Amazon Bedrock in AWS GovCloud supports NVIDIA Nemotron and OpenAI GPT OSS models

Amazon Web Services in July 2026 expanded the capabilities of Amazon Bedrock in AWS GovCloud (US) by adding support for open-source models OpenAI GPT OSS and the NVIDIA Nemotron series — for the first time on sovereign infrastructure for the U.S. government sector.

Which models are now available in GovCloud?

AWS GovCloud users now have access to two families of open frontier models: open-weight versions of GPT from OpenAI and the full Nemotron lineup from NVIDIA.

• OpenAI GPT OSS 120B — the flagship open model with 120 billion parameters
• OpenAI GPT OSS 20B — a lightweight version for resource-constrained tasks
• NVIDIA Nemotron Nano 9B v2 — a compact second-generation model for edge scenarios
• NVIDIA Nemotron Nano 12B v2 — an expanded Nano with improved reasoning capabilities
• NVIDIA Nemotron Nano 30B — mid-sized model balancing answer quality and inference speed
• NVIDIA Nemotron Super 120B — the series flagship for complex analytical tasks

All models fall into the open-weight category: parameters are published, architecture and system behavior can be independently verified. This is a critically important requirement for government customers who must audit AI systems before deployment in sensitive scenarios.

Why government agencies need GovCloud

AWS GovCloud (US) is an isolated cloud region for U.S. federal, state, and defense organizations. It complies with FedRAMP High, DoD Impact Level 2/4, and ITAR requirements: data remains physically within the U.S., access is restricted to U.S. citizens, and the entire stack has been audited for government security standards compliance.

Until this announcement, government organizations using Bedrock worked primarily with proprietary models — Anthropic Claude, Amazon Titan, and others. The arrival of open-weight models changes the situation fundamentally: agencies gain systems with transparent weights suitable for independent verification. This reduces "black box" risks and enables organizations to conduct their own security assessments before deployment.

The ability to keep data residency within GovCloud guarantees that model requests and responses do not leave the protected region — a principle critical when working with citizen personal data, intelligence materials, and national security documentation.

How to get started with new models

Amazon Bedrock offers several inference modes in GovCloud: via API using AWS SDK, through the Bedrock web console, and with support for managed inference providers for flexible load distribution. Two service tiers are available — on-demand (pay-per-call) and Provisioned Throughput (capacity reservation for predictable performance).

NVIDIA Nemotron models are trained with emphasis on enterprise and government tasks, including RLHF applied to data reflecting corporate and government policies. The lineup spans a wide range: from lightweight Nano versions for rapid inference to the powerful Super 120B for comprehensive analysis.

What this means

The arrival of frontier open-weight models in AWS GovCloud marks a new milestone for government AI: transparent, auditable models with sovereign data storage become the standard for regulated sectors. Competition between NVIDIA and OpenAI for a share of the government segment is now unfolding directly on federal-level infrastructure.