Keep pulling the thread on Neil Tiwari.
CoreWeave began training models for OpenAI in early 2023, which marked a significant growth phase for the company due to the massive compute requirements for large language models.
Capital expenditure on AI compute and infrastructure from hyperscalers is projected to be between $660 billion and $690 billion in 2026.
Total capital expenditure on AI compute and infrastructure is expected to scale to trillions of dollars over the next several years.
Early debt financing structures for AI compute build-outs were structured as Special Purpose Vehicles (SPVs) where the primary collateral was contracted cash flows from investment-grade customers, not the GPUs themselves.
The primary bottleneck for scaling AI compute has shifted from chip availability to the infrastructure required to operate them, including power, data centers, and skilled labor.
Data from SemiAnalysis indicates that NVIDIA's Blackwell GPUs are 90 to 100 times more efficient for inference tasks compared to Hopper GPUs, exceeding NVIDIA's own claim of a 30x improvement.
For AI application companies, compute is the single largest component of their Cost of Goods Sold (COGS).
Power availability is the primary limiting factor for the continued build-out of AI compute capacity.
In the next 6 to 12 months, the primary bottlenecks for building new data center and power infrastructure are shortages of structural steel, electricians, substations, transformers, and air chillers.
Sovereign nations, including the United States, increasingly view AI compute capacity as a matter of national security.
Magnetar Capital is an alternative asset manager with $22 billion in assets under management.
Magnetar Capital's first investment in CoreWeave occurred in 2021, as the company was pivoting from Ethereum mining to high-performance computing.