NVDA

ACCE thesis

NVIDIA designs and sells GPUs, networking hardware, and the CUDA software stack that has become the de facto operating system for AI model training and inference at scale. The switching costs embedded in CUDA-after 15+ years of developer adoption-mean hyperscalers and frontier AI labs don't just buy H100s and B200s, they build entire engineering cultures around NVIDIA's toolchain. Blackwell demand has outstripped supply since launch, with management guiding data center revenue to continue accelerating through fiscal 2026 even as the consumer GPU segment remains secondary to the story. At a forward PE of 17.7x on 85% revenue growth and 114% ROE, the market is pricing in a material deceleration that hasn't shown up in the order books yet-Microsoft, Google, Meta, and Amazon are all on record committing to capex expansions anchored to NVIDIA silicon. The gross margin holding at 74% through a full product transition to Blackwell is the sleeper data point here; it tells you pricing power is intact and the competitive response from AMD and custom silicon hasn't dented the moat in any measurable way.

Snapshot