DevelopingChips

Flatkey launches dedicated NVIDIA B300 clusters for AI inference

Flatkey is offering a 'NeoCloud' where customers get exclusive, single-tenant hardware to run production AI workloads without sharing resources.

24 hoursTime to receive a formal quote

The short version

Each cluster is reserved for one customer, ensuring no other tenant's workload can affect performance.
The service uses NVIDIA B300 GPUs specifically designed for AI inference rather than general-purpose computing.
Flatkey manages the infrastructure, meaning customers do not need their own internal teams to run the hardware.
The company provides GDPR-compliant security and does not retain any request content from its customers.
Customers can request custom configurations for GPU count, region, and network topology.

Why it matters

For companies moving AI from experiments to core products, shared cloud hardware can cause unpredictable delays and security risks. This model offers a middle ground between managing your own data center and using a public cloud.

Moving beyond shared clouds

Many AI teams currently rely on general-purpose cloud providers to access GPU capacity. While these platforms are convenient for testing and small experiments, they often involve many different customers sharing the same physical hardware. This shared environment can lead to unpredictable performance issues where one user's heavy workload slows down another's.

Flatkey is addressing these limitations by launching Flatkey Compute, which it describes as a NeoCloud. This service focuses exclusively on AI inference, which is the process of using a trained model to generate outputs. By specializing in this specific task, the company aims to provide a more stable environment than broad-market cloud providers.

The primary goal of this new offering is to remove the variables that plague multi-tenant systems. For companies running production-level AI, consistency is vital. Flatkey wants to ensure that a customer's performance is never impacted by the activity of strangers on the same machine.

The dedicated cluster model

The core of Flatkey Compute is the dedicated NVIDIA B300 cluster. Unlike standard cloud instances where you pick from a list of pre-set sizes, Flatkey builds and sizes each cluster specifically for a customer's unique workload. This includes customizing the GPU count, the geographic region, and the network topology for every deployment.

Once a cluster is live, Flatkey takes over the responsibility of operating the hardware. This means that a company can benefit from high-end infrastructure without needing to maintain its own internal team of engineers to manage the servers. It provides a managed experience that still feels like private hardware.

Because the entire cluster belongs to a single customer, it simplifies the security review process. Regulated industries often need to document exactly who has access to their machines. In a dedicated setup, there are no co-tenants to account for, which makes it easier to meet strict compliance standards.

Security and compliance standards

Data protection is a central pillar of the Flatkey Compute offering. The company states that its infrastructure is GDPR compliant, which is a major standard for data privacy in the European Union. To ensure these standards are met, Flatkey uses a tool called Vanta to monitor its security controls on a continuous basis.

A significant feature for privacy-conscious companies is that Flatkey does not retain any request content. This means that the data being processed by the AI models is not stored by the provider after the inference is complete. This helps maintain a high level of trust for customers handling sensitive information.

The company's focus on AI-only workloads further reinforces this security posture. By not offering general-purpose computing, Flatkey can optimize its infrastructure and security protocols specifically for the needs of AI developers. This specialization allows them to build a more trusted environment for production-grade tasks.

Pricing and availability

Flatkey has structured its pricing to accommodate different scales of commitment. Customers can request capacity through their website by specifying their needs for GPU count, region, and the length of their term. Flatkey then provides a formal quote within 24 hours that outlines the specific price and provisioning date.

The company offers reserved terms of one, three, and five years. Those who commit to longer terms are eligible for lower pricing, which helps companies plan their infrastructure costs over a longer period. A reference GPU price index is also available on their site to give potential customers an idea of indicative rates.

While the dedicated clusters are the primary focus, Flatkey also offers a more flexible option for smaller teams. These users can rent a few GPUs on demand through the Flatkey console. These smaller requests are billed by the hour, providing a way for smaller projects to access the same hardware without a long-term commitment.

The shift to production AI

The launch of Flatkey Compute reflects a broader trend in the technology industry. As more companies integrate AI into their core products, they are moving away from experimental setups. Hunter Guo, the founder of Flatkey, noted that inference is moving from something teams experiment with to something their product depends on.

This shift creates a demand for infrastructure that feels like private hardware but functions like a cloud service. Companies want to know exactly where their hardware is located and who is running it. Flatkey is positioning itself to meet this demand by providing a managed, dedicated environment for the NVIDIA B300 chips.

Looking ahead, the success of this model will depend on how many enterprise customers choose dedicated clusters over the convenience of public clouds. As production AI becomes more common, the need for predictable latency and high security will likely drive more companies toward these types of specialized NeoClouds.

try_itale_content_summary_not_needed_for_json_but_kept_in_mind_for_structure_only_as_per_instructions_no_extra_fields_allowed_except_those_listed_in_json_schema_the_user_wants_one_article_in_json_format_no_prose_around_it_the_output_should_be_only_the_json_object

try_it_content_summary_not_needed_for_json_but_kept_in_mind_for_structure_only_as_per_instructions_no_extra_fields_allowed_except_those_listed_in_json_schema_the_user_wants_one_article_in_json_format_no_prose_around_it_the_output_should_be_only_the_json_object

try_it_content_summary_not_needed_for_json_but_kept_in_mind_for_structure_

Sources and links

Officialflatkey.ai/compute flatkey.ai
Officialconsole.flatkey.ai/gpu-cluster console.flatkey.ai
ReportingPR Newswire: the article we sourced

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.

Listen to the AI Wire headlines

Listen to the AI Wire

Listen to our daily AI shows every morning: on your commute, in the shower, wherever you are

Listen to the daily shows

First reported by PR Newswire · Read the original. 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.