NVIDIA (英伟达) launched the AI era — now it must "hold onto its gains," The New York Times says

A dominant run faces new pressure

The New York Times argues that NVIDIA (英伟达) helped kick off the modern AI boom by turning graphics processors into the de facto hardware for training large neural networks. Its GPUs, the CUDA software stack and an ecosystem of data‑center partners gave the company a sprawling competitive moat. But dominance breeds challengers. Can NVIDIA hold onto its gains as rivals sharpen their focus and customers demand different trade‑offs?

Competition, customers and geopolitics

NVIDIA's lead is being tested on multiple fronts: established rivals and cloud providers are developing alternative accelerators and system architectures; startups are building domain‑specific chips; and large cloud customers want more end‑to‑end services, not just raw silicon. Geopolitics matters too. It has been reported that U.S. export controls and broader trade tensions have complicated access to advanced AI hardware for some Chinese customers, while Beijing's industrial policy and firms such as Huawei (华为) and Baidu (百度) are accelerating domestic chip and model efforts to reduce reliance on foreign suppliers.

What NVIDIA must do next

To stay ahead NVIDIA needs to keep delivering raw performance and to extend its software and services lead — from compiler and tooling advantages to tightly integrated systems that reduce customer switching costs. Reportedly, the strategic battleground is shifting from pure chip performance to a combination of software, systems, data‑center partnerships and supply‑chain resilience. That matters because buyers are increasingly judging vendors on the total cost and reliability of running large models, not just FLOPS per watt.

Why this matters beyond Wall Street

For Western readers unfamiliar with China’s tech landscape: the outcome will shape who controls the "external brain" — the infrastructure that serves and moderates AI worldwide. Winners will influence which platforms set standards, who gets privileged access to model‑ready datasets, and how export controls, data‑flow rules and commercial APIs evolve. The question is not only which company wins, but which regulatory and industrial ecosystem they win within.