黄仁勋：L'IA entre dans l'ère du « faire le travail », le Token devient en train de se transformer en « argent »

北京时间3月17日凌晨，“AI界春晚”英伟达GTC 2026大会在加州圣何塞正式拉开帷幕。
On March 17th at midnight Beijing time, the “Spring Festival Gala of AI” NVIDIA GTC 2026 conference officially kicked off in San Jose, California.

在2个多小时的演讲中，英伟达CEO黄仁勋讲的不仅是芯片参数，更是一张关于“钱往哪流”的蓝图。
In a speech lasting more than two hours, NVIDIA CEO Jensen Huang spoke not only about chip parameters but also about a blueprint for “where the money flows.”

他抛出的一张“万亿美金”大饼后，市场立刻买账。
After he presented a “trillion-dollar” pie, the market immediately bought in.

直接推动英伟达股价盘中一度飙升近5%，纳指也随之大涨近2%。
This directly pushed NVIDIA’s stock price to soar nearly 5% at one point, with the Nasdaq also rising nearly 2%.

不过，黄仁勋这次不仅画了饼，更说明了背后的商业逻辑：AI正在从“训练”转向“推理”，从“生成”转向“干活”。
However, Jensen Huang not only painted the pie this time but also explained the underlying business logic: AI is shifting from “training” to “inference,” from “generating” to “doing work.”

在这个关键的转折点上，算力需求正被放大。
At this critical turning point, the demand for computing power is being amplified.

东吴证券此前预计，英伟达GTC 2026锚定四大核心赛道——芯片、智能体、光模块、电力，成为贯穿未来AI投资的主线。
Dongwu Securities previously predicted that NVIDIA GTC 2026 will anchor four core tracks—chips, intelligent agents, optical modules, and electricity—as the main line running through future AI investments.

而这一切的底层逻辑，都指向同一个方向，Token正在成为AI时代的硬通货。
And all of this underlying logic points in the same direction: Tokens are becoming the hard currency of the AI era.

正如黄仁勋所言：“在这个Token工厂里，你的吞吐量和Token生成速度，将直接转化为你明年的收入。”
As Jensen Huang said, “In this Token factory, your throughput and Token generation speed will directly translate into your revenue for next year.”

一场有关谁是“Token之王”的争夺战，才刚刚开始。
A battle for who will be the “King of Tokens” has only just begun.

01
AI推理拐点已至
01
The Inflection Point of AI Inference Has Arrived

在演讲中，黄仁勋回顾了过去两年推动AI浪潮的三件大事：ChatGPT与生成式AI的开启、推理AI的崛起，以及Claude Code与智能体AI的诞生。
In his speech, Jensen Huang reviewed three major events over the past two years that have driven the AI wave: the launch of ChatGPT and generative AI, the rise of inference AI, and the birth of Claude Code and intelligent agent AI.

黄仁勋阐述道：“现在我们不再只是问AI‘什么、在哪里、什么时候、怎么样’这些基础问题。
Jensen Huang explained, "Now we are no longer just asking AI basic questions like ‘what, where, when, how.’

我们开始让它去创造、执行、构建。
We are starting to let it create, execute, and build.

它学会了使用工具，接入你的上下文环境、读取文件。
It has learned to use tools, access your contextual environment, and read files.

它能智能地把复杂问题拆解开来，一步步推理、反思，最终不仅找到答案，还能实际执行任务。”
It can intelligently break down complex problems, reason and reflect step by step, ultimately not only finding answers but also executing tasks in practice."

“AI的每一个部分，每一次都需要思考、推理、行动、生成Token。”
“Every part of AI needs to think, reason, act, and generate Tokens each time.”

黄仁勋强调，“现在已经过了单纯训练的阶段，进入了推理的领域。
Huang emphasized, "We have now moved past the pure training phase and entered the realm of inference.

所以推理的转折点已经到来，而此时所需的Token量、计算量，相比之前增加了大约一万倍。”
Therefore, the inflection point of inference has arrived, and the amount of Tokens and computing power required has increased by approximately ten thousand times compared to before."

“如果能有更多算力，就能生成更多Token，收入就会增加，更多人能够使用AI，AI就能变得更先进、更智能。”
“If there is more computing power, more Tokens can be generated, revenue will increase, more people can use AI, and AI can become more advanced and intelligent.”

黄仁勋说，AI推理的转折点已经到来。
Huang stated that the inflection point of AI inference has arrived.

02
从芯片到Token工厂
02
From Chips to Token Factories

“Token工厂，就是AI工厂，这就是你们的收入来源。”黄仁勋在演讲中直言。
“The Token factory is the AI factory; this is your source of income,” Huang stated bluntly in his speech.

演讲中，黄仁勋大胆预测，公司到2027年底将创造至少1万亿美元收入。
During the speech, Huang boldly predicted that the company would generate at least $1 trillion in revenue by the end of 2027.

去年，他预计到2026年底这一数字约为5000亿美元。
Last year, he estimated that this figure would be around $500 billion by the end of 2026.

短短一年，预期翻倍。
In just one year, the expectation has doubled.

再次凸显AI基础设施投资浪潮仍在快速扩张。
This further highlights that the wave of investment in AI infrastructure is still rapidly expanding.

