Nvidia’s AI chip lead faces a broader field of rivals

Nvidia remains the leading AI hardware supplier, with data center systems built around Ampere, Hopper, Blackwell and the next Vera Rubin platform. Its strength comes from GPU flexibility, cloud availability, DGX Cloud and a mature CUDA software ecosystem, while AMD is gaining traction with Instinct accelerators and Intel is repositioning around rack-scale AI systems after cancelling Falcon Shores for Jaguar Shores.

The market is splitting between general-purpose GPUs and custom ASICs. GPUs remain attractive for varied training and inference workloads, but ASICs from Google, AWS, Groq, Cerebras and SambaNova promise better performance per watt for targeted tasks. TrendForce projects custom ASIC shipments from cloud providers to grow 44.6% in 2026, compared with 16.1% for GPU shipments, reflecting hyperscalers’ growing investment in proprietary silicon.

Cloud providers and startups are pushing alternatives across inference, edge deployments and large-model serving. AWS uses Trainium and Inferentia, Google is advancing TPU generations including Ironwood, and OpenAI is working with Broadcom and TSMC on custom chips. TSMC remains central to the sector through 5nm and 3nm manufacturing, while U.S. export rules and a 25% tariff on Taiwan-made chips routed for testing are encouraging Chinese suppliers such as Huawei, Alibaba, Cambricon and Baidu to accelerate domestic development.