The NVIDIA Blackwell era of AI is here

Be among the first to access NVIDIA Blackwell Ultra GPUs NVIDIA Blackwell platform brings groundbreaking advancements for generative AI and accelerated computing with up to 30x faster real-time LLM performance.

Reserve capacity now NVIDIA HGX B300 now available

NVIDIA HGX B300 on CoreWeave:
Built for agentic AI

CoreWeave’s NVIDIA HGX B300-based instances are purpose-built to run large reasoning and agentic AI systems. By combining expanded memory, full line-rate networking across large GPU clusters, and deep observability into GPU and interconnect performance, you can deploy, operate, and scale agentic AI with confidence.

‍

Expanded memory and inference capacity

With 270GB of HBM3e GPU memory, NVIDIA HGX B300 offers 50% more memory per GPU and 50% higher NVFP4 performance than NVIDIA HGX B200. It’s built to supercharge your AI reasoning and agentic systems.

Frictionless connectivity

CoreWeave’s non-blocking NVIDIA Quantum-X800 InfiniBand networking delivers 1.95x higher measured collective bandwidth on DeepSeek-R1. Get ready for faster GPU synchronization and more consistent performance (vs. NVIDIA HGX H200) .

Actionable observability

With CoreWeave ARENA, services like W&B Weave, and Kubernetes-native orchestration via CoreWeave Kubernetes Service and SUNK, you can reduce iteration cycles, surface issues faster, and move from run to results with predictable speed at scale.

Setting a new standard with NVIDIA GB300 NVL72

NVIDIA GB300 NVL72 on CoreWeave delivers an unprecedented 50x increase in output for AI inference workloads, enabling rapid deployment of larger, more sophisticated AI models when compared to the NVIDIA Hopper platform.

‍

Next-level AI reasoning performance

NVIDIA GB300 NVL72 delivers up to 10x improved user responsiveness, 5x greater throughput per watt, and 1.5x more FP4 FLOPs compared to previous-generation architectures.

Expanded GPU memory

With up to 21TB of HBM3e high-bandwidth GPU memory per rack, the NVIDIA GB300 NVL72 enables larger batch sizes and handles complex, memory-intensive AI models more efficiently.

Next-generation connectivity

NVIDIA GB300 NVL72 features fifth-generation NVIDIA NVLink™ technology with 130TB/s aggregate bandwidth, paired with NVIDIA Quantum-X800 InfiniBand switches and ConnectX-8 SuperNICs for dedicated 800 Gb/s connectivity per GPU. This ensures seamless GPU-to-GPU communication and maximum efficiency.

Maximize your potential with the NVIDIA GB200 NVL72

CoreWeave’s NVIDIA GB200 NVL72-powered cluster is built on the NVIDIA GB200 Grace Blackwell Superchip, fifth-generation NVIDIA NVLink with NVLink switch trays, and NVIDIA Quantum-2 InfiniBand networking. It’s purpose-built to meet the demands of all your next-generation AI workloads.

‍

Exponentially accelerate real-time inference and AI training

NVIDIA GB200 NVL72 clusters on CoreWeave deliver up to 1.4 exaFLOPS of AI compute power per rack. They enable up to 4x faster training and 30x faster real-time inference of trillion-parameter models compared with previous-generation GPUs.

Advancing data processing and physics-based simulation

NVIDIA Grace Blackwell GB200 NVL72—with the tightly coupled CPU and GPU in the GB200 Superchip—brings new opportunities in accelerated computing for data processing and engineering design and simulation.

Accelerated networking platforms for AI

Paired with NVIDIA Quantum-X800 InfiniBand, Spectrum-X Ethernet, and BlueField-3 DPUs, GB200 delivers unprecedented levels of performance, efficiency, and security in massive-scale AI data centers.

NVIDIA Blackwell Architecture

‍

New Class of AI Superchip

Second-Gen Transformer Engine

Faster and wider fifth-gen NVIDIA NVLink interconnect

High-performance confidential computing and secure AI

NVIDIA Decompression Engine to identify potential faults that may occur early on to minimize downtime

NVIDIA RAS Engine provides in-depth diagnostic information that can identify areas of concern and plan for maintenance

Scale your AI ambitions with the NVIDIA HGX B200

The NVIDIA HGX B200 is designed for some of the most demanding AI, data processing, and high-performance computing workloads. Get up to 15X faster real-time inference performance.

Real-time large language model inference

The second-generation Transformer Engine in the NVIDIA Blackwell architecture features FP4 precision enabling a massive leap forward in accelerating inference. The NVIDIA HGX B200 achieves up to 15X faster real-time inference performance compared to the Hopper generation for the most massive models such as the GPT-MoE-1.8T.

Supercharged AI training

The faster, second-generation Transformer Engine—which also features FP8 precision—enables the NVIDIA HGX B200 to achieve up to a remarkable 3X faster training for large language models compared to the NVIDIA Hopper generation.

Advancing data analytics

With support for the latest compression formats, such as LZ4, Snappy, and Deflate, NVIDIA HGX B200 systems perform up to 6X faster than CPUs and 2X faster than NVIDIA H100 Tensor Core GPUs for query benchmarks using Blackwell’s new dedicated Decompression Engine.

NVIDIA RTX PRO 6000 Blackwell Server Edition

Accelerate your GenAI workloads with NVIDIA RTX PRO 6000-based instances, now available first on CoreWeave.

Reserve capacity now

Fast track inference and fine-tuning at scale

Optimized for efficient inference and fine-tuning of small to medium-sized models, these instances combine 96GB of GDDR7 memory and 1597 GB/s of GPU memory bandwidth with powerful Intel Platinum CPUs. Designed to deliver exceptional performance and throughput.

Full performance on a purpose-built platform

CoreWeave’s bare-metal deployment, proactive node health checks, and automated failure remediation ensure rock-solid reliability for latency-sensitive workloads. Our resilient infrastructure keeps your models running smoothly, without the performance trade-offs of legacy cloud platforms.

Powering multi-workload acceleration

Built on the groundbreaking NVIDIA Blackwell architecture, the NVIDIA RTX PRO™ 6000 Blackwell Server Edition delivers a powerful combination of AI and visual computing capabilities. Accelerate agentic AI, physical AI, scientific computing, 3D graphics, rendering, and more.

When speed and efficiency matter, CoreWeave is your partner

Get to market faster with our fully managed cloud platform, built for AI workloads and optimized for efficiency. We get your cluster online quickly so that you can focus on building and deploying models, not managing infrastructure.

Accelerated time-to-market

CoreWeave is proud to be one of the first major cloud providers to bring up an NVIDIA GB200 cluster, continuing our tradition of bringing state-of-the-art accelerated computing cloud solutions at lightning-fast speed and scale.

Fully-managed infrastructure

When you’re burdened with infrastructure overhead, you have less time and fewer resources to focus on building your products. CoreWeave’s fully-managed cloud infrastructure frees you from these constraints and empowers you to get breakthroughs to market faster.

Optimize ROI

CoreWeave ensures your valuable compute resources are only used to run value-adding activities like training, inference, and data processing. This means you’re getting the best performance out of your resources without sacrificing performance.