Jensen Huang: Future data centers will become token manufacturing plants; the AI ​​chip and infrastructure market may reach $1 trillion by 2027.

On March 16th, local time, at NVIDIA GTC 2026, NVIDIA founder Jensen Huang shared the company's overall vision for the future of the artificial intelligence industry: from next-generation AI computing architecture and data center business models for the inference era, to the software ecosystem and industry alliances built around agents, this conference presented not just upgrades to single hardware products, but a complete AI infrastructure system centered around computing power. In his speech, Huang boldly predicted that by 2027, the market size surrounding AI chips and infrastructure could reach $1 trillion. Furthermore, beyond technology, Huang proposed a new AI industry narrative: "Data centers are factories that produce tokens; inference is the workload, tokens are new commodities, and computing power equals revenue; in the future, every CEO will have to keep an eye on the efficiency of their token factory." In his view, the development of AI is experiencing a new inflection point. From chatbots to systems with inference capabilities, and then to agents capable of performing tasks, each leap in capabilities significantly increases the computing power required for a single inference, while also driving rapid growth in overall usage. Based on this trend, NVIDIA has proposed a new tiered AI service model, ranging from a free tier to an Ultra tier, corresponding to different model sizes, context lengths, and response speeds, as well as different token prices. Under this system, computing infrastructure directly determines the economic viability of AI services, while more advanced AI services require more powerful computing platforms. (AIPress)