OpenAI GPT OSS and NVIDIA Nemotron models arrive on Bedrock in AWS GovCloud (US)

Amazon Bedrock now offers OpenAI's open-weight GPT OSS models (120B and 20B parameters) and NVIDIA's Nemotron family (Nano 9B v2, Nano 12B v2, Nano 30B, and Super 120B) inside AWS GovCloud (US). The release brings US-origin frontier open-weight models to environments with strict data-residency requirements, targeting government agencies and regulated industries that cannot use commercial cloud regions.

GovCloud availability matters because it unlocks a segment — federal, defense, and critical-infrastructure customers — that requires physically and logically isolated infrastructure with US-person operational controls. Offering open-weight models there means these customers can run inference on data-resident, auditable infrastructure without shipping sensitive data to standard commercial regions or relying solely on closed API models.

The move dovetails with a broader sovereignty theme this week: Palantir separately launched an engine to deploy NVIDIA Nemotron open models in classified, air-gapped environments with explicit data authorization, secure-perimeter enforcement, and full auditability. Open-weight models have become the preferred substrate for sovereign and national-security deployments precisely because they can be inspected, fine-tuned, and run entirely within a customer's control.

Competitively, this positions AWS GovCloud against Azure Government and Google's sovereign offerings, with the open-weight angle differentiating from closed frontier APIs. GPT OSS gives OpenAI a foothold in data-residency workloads it can't serve via its hosted API, while NVIDIA's Nemotron continues its enterprise-open-model expansion. Watch which agencies adopt first and whether GPT OSS performance in restricted environments matches its commercial siblings.