Age of AI Factories: Reconstructing Industrial Competition Logic

NVIDIA CEO Jensen Huang delivered a speech at GTC, focusing on the arrival of the “AI Factory” era. In this new era, data centers will shift from merely storing files to becoming factories that produce Tokens (the smallest unit of processing in large models, used for understanding and generating content). Every cloud service provider and AI company will, in the future, measure their success by “Token Factory Efficiency.”

In recent years, the core narrative of the AI industry has revolved around “model competitions”: from hundreds of billions to trillions of parameters, from single-language models to multimodal models. Companies have fallen into the misconception that higher parameters mean stronger capabilities, neglecting a key issue—models that lack cost control and efficiency are ultimately difficult to scale commercially. Huang’s speech aims to break this imbalance: data centers will no longer be just “warehouses” for storing files but will produce Tokens in bulk, efficiently, and at low cost—these digital “bulk commodities” carrying AI capabilities will become the core carriers for future AI applications. “Token Factory Efficiency,” measured by the number of Tokens produced per unit of computing power, the cost of Token generation, and iteration speed, will replace “model parameters” as the key operational metric for cloud providers and AI companies.

This shift marks the transition of the AI industry from the “technological exploration phase” into the “industrialization phase.” More fundamentally, Huang’s vision of the “AI Factory” era is not just a technological and business model upgrade but a profound change in the development logic of the AI industry—transforming AI from “black technology in labs” into a practical productivity tool across various industries. Tokens, as the product of the “AI Factory,” with layered pricing and efficient supply, will break down the cost barriers for AI deployment, enabling AI to truly integrate into finance, healthcare, manufacturing, transportation, and other fields. The competition for “Token Factory Efficiency” will drive the industry from “rough development” to “refined operation,” pushing companies to continuously invest in computing power optimization, algorithm iteration, and ecosystem collaboration, ultimately achieving high-quality development of the AI industry.

However, the arrival of the “AI Factory” era also presents challenges. Standardized production models may limit technological innovation diversity; over-reliance on a single ecosystem could lead to industry imbalance; and issues related to the security, compliance, and ethics of Tokens will become more prominent as production scales up. Nonetheless, the future of the AI industry will no longer be a solitary “battle of models” but a collaborative industrialization across the entire sector.

(Edited by: Zhang Yan)

【Disclaimer】This article reflects only the author’s personal views and is not related to Hexun.com. Hexun.com remains neutral regarding the statements and opinions expressed herein and does not guarantee the accuracy, reliability, or completeness of the content. Readers are advised to use this information for reference only and bear all responsibilities themselves. Email: news_center@staff.hexun.com