Public Sector

New updates on AWS GovCloud (US) | Issue #4 (2026)

July 1–15, 2026 | New updates on AWS GovCloud (US): This issue features insights on deploying AI at production scale from AWS Summit Washington, DC, the $1 billion AWS Forward Deployed Engineering investment, and getting started with Anthropic Claude Opus 4.8 in AWS GovCloud (US), plus the latest features and services.

Executive Insights

From Evaluation to Deployment, AWS Invests $1 Billion to Accelerate AI

The Lincoln Memorial, Washington Monument, and U.S. Capitol at sunset with a vibrant orange and blue sky.

At the 2026 AWS Summit in Washington, DC, government technology leaders from the U.S. Army, the Department of War, Idaho National Laboratory, and the Intelligence Community came together for a candid conversation on deploying AI at production scale in regulated environments. Their message was unified: speed and compliance are not in tension. From GenAI.mil serving all military branches as an enterprise IT service, to INL accelerating nuclear energy research, to governance frameworks enabling responsible AI adoption across the Intelligence Community, these leaders shared a common playbook: executive sponsorship that treats AI as infrastructure rather than experimentation, governance models built into the architecture from day one, and a bias toward production-ready platforms that compress timelines without compromising compliance.

To help AWS move faster with AI adoption, AWS announced a $1 billion investment to create the AWS Forward Deployed Engineering (FDE) organization, embedding thousands of expert AI engineers directly with customer teams to co-develop and deploy agentic AI solutions. The AWS FDE model is agentic-first, compresses deployment timelines from months to days, and is designed so customers are self-sufficient when an engagement ends. Unlike traditional consulting, AWS FDE builds for the long term: customers leave engagements with both new solutions and new engineering capabilities, including agentic systems running in their own AWS environment, lasting AI skills, workflows, and patterns they can use to innovate independently. AWS FDE is built for organizations operating in regulated industries and government, where security, governance, and speed to production are non-negotiable.

With up to $50 billion in purpose-built AI and supercomputing infrastructure, a new $1 billion commitment to forward-deployed engineering, and new compute, model, and security capabilities launching throughout 2026, AWS GovCloud (US) is ready to support what comes next, today.

Solutions Insights

Getting Started with Anthropic Claude Opus 4.8 in AWS GovCloud (US)

Here's what practitioners need to know to get started with Claude Opus 4.8 in AWS GovCloud (US).

Claude Opus 4.8 is accessible through the Converse API on bedrock-runtime in us-gov-west-1, using the cross-region inference profile ID us-gov.anthropic.claude-opus-4-8. The cross-region inference profile is required: using the raw model ID directly through the Converse API returns a ValidationException. The model supports up to 1 million tokens of context and 128K max output tokens, with text and image input modalities. Claude Opus 4.8 uses adaptive thinking with configurable reasoning effort (low, medium, high, xhigh), allowing teams to tune chain-of-thought depth against latency and cost per task. Fixed thinking budgets are not supported, and temperature, top_p, and top_k parameters are all rejected, so teams should guide model behavior through prompting.

Access requires a two-account setup: a designated administrator must first accept the Claude Opus 4.8 EULA in the linked commercial AWS account (us-east-1 or us-west-2), then a AWS GovCloud (US) administrator enables the model in the paired AWS GovCloud (US) account in us-gov-west-1. The EULA cannot be accepted directly in AWS GovCloud (US). Once access is active, the runtime identity needs bedrock:InvokeModel and bedrock:InvokeModelWithResponseStream permissions scoped to the cross-region inference profile ARN and the underlying foundation-model ARNs in both destination Regions (us-gov-west-1 and us-gov-east-1)..

For teams building mission-critical workflows, Claude Opus 4.8 supports three composable capabilities: adaptive-thinking reasoning for analysis that separates observed facts from inferences, structured JSON output for producing validated incident records with application-side schema enforcement, and tool use through a two-call pattern that connects model reasoning to local control lookups while keeping execution authority with the application. All inference stays within the AWS GovCloud (US) physically and logically isolated boundary, served on FIPS-enabled endpoints with Nitro isolation/zero operator access.

Learn more: Bedrock Service Documentation | Opus WNP