Asian Tech NVDA 1QFY27 key takeaways

研报英文原文证据摘录

Asian Tech NVDA 1QFY27 key takeaways

reach

$3-4T by 2030, which implies sustained high-double-digit growth rate into 2029/30.

Additionally, NVDA expects neoclouds, sovereign AI, and enterprises (e.g. industrial

companies) will grow even faster than CSPs. Strong AI demand outlook bodes well for

the entire server foodchain. Key downside risks are the depletion of free cash flow and

potential financing difficulties for this buildout.

• AI infrastructure demand continues to accelerate, as evidenced by rising GPU

rental prices: Management highlighted that demand for AI infrastructure remains

strong, with data center builds by NVDA’s partners now exceeding 10MW—roughly

double YoY. Reflecting tight supply, rental prices for H100 and A100 GPUs have

increased by ~20% and ~15% YTD, respectively. Against this backdrop, most U.S. CSPs

have raised their 2026 capex guidance in 1Q26 earnings. We now estimate that aggregate

capex for the top 4 US CSPs will rise ~70% YoY. We expect the AI upcycle to persist at

least through 2027, with global semi revenue projected to grow ~18% in 2026 and ~11%

in 2027.

• LPU attach rate likely to start low, may reach 20% over time: Management

characterized LPX as a niche product built for low‑latency, high‑token‑rate use cases,

while acknowledging limitations in throughput, model‑size capacity, and context

processing due to its SRAM‑based architecture. The current addressable market is

relatively narrow—suited for providers with a portfolio of token services where

premium, high‑interactivity workloads justify the cost. Management indicated the attach

rate for LPX‑type accelerators is small today but could reach ~20% over time as premium

token services develop. We believe LPU is directionally correct given surging agentic AI