CoreWeave is a U.S.-based cloud computing services company that specializes in providing GPU-based (Graphics Processing Unit) infrastructure to AI developers and enterprise customers. The original founders of the firm were planning to use their computers to mine cryptocurrencies, but decided to pivot into cloud computing in 2019, utilizing their existing inventory of GPUs to run high-performance AI workloads. Today, CoreWeave operates multiple datacenters in both the United States and Europe and supports many of its top-tier clients with flexible, scalable compute solutions.
Founded as Atlantic Crypto in 2018 by Michael Intrator, Brian Venturo, Brannin McBee, and Peter Salanki to develop a cryptocurrency mining operation using GPUs.
After the 2018 crypto crash, CoreWeave was renamed and began developing GPU-based cloud infrastructure for businesses.
Announced investment of $100 million in Nvidia H100 chips and initiated a new accelerator program to offer AI startup access to compute credits for cloud-based AI development.
Constructed a $1.6 billion supercomputer datacenter in Plano, Texas, for Nvidia that has been referred to as the world's fastest AI supercomputer.
Launched its London headquarters and established its first UK data centers and hired major executives to lead its global operations.
Increased number of data centers to 32 and increased the amount of available GPUs to 250,000 in total across all data centers in the US and Europe.
32 owned data centers in US and Europe with 250,000+ GPUs including dedicated supercomputer facilities
Specialized GPU cloud platform for AI model training, inference, and high-performance computing workloads with proprietary chip management software
Based on company descriptions from Wikipedia, CB Insights, and official positioning as NVIDIA GPU specialists
| Service | Cost | Details | Source |
|---|---|---|---|
| GPU Compute (NVIDIA H100) | Custom contract pricing | Premium pricing for GenAI workloads, higher revenue per GPU than hyperscalers | CoreWeave investor guidance |
| AI Infrastructure Capacity | $55.6B backlog across multi-year contracts | Long-term commitments with OpenAI, Meta, Microsoft through 2029-2030 | Q4 2025 earnings |
| Enterprise Contracts | Custom quotes | Dedicated GPU clusters for model training and inference | Customer announcements |
| Feature | CoreWeave | AWS | Azure | Google Cloud |
|---|---|---|---|---|
| GPU Specialization | AI-native GPU clusters | General cloud + AI | General cloud + AI | General cloud + AI |
| NVIDIA Partnership | Exclusive 5GW AI factories | Standard access | Standard access | Standard access |
| GenAI Workload Focus | Premium pricing | Competitive pricing | Competitive pricing | Competitive pricing |
| Revenue Backlog | $55.6B | |||
| Customer Concentration | OpenAI, Meta, Microsoft | Broad enterprise | Broad enterprise | Broad enterprise |
| Infrastructure Scale | 250k+ GPUs planned | Millions servers | Millions servers | Millions servers |
| AI Factory Buildout | 5GW by 2030 w/NVIDIA | Custom regions | Custom regions | Custom regions |
| Pricing Power | Higher per GPU revenue | Volume discounts | Volume discounts | Volume discounts |
| Execution Risk | Power/construction delays | Mature operations | Mature operations | Mature operations |
Premium pricing model and specialized NVIDIA GPU hardware provide higher revenue per GPU for CoreWeave when compared to AWS's broader cloud service offerings at lower prices.
For high performance AI training/inference, we recommend CoreWeave. For general cloud and AI hybrid requirements, we recommend AWS.
Both AWS and CoreWeave serve OpenAI, however, CoreWeave provides dedicated capacity with premium economics vs Azure which has larger enterprise relationships and more mature operations but has been under margin pressure in pure AI compute.
CoreWeave has an advantage in terms of high-performance AI factories that utilize GPU dense architectures. We believe that Azure has an advantage in terms of being able to support the Microsoft integrated ecosystem.
Google has an advantage over other companies when it comes to TPUs for specific workloads while CoreWeave dominates the NVIDIA GPU space. The large $55 billion backlog for CoreWeave indicates that they have captured significantly more of the AI specific demand.
CoreWeave has the largest share of the NVIDIA GPU market and Google Cloud has the largest share of the TPU optimized inference market.
CoreWeave has the ability to grow much faster than AWS with the type of hyperscale level contracts that are typically needed for a company of their size and scale. Additionally, because of the NVIDIA investment and the available power capacity, CoreWeave also has a significant infrastructure advantage.
For enterprise scale AI requirements, we recommend CoreWeave. For startup experimentation with AI, we recommend Lambda.
| Hardware Type | Primary Use Case | CoreWeave Support | Specialization |
|---|---|---|---|
| NVIDIA GPUs | Training and inference at scale | Primary platform - early access to cutting-edge | General-purpose accelerated computing |
| High-Performance Storage | AI model and dataset management | Purpose-built AI Object Storage (>$100M ARR) | Optimized storage layer for AI workloads |
| Workload Type | CoreWeave Capability | Key Features | Performance Benefit |
|---|---|---|---|
| Model Training | Native support across stack | Orchestration via Slurm and Kubernetes, managed services | Train and fine-tune models faster at scale |
| Model Inference | Production-ready infrastructure | Day 1 production workloads, high-performance clusters | Deploy models faster with lower total cost of ownership |
| AI Agent Training | OpenPipe acquisition integration | Dedicated platform for training AI agents | Specialized infrastructure for agent development |
| Development and Validation | CoreWeave ARENA lab | Production AI lab for testing before commitment | Assess performance, scaling, and cost pre-production |
