CoreWeave Advances AI-Native Cloud Platform for the Next Phase of Production-Scale AI

Key Terms

reinforcement learning medical

A type of artificial intelligence that learns by trial and error, receiving feedback from its actions to favor choices that lead to better outcomes. Think of it like a salesperson learning which pitches close deals by trying different approaches and keeping the ones that work. For investors, reinforcement learning matters because it can power smarter trading systems, optimize business operations, or improve products—potentially boosting efficiency and profits while also introducing model and execution risks.

inference technical

Inference is the process of drawing a conclusion from available evidence or data, like a detective piecing together clues to form a likely story. For investors it matters because these judgments turn raw reports, test results, or market signals into expectations about future performance, risk, or regulatory outcomes—so how someone infers from the same facts can change investment decisions and valuation.

infiniBand technical

Infiniband is a high-speed data transport technology used inside data centers to move large amounts of information quickly and with very little delay, often used for servers, storage and computing clusters. Investors should care because it acts like a multilane expressway for data—companies that build, use or support such fast networks can gain competitive advantages in cloud services, high-performance computing and AI workloads, which can affect costs and revenue potential.

multimodal technical

Multimodal describes an approach, product, or system that uses two or more different types of inputs, methods, or channels — for example combining text, images and audio in a technology product, or blending drugs, devices and therapy in medical care. For investors, multimodal solutions can broaden market reach and competitive differentiation but also add development cost, operational complexity and regulatory hurdles; think of it like a hybrid car that offers more capabilities but requires more parts and oversight.

mlperf technical

A standardized set of performance tests for machine learning systems that measures how fast and efficiently hardware and software can train and run AI models. Like a car mileage test for computers, MLPerf lets investors compare different vendors on speed, energy use and cost for AI workloads, helping predict which technologies will be more competitive, scalable and profitable as demand for AI grows.

New capabilities to develop, deploy, and iterate AI faster for mission critical workloads across financial services, robotics, and industrial use cases

SAN JOSE, Calif.--(BUSINESS WIRE)-- CoreWeave, Inc. (Nasdaq: CRWV), The Essential Cloud for AI™, today announced an expansion of its purpose-built AI cloud platform at NVIDIA’s GTC conference. The expansion brings NVIDIA HGX B300 to the CoreWeave Cloud, unlocking a new generation of performance for AI workloads, alongside new Weights & Biases capabilities that streamline reinforcement learning (RL) and agent development workflows.

NVIDIA HGX B300, now generally available on CoreWeave.

As the industry shifts from large-scale training to iterative RL, infrastructure requirements are rapidly evolving. By combining the latest hardware with CoreWeave’s AI optimized cloud services and advanced development workflows, CoreWeave is closing the gap between training and production to support the next generation of self-improving agents and physical AI workloads.

“The next phase of AI is being defined by how efficiently AI systems can run and scale in production,” said Michael Intrator, CEO, co-founder, and chairman at CoreWeave. “By pairing the massive compute power of NVIDIA’s latest hardware with CoreWeave’s cloud services, we’re enabling enterprises to build and refine autonomous agents faster and more reliably than ever before. This expansion reinforces our position as the essential partner for any organization navigating the complexities of frontier-scale AI.”

“The era of AI is shifting from training models to operating agents at scale,” said Jensen Huang, founder and CEO, NVIDIA. “CoreWeave is a world-class new generation AI-Native cloud. We are thrilled to partner with them to build out NVIDIA computing infrastructure to power the world’s AI.”

Purpose-Built for Agentic AI: NVIDIA HGX B300
CoreWeave Cloud unlocks NVIDIA HGX B300 performance at scale, marking a major step forward for frontier and agentic workloads. With the general availability of NVIDIA HGX B300, part of the NVIDIA Blackwell Ultra platform, customers will have access to:

• Performance Leap: NVIDIA HGX B300 is designed for AI reasoning and inference with enhanced compute and increased memory.
• Massive Memory: Featuring 2.1 TB of HBM3e memory—a 50% increase over HGX B200 instances—enabling teams to run long-context inference with low latency and train models with 100B+ parameters on a single node.
• Unprecedented Bandwidth: Next-generation NVIDIA Quantum-X800XDR InfiniBand support doubles node-to-node bandwidth, eliminating interconnect bottlenecks.
• Liquid-Cooled Reliability: Every HGX B300 server on CoreWeave Cloud is managed by state-of-the-art liquid cooling to eliminate thermal throttling and allow sustained peak performance.

