Radiant

Running GPU infrastructure for multiple customers simultaneously requires keeping their workloads isolated and securely preparing servers when they change hands. At Radiant, we securely wipe, reconfigure and verify each server before its next deployment. Across thousands of GPUs, that time adds up. Automating the process gets capacity back into use sooner while ensuring every security check passes. In this joint post NVIDIA’s Jake Bloom and Erez Kirson, alongside Radiant’s João Coelho explain how Radiant FlightDeck works with NVIDIA Infra Controller (NICo) and BlueField DPUs to extend tenant isolation to every server and automate bare-metal activation and hand-over. They offer a closer look at how the Radiant and NVIDIA teams have worked together to enable secure multi-tenancy throughout the compute lifecycle, from initial deployment to hand-over between customers. Read the guest post from the NVIDIA team. Link in the comments

The bottleneck in AI isn't the model. It's the megawatt. Most infrastructure discussions focus on chips. But chips without power, land, and capital are just expensive paperweights. The real constraint is the full stack — permitted land connected to reliable power, capital structured to match infrastructure timelines, compute deployed at scale, and software intelligent enough to orchestrate it all. Strip any one layer out and the system fails. This is why cost of compute is ultimately a function of cost of capital. Institutional capital deployed at the source of power changes the unit economics of AI entirely. Long-duration capital matched to long-duration infrastructure removes the arbitrage that keeps compute expensive. GPU utilization climbs. Idle capacity shrinks. The marginal cost of intelligence approaches zero. Radiant's thesis isn't speculative. It's structural: land → power → compute → intelligence, owned vertically, deployed at speed. The companies that control this chain don't just reduce cost — they compress the time between idea and inference to near zero. Abundance isn't a feature. It's an infrastructure outcome. https://lnkd.in/dRSdwqsK #AIInfrastructure #AIAbundance #VerticalIntegration

Well that came together quickly....looking tight. Again, we need a wider angle lens. #buildersbuild

Radiant is extending its NVIDIA Vera Rubin deployment with the NVIDIA #OpenShell Reference Architecture. Our cloud customers get model compute, agent execution and security controls in a single stack. The work breaks down like this: 1. NVIDIA Vera Rubin NVL72 runs model inference and reasoning. 2. NVIDIA Vera CPUs run agent harnesses and the NVIDIA OpenShell secure runtime. 3. NVIDIA BlueField-4 DPUs run NVIDIA Sentry, enforcing policy from a hardware-isolated security domain that operates independently of the host. Enforcement sits outside the system the agent runs on. It does not depend on the host staying trustworthy. Our aim is to let customers deploy agentic services with clear boundaries on what agents can access and do, backed by coordinated runtime governance and independent security enforcement. Learn more here: https://lnkd.in/gnsz3nJP

The global AI race is hitting a physical wall: the power grid. ⚡️Morgan Stanley is projecting a massive 49 GW shortfall in U.S. data center capacity by 2028, which makes the traditional 3-to-4-year timeline for building AI factories unviable. So, how do we fix the infrastructure bottleneck? In a great new piece, Manesh Patel PhD CEng MIET outlines how Radiant is fundamentally rewriting the deployment playbook, compressing timelines from years to a few quarters. Instead of the flawed, sequential approach of waiting on grid interconnects and piecing together hardware, Radiant is leveraging a vertically integrated utility model to parallelize delivery: 💡 Power First: Pre-secured powered land and behind-the-meter generation to bypass grid queues. 💰 Structural Advantage: Infrastructure-grade capital to absorb upfront costs and deliver predictable pricing. ❄️ High-Density Design: Purpose-built cooling architecture designed for frontier GPUs such as NVIDIA Rubin and Blackwell Ultra. 🧠 Software Intelligence: In-house orchestration achieving maximum GPU utilization. We can't afford to wait for the grid. For nations, enterprises and telcos building AI today, infrastructure deployment velocity is the ultimate strategic advantage. Highly recommend reading the full breakdown here: https://lnkd.in/gfhUifyc #AIFactories #DataCenters #AIInfrastructure

Training a model with reinforcement learning is the fun part. Getting the checkpoint you’re happy with into production is where things tend to go off the rails. Baseten has closed that gap. With Baseten Loops, teams run RL on open source models through a Python SDK, then ship the trained checkpoint to production inference with a single command. Radiant runs the managed GPU infrastructure, Kubernetes, networking and storage underneath, so Baseten can scale post-training reliably and its customers can spend their time on the model instead of the cluster. Great team building a very useful product, and we’re glad to be part of it. Customer story in the comments.