prices cover the full hardware lifecycle. One-, three-, and five-year terms may improve revenue visibility while transferring demand and technology-obsolescence risk to customers.

For hyperscalers and NeoCloud competitors such as CoreWeave, the launch is a competitive signal, not evidence of material share loss. Shared instances should remain attractive for bursty, price-sensitive workloads; the likely response is more reserved, isolated inference offerings rather than a wholesale shift away from general-purpose clouds. NVIDIA may benefit if dedicated deployments expand B300 demand, but this announcement alone does not establish order volume or incremental GPU supply.

The main diligence gap is who owns or finances the hardware and what utilization, pricing, and service-level commitments support the economics. “GDPR compliant” and no request-content retention do not, by themselves, establish suitability for every regulated workload; customers will still scrutinize controls, data location, and independent assurance. Near term, the launch is unlikely to change public-company earnings estimates. Over 1–3 months, signed deployments and disclosed capacity would be more informative; over 6–18 months, utilization, renewals, and inference price declines will determine whether dedicated clouds are durable businesses or capital-intensive niches. Contrarian point: dedicated clusters can improve customer control while worsening provider asset efficiency.

Key Decisions for Investors

• No event-driven position on this launch: Flatkey’s scale, funding, hardware ownership, pricing, and committed capacity are undisclosed, so the announcement does not support a defensible public-equity earnings trade.
• Track NVIDIA and public cloud/AI-compute providers for evidence of B300 allocation, dedicated-inference product launches, and customer migration. Treat the launch as a competitive watch item, not a reason to assume near-term share loss at hyperscalers.
• Before underwriting the NeoCloud model, verify contracted utilization or take-or-pay terms, GPU financing/ownership, power and network costs, uptime remedies, and renewal economics. Weak utilization or falling rental prices would challenge the dedicated-capacity thesis.
• Reassess only if Flatkey or competitors disclose material capacity commitments or customers identify production workloads moving from shared clouds. Conversely, limited deployments, delayed provisioning, or customers retaining burst workloads on shared infrastructure would falsify a broad displacement thesis.