$NVDA - NVIDIA maintains dominant position in AI infrastructure for both training and inference workloads. Customers are committing to longer contracts (5 years vs 3 years) and larger deployments, with no material demand for alternative silicon. The CUDA ecosystem remains the most efficient, scalable, and reliable platform.
$AILABS - AI labs are demonstrating strong conviction in scaling laws by demanding longer contract durations (5 years), larger deployments, and showing no signs of pullback. CoreWeave serves 9 of top 10 AI labs globally with unrelenting demand for inference infrastructure.
$DATACENTER - Data center infrastructure, particularly 'powered shells' with electricity and cooling, has become the primary bottleneck rather than GPU availability. Supply chain constraints including electricians, transformers, and backup batteries create structural scarcity with no near or medium-term solution to satiate demand.
$CORWEAVE - CoreWeave has demonstrated execution excellence with over 1 gigawatt of active power deployed, achieved investment-grade debt ratings at SOFR+225, and raised over $21B in financing. Customer diversification is progressing with financial services backlog approaching $10B and twice as many new clients added in Q4 versus prior quarters.
$CLOUDGPU - GPU compute is not fungible across providers due to operational excellence differences in goodput and model flops utilization (MFU). This prevents commoditization and creates sustainable competitive moats for best-in-class operators, with complexity increasing rather than decreasing over time.
Bearish:
$CUSTOMSILICON - Despite announcements from Microsoft and others about custom silicon, CoreWeave sees no material customer demand for non-NVIDIA infrastructure. Customers continue to commit exclusively to NVIDIA for multi-year, multi-billion dollar contracts, suggesting custom silicon efforts may not gain meaningful traction.