Hyperscaler AWS, Google, and Meta each disclosed custom AI inference chips (Trainium, TPU, Cerebra successors) gaining traction in workload allocation.

NVIDIA owns the AI compute market, yet every major customer is spending billions to reduce dependence on the company. Trading between a 52-week low of $158.18 and high of $236.26, NVIDIA carries a market capitalization of $4.75 trillion. In Q1 FY27, revenue reached $81.61 billion, up 85.2% year over year, with data center revenue of $75.25 billion.

Critically, approximately half of that data center revenue comes from hyperscalers—Amazon, Google, Microsoft, and Meta—who simultaneously bankroll proprietary silicon alternatives: Amazon Trainium, Google TPU, Microsoft Maia, and Meta MTIA. This concentration represents both NVIDIA's largest market and its greatest competitive vulnerability.

Growth momentum remains strong. Management guided Q2 FY27 revenue to $91.0 billion with non-GAAP gross margins expected to hold at 75%. Networking revenue surged 199% year over year to $14.8 billion, extending NVIDIA's moat beyond GPUs into InfiniBand, Spectrum-X, and NVLink. The company is ramping Blackwell Ultra and announced Rubin. CEO Jensen Huang called the AI infrastructure buildout "the largest infrastructure expansion in human history." The board raised the quarterly dividend to $0.25 and authorized an $80 billion additional buyback.

The structural threat is clear. Amazon has disclosed that Trainium is now a multi-billion-dollar business, and every hyperscaler funding NVIDIA's data center segment also funds an alternative. Custom silicon "not only gives you a differentiation factor where you can be cheaper than competitors, but it also allows you to have some leverage over NVIDIA in negotiations." NVIDIA's NVLink Fusion pivot seeks to lock in its interconnect layer even as hyperscalers migrate compute to custom chips—a strategy that implicitly concedes the battle over GPUs.

Several headwinds compound the challenge. China's data center compute revenue is effectively zero. CFO Colette Kress acknowledged that losing this market, which NVIDIA estimates at "close to about $50 billion in the future," would be material. Supply commitments of $119 billion amplify demand risk if hyperscaler orders decelerate. Recent insider transactions show a pattern of net selling, with 16 transactions logged.