We Are the Frontier

Dedicated B300 GPU Clusters Go to Single Customers

Flatkey Compute, launched November 6, 2026, offers reserved single-tenant NVIDIA B300 capacity as AI inference demands isolation.

Updated on Nov. 5, 2026 in Tech

Rows of dark server cabinets line a clean raised-floor data hall under cool, even light, with overhead cooling pipes and a cable bundle across the floor.
Flatkey Inc. launched Flatkey Compute on November 6, 2026, offering reserved, single-tenant NVIDIA B300 GPU clusters to customers who need isolated AI capacity. AI Illustration. Upload story photo >

Flatkey Inc. launched Flatkey Compute on November 6, 2026, a service that builds and operates dedicated NVIDIA B300 GPU clusters, each reserved for a single customer. The service is available now, not announced for later, and Flatkey says it has already delivered capacity to dozens of major customers.

Why it matters

The launch reflects a shift toward dedicated, single-tenant AI compute, in which customers pay for isolated GPU clusters rather than sharing general-purpose cloud infrastructure. Production inference workloads that need predictable latency, and regulated customers who must document who else can access their machines, are the buyers this model targets.

Flatkey Compute builds dedicated clusters of NVIDIA B300 GPUs, the company's current-generation accelerator, with each cluster reserved for one customer. The company reports serving dozens of major customers before launch, but no throughput, latency, or utilization benchmarks were published alongside the announcement.

The players

Flatkey Inc.

A San Jose-based provider of dedicated NVIDIA B300 GPU clusters for AI inference and training, which also offers access to more than 100 official AI models through one API key.

The details

A NeoCloud is a cloud provider specialized in GPU capacity for AI workloads rather than general-purpose computing. Flatkey does not offer general-purpose computing; it provisions whole GPU clusters for individual customers, so no other tenant's workload shares the same machines. Customers request capacity at flatkey.ai/compute by stating GPU count, region and contract term, and Flatkey responds with a formal quote that fixes price, configuration, region and provisioning date. Flatkey says it does not retain request content, is GDPR compliant, and monitors its security controls continuously through Vanta, a compliance-automation platform.

Timeline

  1. July 2026: Flatkey launched as a company.

  2. November 6, 2026: Flatkey Compute launched publicly.

The Tech Race

Flatkey enters the NeoCloud segment by selling single-tenant NVIDIA B300 clusters, positioning itself against general-purpose cloud providers that share hardware across customers. The offering is a bet that isolation and contractual control will matter more to production AI buyers than the flexibility of shared infrastructure.

Teams running production AI inference can now request reserved B300 capacity with 1, 3 or 5 year terms, receiving a quote within 24 hours. Regulated customers gain a single-tenant option where they can document exactly who has access to the machines.

The takeaway

Watch whether Flatkey publishes throughput or utilization benchmarks for its B300 clusters, since the launch currently rests on customer-count and terms claims alone. Buyers evaluating single-tenant GPU capacity should compare the quoted price per reserved cluster against the cost of shared alternatives once Flatkey discloses pricing.

Further reading

For broader context on how dedicated GPU capacity is being built out, see Data Centers coverage in San Jose.

More information

Customers can submit a capacity request on the Flatkey Compute capacity request page, which states GPU count, region and term.

Source note: This article includes information reported by The Manila times.