支撑这个“翻倍”预期的，不仅仅是GPU卖得更多了，而是英伟达对未来的定义变了——未来的数据中心，将不再是存储数据和运行软件的机房，而是“生产Token的工厂”。
Supporting this “doubling” expectation is not only that more GPUs are being sold but that NVIDIA’s definition of the future has changed—the future data center will no longer be a room for storing data and running software but a “Token production factory.”

“未来，每一家云服务提供商、计算机公司、云端企业、AI 公司，乃至所有企业，都将关注其 Token 工厂的效率，”黄仁勋说。
“In the future, every cloud service provider, computer company, cloud enterprise, AI company, and indeed all enterprises will focus on the efficiency of their Token factories,” Huang said.

这就是未来的核心工厂，因为在座的每一位都由智能驱动，而未来，这种智能将由Token 赋能。
This is the core factory of the future because each one present is driven by intelligence, and in the future, this intelligence will be empowered by Tokens.

Token已经成为一种分层定价的商品。
Tokens have become a commodity with tiered pricing.

黄仁勋将AI服务划分为多个商业层级：免费层（高吞吐、低速度）、中级层（约每百万Token 3美元）、高级层（约每百万Token 6美元）、高速层（约每百万Token 45美元）以及超高速层（约每百万Token 150美元）。
Huang categorized AI services into multiple commercial tiers: free tier (high throughput, low speed), intermediate tier (about $3 per million Tokens), advanced tier (about $6 per million Tokens), high-speed tier (about $45 per million Tokens), and ultra-high-speed tier (about $150 per million Tokens).

免费层级吸引更多客户，高端层级服务最有价值的客户，综合起来，直接决定收入。
The free tier attracts more customers, while the high-end tier serves the most valuable customers, which together directly determines revenue.

中信证券研报认为，伴随AI带来的产业技术变革，未来用户可能会从买工具变为买结果，软件公司的商业模式可能会逐步围绕按“结果”“Token调用量”等模式收费。
CITIC Securities research report believes that with the industrial technological transformation brought by AI, users may shift from buying tools to buying results, and the business models of software companies may gradually revolve around charging based on “results” and “Token usage.”

AI时代国内软件应用公司通过开发原生的AI应用（AI-Native）带来价值量持续提升，也有望直接升级至按“结果”付费。
In the AI era, domestic software application companies can continuously enhance value by developing native AI applications (AI-Native) and are also expected to directly upgrade to a “result”-based payment model.

03
“养龙虾”的经济学
03
The Economics of “Farming Lobsters”

OpenClaw，成为GTC全场欢呼声最高的“顶流”。
OpenClaw has become the most cheered “top star” at GTC.

黄仁勋在英伟达年度GTC大会上谈及OpenClaw对软件行业的影响时称，OpenClaw的重要性堪比Linux与HTML——即下一代互联网和软件的底层地基。
In his address at NVIDIA’s annual GTC conference, Huang mentioned that the significance of OpenClaw to the software industry is comparable to that of Linux and HTML—it’s the foundational base for the next generation of the internet and software.

黄仁勋给了它极高的评价：“这是人类历史上最受欢迎的开源项目，仅用几周时间就超越了Linux 30年的成就。”
Huang gave it a high评价: “This is the most popular open-source project in human history, surpassing 30 years of achievements by Linux in just a few weeks.”

“OpenClaw 让个人智能体的创建成为可能，其影响极为深远。”黄仁勋还说。
“OpenClaw makes the creation of personal intelligent agents possible, and its impact is profound,” Huang added.

东海证券此前表示，当下OpenClaw带动的浪潮将AI Agent概念推向了规模化应用的前夜，重新定义了人机交互范式并倒逼安全体系升级，同时其大量的推理Token需求进一步拉动了AI产业链的算力增长。
Donghai Securities previously stated that the wave driven by OpenClaw is pushing the concept of AI Agents to the brink of large-scale application, redefining the human-computer interaction paradigm and forcing an upgrade of security systems, while the massive demand for inference Tokens is further driving the growth of computing power in the AI industry chain.

为什么资本市场要关心一个开源软件？因为它开启了一个新的商业模式。
Why should the capital market care about an open-source software? Because it opens up a new business model.

黄仁勋预言，“在OpenClaw 与智能体技术普及之后，行业格局将彻底改变。
Huang predicted, "After the popularization of OpenClaw and intelligent agent technology, the industry landscape will be completely transformed.

最非凡的变革在于，每家信息技术企业、每家软件即服务企业，都将转型为智能体即服务企业，这是必然趋势。”
The most extraordinary change is that every information technology company and every software-as-a-service company will transform into an intelligent agent-as-a-service company; this is an inevitable trend."

回顾整场演讲，英伟达正在下一盘大棋，通过收购Groq补齐推理短板，通过开源模型和NemoClaw培育智能体生态，通过Vera CPU和BlueField DPU渗透数据中心每一个角落，最后通过“Token工厂”的概念，把所有生意都装进自己的商业模型里。
Looking back at the entire speech, NVIDIA is playing a bigger game by acquiring Groq to fill the inference gap, nurturing the intelligent agent ecosystem through open-source models and NemoClaw, penetrating every corner of data centers through Vera CPUs and BlueField DPUs, and finally integrating all businesses into its commercial model through the concept of a “Token factory.”

只要Token还在生成，英伟达就在收钱。
As long as Tokens are still being generated, NVIDIA is making money.

这或许就是那1万亿美元的底气和硬逻辑。
This may be the reason and hard logic behind that $1 trillion.