Bedrock Brief 12 Aug 2026

Pleri AI

Pleri AI

— 5 min read
Share
Bedrock Brief 12 Aug 2026

Well, that escalated quickly. One minute AWS is helping Novo Nordisk shave 15 weeks off clinical documentation workflows, and the next we're reading about a 7.65-gigawatt natural gas plant in Texas that could become America's single largest CO₂ emitter. Welcome to the contradictions of AI at scale: we're curing diseases faster while simultaneously making the planet warmer faster. Amazon's new "behind the meter" power plant, authorized to release 33 million tons of greenhouse gases annually, is a stark reminder that every agent you deploy and every foundation model you fine-tune has a carbon cost someone, somewhere, is paying.

The Novo Nordisk partnership is genuinely impressive from a technical standpoint. AWS is embedding engineers directly into Novo's London facility to build AI agents that can identify drug targets, design therapies, and wrangle messy clinical, genomic, and imaging datasets. They've already cut documentation time by over 90% using Claude 3.5 on Bedrock, freeing up medical pros to focus on validation instead of paperwork. Over 25,000 Novo employees are now using Bedrock-powered chatbots for everything from info retrieval to document drafting. If you're wondering whether agentic AI can actually move the needle in the real world, pharmaceutical R&D is starting to provide some convincing receipts.

But here's the thing: as we sprint toward faster drug discovery and smarter clinical workflows, we're also building data centers that require their own dedicated power plants. Amazon pledged net-zero emissions by 2040, yet its carbon footprint keeps climbing, year after year. Microsoft is in the same boat, and both companies keep insisting they're "committed" to their climate goals while simultaneously lighting up natural gas turbines to keep the AI lights on. The irony is almost poetic. AI can help us solve humanity's hardest problems, but only if we're willing to reckon with the infrastructure costs of running it at scale.

Fresh Cut

  • You can now track which IAM users and roles are spending money on AI model inference requests through Bedrock's bedrock-mantle endpoint by tagging them with team or project labels and viewing the breakdown in AWS Cost Explorer. Read announcement →
  • You can deploy three new AI models on SageMaker: one that finds objects in images using parallel box decoding, one that simulates how software agents interact with tools and operating systems, and a 122-billion parameter model that only activates 10 billion parameters per query to save on compute costs. Read announcement →
  • NVIDIA's Nemotron 3.5 Lightning processes 410 tokens per second using only 3 billion active parameters at a time (out of 30 billion total) and handles up to 1 million tokens of context, making it useful for building AI agents that need to work with large documents or long conversations. Read announcement →
  • AWS Glue's console gets a button that opens SageMaker Unified Studio with your current IAM role, so you can query data or build pipelines without manually configuring permissions or switching between services. Read announcement →
  • Amazon's SageMaker lets you deploy FLUX.2-small-decoder (which decodes generated images 1.4x faster using less GPU memory) and gemma-4-12B (a 12-billion parameter model that handles text, images, and audio together while running on just 16GB RAM). Read announcement →
  • Three new models let you cache LLM queries by meaning to cut costs, generate code with a 12B-parameter model that only uses 2.5B per token through mixture-of-experts, or convert documents to text without traditional OCR pipelines. Read announcement →
  • Three new AI models are available for deployment: GLM-5.2 FP8 handles 1 million token contexts for full software development workflows, NVIDIA-Nemotron-Nano-12B-v2 uses a hybrid Mamba-2/Transformer architecture for 6x faster inference, and GLM-OCR converts complex documents into Markdown/JSON with only 0.9 billion parameters. Read announcement →
  • AWS launched servers in Milan with custom Intel Xeon 6 chips that run PostgreSQL 30% faster, web servers 60% faster, and AI recommendation models 40% faster than the previous generation while costing 15% less. Read announcement →
  • AI coding agents can automatically configure user authentication including OAuth 2.0 flows and JWT tokens through a new toolkit that executes AWS CLI commands with safety guardrails. Read announcement →
  • Amazon Bedrock's AgentCore platform is available in the government cloud region with short and long-term memory for AI agents, natural language policies that convert to the Cedar policy language, and a managed runtime that eliminates the need to write orchestration code. Read announcement →

The Quarry

How we built an MCP bridge to give our AgentCore-hosted AI agent access to local MCP tools

Amazon built a clever workaround to let their cloud-hosted Bedrock AgentCore agents access local tools and files on a user's laptop without opening firewall ports or requiring a VPN. The secret sauce is a browser extension that tunnels signed MCP (Model Context Protocol) messages bidirectionally over an existing WebSocket connection, then uses Chrome's native messaging API to route those calls to local MCP servers running on the machine. It's basically reverse proxy magic that keeps security teams happy while giving agents real access to your filesystem and local dev tools. Read blog →

More posts:

Core Sample

AWS Security Hub AI Inventory for Organization-Wide Visibility of AI Assets

AWS Security Hub just got an AI inventory feature that automatically discovers and catalogs all your AI workloads across managed services like Bedrock and SageMaker, plus self-hosted models running on EC2, ECS, and EKS through runtime analysis. It maps each AI asset to its underlying infrastructure (compute, networking, IAM roles, and data stores) and correlates everything with security signals from GuardDuty, so when anomalous activity pops up, you can trace it back to the exact infrastructure involved. Best part? It's included free in your Security Hub Essentials plan, giving you organization-wide visibility without opening your wallet. Watch video →

More videos:

Read more