Huang Renxun: AI enters the "work" era, tokens are turning into "money"

At 2 a.m. Beijing time on March 17, the “AI World Spring Festival” NVIDIA GTC 2026 conference officially kicked off in San Jose, California.

During the over two-hour keynote, NVIDIA CEO Jensen Huang discussed not only chip specifications but also a blueprint for “where the money flows.”

After unveiling a “trillion-dollar” vision, the market responded immediately. NVIDIA’s stock price surged nearly 5 intraday, and the Nasdaq also rose nearly 2%.

However, Huang didn’t just make promises this time; he explained the underlying business logic: AI is shifting from “training” to “inference,” from “generation” to “doing work.” At this critical turning point, demand for computing power is being amplified.

Previously, Dongwu Securities predicted that NVIDIA GTC 2026 would focus on four core sectors—chips, intelligent agents, optical modules, and power—becoming the main investment theme for future AI. All of this is driven by a common underlying logic: tokens are becoming the hard currency of the AI era.

As Huang said, “In this token factory, your throughput and token generation speed will directly translate into your revenue next year.”

The battle for who is the “Token King” has just begun.

01

The AI inference turning point has arrived

In his speech, Huang reviewed three major events that have driven the AI wave over the past two years: the advent of ChatGPT and generative AI, the rise of inference AI, and the birth of Claude Code and intelligent agents.

Huang explained, “We are no longer just asking AI ‘what,’ ‘where,’ ‘when,’ ‘how.’ We are now letting it create, execute, and build. It has learned to use tools, access your context, and read files. It can intelligently break down complex problems, reason step by step, reflect, and ultimately not only find answers but also perform tasks.”

“Every part of AI, every step, requires thinking, reasoning, action, and token generation.”

Huang emphasized, “We have moved beyond pure training into the realm of inference. The inflection point for inference has arrived, and the amount of tokens and computation required has increased roughly ten thousand times compared to before.”

“With more computing power, more tokens can be generated, increasing revenue. More people can use AI, making it more advanced and intelligent.”

Huang stated that the inflection point for AI inference has arrived.

02

From chips to Token factories

“The token factory is an AI factory—that’s your revenue source,” Huang said plainly during his speech.

He boldly predicted that by the end of 2027, the company would generate at least $1 trillion in revenue.

Last year, he estimated that this figure would reach about $500 billion by 2026. In just one year, the forecast doubled, highlighting that the wave of AI infrastructure investment is still rapidly expanding.

This “doubling” expectation is supported not just by increased GPU sales but also by NVIDIA’s redefinition of the future: data centers will no longer just store data and run software but will become “Token-producing factories.”

“Going forward, every cloud service provider, computer company, cloud enterprise, AI company, and all businesses will focus on the efficiency of their token factories,” Huang said. This will be the core factory of the future, driven by intelligence, which in turn will be empowered by tokens.

Tokens have become a tiered-priced commodity. Huang divided AI services into multiple business levels: free (high throughput, low speed), mid-tier (about $3 per million tokens), high-tier (about $6 per million tokens), ultra-high-speed (about $45 per million tokens), and super-high-speed (about $150 per million tokens).

The free tier attracts more customers, the high-end tier serves the most valuable clients, and together, they directly determine revenue.

CITIC Securities’ research report suggests that with the technological transformation brought by AI, future users may shift from buying tools to buying results. Software companies’ business models may gradually charge based on “results” and “token usage.” Native AI applications (AI-Native) developed by domestic software firms will continue to create value and may directly upgrade to result-based payment models.

03

The economics of “Lobster Farming”

OpenClaw has become the most talked-about “top star” at GTC.

Huang Huang praised OpenClaw’s significance in the NVIDIA annual GTC, comparing it to Linux and HTML—foundations of the next-generation internet and software.

He highly praised it: “This is the most popular open-source project in human history, surpassing Linux’s 30-year achievement in just a few weeks.”

“OpenClaw makes the creation of personal intelligent agents possible, with profound impact,” Huang said.

Earlier, Donghai Securities noted that the current wave driven by OpenClaw has pushed the AI agent concept to the brink of large-scale application, redefining human-computer interaction paradigms and forcing security system upgrades. Its massive inference token demand further boosts the computing power growth across the AI industry chain.

Why does the capital market care about open-source software? Because it opens a new business model.

Huang predicted, “After the widespread adoption of OpenClaw and intelligent agent technology, the industry landscape will be fundamentally changed. The most extraordinary transformation is that every IT company and SaaS enterprise will shift to becoming an intelligent agent-as-a-service provider—that’s an inevitable trend.”

Looking back at the entire speech, NVIDIA is playing a big game: acquiring Groq to fill inference gaps, cultivating intelligent agent ecosystems through open models and NemoClaw, penetrating every corner of data centers with Vera CPU and BlueField DPU, and ultimately integrating all businesses into its “Token factory” concept.

As long as tokens keep generating, NVIDIA keeps earning. Perhaps this is the foundation and logic behind the $1 trillion vision.