“We’ve already run production workloads with CoreWeave on NVIDIA HGX B200, and that experience built real confidence in their ability to operate at scale,” said Aman Sanger, Co-Founder of Cursor, an AI-powered code editor. “We’re focused on collaborating with companies who deliver predictable performance, operational reliability, and ongoing support as our requirements evolve. As we move toward HGX B300, that proven operating model gives us confidence to focus on building more capable AI code generation systems rather than infrastructure risk.”

CoreWeave also expects to be among the first cloud providers to deploy the NVIDIA Vera Rubin NVL72 platform and NVIDIA Vera CPU rack in production in the second half of 2026. This expansion will further support large-scale inference, reasoning, and the most demanding agentic AI applications.

Direct "On-the-Job" Training for Agents with Environment-Free RL
RL is the most effective method for post-training LLMs on agentic tasks, but it traditionally requires costly, time-consuming, and difficult-to-model simulated environments. Weights & Biases now addresses this with environment-free training in its Serverless RL offering, eliminating the need to build simulated environments, and allowing agents to learn through continual "on-the-job" training.

• Efficiency: Achieve 1.4x faster training at up to 40% lower cost compared to self-managed clusters.
• Superior Results: Deliver inference with up to 5x lower cost and up to 60x lower latency without quality loss.

Production Agent Evaluations Accelerate Agent Development
Research evaluations often rely on limited datasets and synthetic scenarios, leaving agents unprepared for real-world production. This gap makes it difficult to feed production insights back into the research loop, slowing time-to-market and resulting in agents that fail to meet expectations.

Production agent evaluations in W&B Weave close that gap, connecting research and production, enabling teams to learn from real user interactions. It automatically detects production failure modes, alerts researchers, and feeds insights back to improve evaluation datasets and accelerate iteration. By closing this loop, W&B Weave helps teams ship more reliable agents faster.

• Agent Self-Improvement: Incoming traces are automatically scored by LLM judges. Lower-quality outputs are routed for human review and turned into test cases, making evaluation suites smarter over time. Specialized classifier models then analyze the production traces and identify failure modes.
• Accelerated Development: Smart alerts notify when new failure modes are detected, allowing customers to refine agents in production using data from real user interactions and continuous monitoring.

Accelerating Robotics AI Development
W&B Models now allows robotics and embodied AI customers to track, compare, and reproduce multimodal experiments faster. Teams can log training metrics, simulation outputs, and video in a single workspace, and use a new comparison panel to evaluate up to four images or videos side-by-side. In collaboration with NVIDIA, Weights & Biases is releasing two blueprints for scalable training of RL and Vision-Language-Action (VLA) models using NVIDIA Isaac Lab simulations. These blueprints outline a full-stack workflow to help teams quickly start training and fine-tuning embodied AI models.

Monitor Training Anywhere With the First Weights & Biases Mobile App
To enable AI pioneers to manage these massive production runs from anywhere, CoreWeave introduced the first iOS app designed specifically for AI model development. The Weights & Biases iOS app provides a mobile-first experience to track training runs in real time, allowing researchers to spot issues early and make confident decisions without being tethered to a desktop.

CoreWeave’s AI cloud delivers industry-leading performance and efficiency through an end-to-end technology stack optimized for AI workloads. It has achieved top MLPerf benchmark for AI workloads, Platinum rankings in both SemiAnalysis ClusterMAX™ 1.0 and 2.0, and NVIDIA Exemplar Cloud validation for training and inference on GB200 NVL72. Visit us at booth #913 at NVIDIA GTC to learn more about today’s announcements.

About CoreWeave
CoreWeave is The Essential Cloud for AI™. Built for pioneers by pioneers, CoreWeave delivers a platform of technology, tools, and teams that enables innovators to move at the pace of innovation, building and scaling AI with confidence. Established in 2017, CoreWeave completed its public listing on Nasdaq (CRWV) in March 2025. Learn more at www.coreweave.com.

View source version on businesswire.com: https://www.businesswire.com/news/home/20260316504524/en/

press@coreweave.com

Source: CoreWeave, Inc.