CoreWeave predominantly deploys NVIDIA's Blackwell architecture in an NBL72 configuration.

CoreWeave currently has over one gigawatt of active power deployed across its data centers for client GPU workloads.

CoreWeave builds its GPU infrastructure to NVIDIA's DGX reference specification, which it considers the most performant way to operate the hardware.

The scarcity of GPU capacity is primarily at large scale, as CoreWeave's clients focus on deployments of hundreds of thousands of GPUs, whereas finding small-scale capacity is more accessible.

CoreWeave has more than 10 clients with contracts valued at over $1 billion each.

CoreWeave finances its GPU assets through asset co-financing structures, specifically Special Purpose Vehicles (SPVs), which hold the infrastructure, data center costs, and debt agreements.

In Q4 of the most recent year, CoreWeave added twice as many new client logos as in any previous quarter, driven primarily by growth in its enterprise customer base.

Vera Rubin is the name of NVIDIA's next-generation GPU architecture that will follow the Blackwell architecture.

The performance of an NVIDIA H100 GPU is not fungible, as its performance varies significantly depending on the cloud provider's operational software and infrastructure management.

CoreWeave's asset-backed financing structures (SPVs) are designed to have a 25% contribution margin of profit that flows up to the parent company after covering debt service and operating costs.

CoreWeave's primary differentiator in GPU performance is its proprietary software stack, which manages operations, predicts failures, and swaps infrastructure to maximize uptime and efficiency.

AI labs are increasing the duration of their committed infrastructure contracts with CoreWeave, moving from 3-year terms a couple of years ago to 5-year terms currently.

Brandon McBee of CoreWeave believes a liquid market for tradable GPU compute is unlikely to emerge in the short term because GPU performance is not fungible across different cloud providers.

Inference workloads represent "well in excess of 50%" of the total infrastructure utilization on CoreWeave's platform.

CoreWeave has raised over $21 billion in financing year-to-date.

CoreWeave provides infrastructure for 9 of the top 10 AI labs in the world, excluding those in China.

Jane Street is a direct client of CoreWeave, using its platform directly rather than through an intermediary AI lab.

CoreWeave is not observing any pullback in inference demand from its AI lab clients; instead, demand remains "unrelenting."

Current 5-year contracts from AI labs with CoreWeave are structured as fixed commitments where the client cannot upgrade, change, or cancel the specified infrastructure for the full term.

CoreWeave has successfully accessed the insurance market as a source of capital for its financing needs.