黄仁勋最新访谈实录：输入电子，输出Token，中间就是英伟达

Interview highlights

Jensen Huang (黄仁勋), founder and CEO of NVIDIA (英伟达), summed up his company’s role in the AI era in blunt, memorable terms in a recent interview with Ifeng: “input electrons, output tokens, and NVIDIA is in the middle.” He argued that modern artificial intelligence is fundamentally a transformation of electrical signals into discrete model outputs — and that GPUs and the associated software stack are the conversion layer. Is NVIDIA simply selling chips? Not according to Huang. He framed the company as a combined hardware‑software platform that accelerates large language models, data‑centre workloads and the nascent generative AI economy.

Technical and market context

What does “electrons to tokens” mean for Western readers unfamiliar with China’s tech scene? Electrons refer to the raw electrical work performed by semiconductors; tokens are the subunits of language models — the words, symbols and embeddings these models output. Huang emphasised co‑design: chips, firmware and developer tools like CUDA work together to turn raw compute into usable AI outputs. That has helped NVIDIA dominate data‑centre GPU sales and become indispensable to cloud providers and AI labs worldwide.

Geopolitical implications

Huang’s comments arrive against a backdrop of tightened US export controls on advanced chips and intense competition to build domestic alternatives. It has been reported that NVIDIA has continued to ship certain product lines to China under licence, while the firm’s highest‑end accelerators are restricted by policy, reportedly limiting some Chinese customers’ access. Chinese cloud and AI companies such as Baidu (百度) and Alibaba (阿里巴巴) rely heavily on GPU capacity, fueling Beijing’s push for indigenous chips and chip design talent. The result? A strategic tug‑of‑war where commercial engineering and geopolitics collide.

What it means going forward

If NVIDIA is the “middle” of today’s AI stack, then the industry question is clear: how long will that middle hold? For developers, customers and regulators on both sides of the Pacific, Huang’s distinction between electrons and tokens underscores why hardware, software and policy decisions now determine who leads the next wave of AI innovation.