The Pulse
Flatkey Launches Dedicated NVIDIA B300 Cloud for AI Inference
Flatkey says its new service gives each customer a dedicated NVIDIA B300 cluster for AI inference, with deployment details set to fit the customer’s workload. The company offers reserved terms of one, three or five years and says smaller te
AI.info Team · Written with AI · fact-checked ·
Flatkey launched Flatkey Compute, offering dedicated NVIDIA B300 clusters for AI inference. The Nov. 4, 2026, announcement describes the service as a compliant, trusted inference cloud built and operated by Flatkey. The company says most AI teams buy GPU capacity from general-purpose clouds, where many tenants share the same physical hardware, and that this works for experiments but makes production inference harder.
According to Flatkey, teams running production workloads need predictable latency, while another tenant’s workload can slow their service. Regulated customers also have to document who else can access the machines. The company presents single-customer clusters as a way to address those concerns.
Why Flatkey Says Shared Clouds Complicate Inference
Each Flatkey Compute cluster is reserved for one customer and is not shared with another tenant. Flatkey says that arrangement prevents another tenant’s workload from affecting the customer’s performance. It says this also means a security review has no co-tenants to account for. The company says every cluster is dedicated to a single customer, rather than shared across multiple workloads.
The company says it builds and operates the clusters itself, setting GPU count, region and network topology for each customer instead of offering only standard instance types. Flatkey says it had delivered compute capacity to dozens of major customers before the launch. It says that once a deployment is live, it operates it, so customers do not need their own infrastructure team to run it.
Flatkey Limits the Service to AI Workloads
The clusters run on NVIDIA B300 GPUs and are built for AI inference. Flatkey says it does not offer general-purpose computing and designs its infrastructure around AI workloads alone. The company says it is GDPR compliant, monitors its security controls continuously through Vanta and does not retain request content.
“Inference is moving from something teams experiment with to something their product depends on,” said Hunter Guo, founder of Flatkey. “At that point, you don't want to be sharing a machine with strangers. You want your own cluster, you want to know where it is, and you want somebody to run it for you. That is what we are building.”
Customers Request Configurations and Terms
Customers request capacity at flatkey.ai/compute by specifying the GPU count, region and term they need. Flatkey says it replies within 24 hours, then sends a formal quote setting the price, configuration, region and provisioning date. The company offers reserved terms of one, three and five years, with lower pricing for longer commitments. A reference GPU price index on the compute page shows indicative rates before a customer gets in touch.
Flatkey Compute is available now by request. Smaller teams needing only a few GPUs can rent them on demand and pay by the hour through the Flatkey console. Flatkey says it also gives developers access to more than 100 official AI models through one API key. The company says it launched in July 2026 and is headquartered in San Jose, California.