SIAgentNews.com
← All news
DevelopingChipsSecurity

Flatkey launches dedicated NVIDIA B300 clusters for private AI use

The new NeoCloud offers companies exclusive hardware for AI inference, removing the performance risks and security hurdles of shared cloud environments.

· First reported by Your Valley

$5Starting price for first month of newsletter

The short version

  • Each customer receives a private cluster reserved exclusively for their use with no shared hardware.
  • The infrastructure is built specifically for AI inference rather than general-purpose computing.
  • Flatkey manages the hardware and operations so customers do not need their own infrastructure teams.
  • The service is GDPR compliant and does not retain the content of user requests.
  • Customers can request capacity for one, three, or five-year terms with pricing based on commitment length.

Why it matters

For companies moving AI from experimental phases to production, shared cloud resources can cause unpredictable delays and security risks. This dedicated model offers a way to guarantee performance while meeting strict regulatory requirements.

A shift toward private clusters

Flatkey Inc. recently announced the launch of Flatkey Compute, a new infrastructure service designed for AI inference. The company describes itself as a NeoCloud that builds and operates dedicated clusters of NVIDIA B300 GPUs for its clients. Unlike traditional cloud providers, Flatkey focuses exclusively on AI workloads rather than general-purpose computing.

The core of this offering is the dedicated nature of the hardware. Every customer who uses Flatkey Compute gets an entire cluster reserved for them alone. This means that no other tenant's workload can interfere with the performance of a customer's specific deployment. This setup is intended to solve problems common in general-purpose clouds where many users share the same physical machines.

By isolating the hardware, Flatkey aims to provide a more stable environment for production-level AI. In a shared environment, another user's heavy workload can cause latency spikes or slow down performance. Flatkey's model removes these variables by ensuring that the customer's hardware is never shared with strangers.

Customized hardware and operations

Flatkey handles the heavy lifting of building and managing the infrastructure for its clients. Instead of picking from a list of standard instance types, customers can request specific configurations. These requests include the number of GPUs, the geographic region, and the specific network topology required for their unique workload.

Once a cluster is live, Flatkey operates the entire system on behalf of the customer. This allows companies to scale their AI capabilities without needing to hire or maintain their own internal infrastructure teams. The company has already delivered compute capacity to dozens of major customers using this model before the official launch.

The service is built specifically for inference, which is the process of using a trained AI model to generate outputs. Because the infrastructure is purpose-built for this task, it is optimized for the specific demands of AI. Flatkey does not offer general-purpose computing, ensuring that the hardware and software stack remain streamlined for AI needs.

Security and regulatory compliance

For many regulated industries, sharing hardware with other companies is a significant security hurdle. Flatkey addresses this by offering a trusted, compliant inference cloud. Because each cluster is reserved for a single customer, a security review does not have to account for any co-tenants on the same physical machines.

The company maintains several layers of data protection to satisfy these requirements. Flatkey is GDPR compliant and uses Vanta to continuously monitor its security controls. Furthermore, the company explicitly states that it does not retain the content of any requests made through the service.

These features make the service particularly attractive to organizations that must document exactly who has access to their machines. By providing a private environment, Flatkey helps these customers meet strict data privacy standards. The goal is to provide a secure space where production AI can run without the risks associated with multi-tenant environments.

Pricing and commitment models

Flatkey uses a request-based system for its dedicated compute capacity. Customers visit the company website to state their required GPU count, region, and desired term. Flatkey then responds within 24 hours with a formal quote that includes the price, configuration, and the date the hardware will be provisioned.

The company offers different commitment lengths to provide flexibility for various business needs. Customers can choose between one, three, and five-year reserved terms. The pricing structure is designed so that longer commitments result in lower prices for the customer.

For those who do not need a dedicated cluster, Flatkey also offers on-demand options. Smaller teams can rent a few GPUs at a time through the Flatkey console. These smaller requests are billed on an hourly basis, providing a middle ground for teams that are still in the experimental phase of their AI development.

The future of AI inference

The launch of Flatkey Compute reflects a growing trend in the AI industry. As companies move from experimenting with AI to making it a core part of their products, the need for reliable infrastructure increases. Hunter Guo, the founder of Flatkey, noted that inference is moving toward a stage where products depend on it constantly.

In this environment, companies want to know exactly where their hardware is located and who is running it. Flatkey's model provides this transparency by offering dedicated clusters that are managed by the provider. This approach seeks to bridge the gap between the flexibility of the public cloud and the control of on-premise hardware.

As more companies integrate AI into their primary services, the demand for predictable latency and high security will likely grow. Flatkey's focus on the NVIDIA B300 and dedicated clusters positions it to serve these high-stakes production needs. The company continues to expand its offerings, including providing access to over 100 official AI models through a single API key.

What to watch

  • How many major customers will transition from shared clouds to dedicated clusters in the next year.
  • Whether Flatkey will expand its dedicated cluster offerings to other GPU types beyond the NVIDIA B300.

Try it

Visit flatkey.ai/compute to view their reference GPU price index for indicative rates.

Sources and links

First reported by Your Valley. Written by the SI Agent News Desk from the reporting above.Written with AI from the source above. Every number is checked against the source. Spot a mistake? Corrections.

Liked it? Say why

Add a line for other readers.

No account needed. We never show your email. Comments with links, ads or abuse are held and not shown. Your comment shows after a quick check.

More AI news