Lambda adds Confidential Bare Metal and agent safety tools at NVIDIA GTC Berlin

Confidential computing on bare metal

Lambda, which calls itself the Superintelligence Cloud, is a Gold sponsor at GTC Berlin. According to the company, the priorities it hears from customers are making AI safer, more secure and more efficient to run at scale. Workloads are moving from single training runs to always-on inference and autonomous agents, which raises questions about protecting data, controlling agents and getting more output from the same power.

The first announcement addresses data protection. Lambda says frontier model builders increasingly rely on cloud providers to train and serve models, and many prefer bare-metal access because it makes working across several providers easier. Running on infrastructure they do not own makes protecting intellectual property and sensitive data a top priority.

Lambda is now supporting Confidential Bare Metal, which extends NVIDIA Confidential Computing to dedicated third-party infrastructure. According to Lambda, hardware-rooted isolation and cryptographic attestation help protect models and data while they run, and let teams verify that the compute environment is trustworthy. Together with Confidential Containers and Confidential Virtual Machines, this allows model builders to deploy their software stack on bare-metal instances to meet their trust boundaries and operational requirements. NVIDIA Confidential Computing is also expanding across NVIDIA Grace Blackwell and NVIDIA Vera Rubin systems.

Agent safety and infrastructure monitoring

On agents, Lambda recalled that in September it joined NVIDIA to support open agent safety. It now supports the NVIDIA OpenShell Reference Architecture, which builds on the NVIDIA OpenShell secure runtime software and extends it into full-stack, industrial-scale infrastructure for agent development and operations.

Building on its NVIDIA Vera Rubin investment, Lambda says it is combining model compute, governed agent execution and independent security enforcement. Because enforcement sits outside the agent's own execution environment, teams can expand what agents do while keeping control over the tools, data and systems they can reach.

On reliability, Lambda argues that compute is not a commodity, since a single degraded component can slow or stall an entire workload as clusters grow. The company plans to deploy NVIDIA Minerva, an AI for IT operations (AIOps) platform that connects telemetry across GPUs, networking, storage, schedulers and AI workloads. A dynamic knowledge graph maps infrastructure components to running jobs, while specialized AI models detect anomalies, predict emerging failures and correlate related alerts to identify likely root causes. Lambda's stated goal for users is fewer workload interruptions, faster fixes, higher goodput and more tokens produced.

More tokens from the same power budget

Power is becoming a major constraint on AI infrastructure, according to Lambda. In a proof of concept on NVIDIA HGX systems, the company used NVIDIA DSX MaxLPS software to run 19 nodes within the same aggregate power budget as a 16-node baseline. Lambda reports about 24% higher cluster-wide token throughput and about 23% higher performance per watt. The results are described in an NVIDIA case study with Lambda.

These figures come from a proof of concept and are reported by the company, not from a production deployment.

What it means for AI teams

Lambda frames the announcements as a package covering security, agent safety, reliability and efficiency, meant to help AI teams protect workloads, keep control of agents and get more from their infrastructure. For model builders that use several cloud providers, Confidential Bare Metal is aimed at keeping bare-metal flexibility while adding verifiable isolation.

Visitors to GTC Berlin can find Lambda at booth 7022 all week. The company says it will show NVIDIA Nemotron and OpenShell in demos, discuss capacity, deployment and production-readiness decisions, and answer questions about its offer for NVIDIA Inception members.