NVIDIA's New AI Plan Will Shock You

NVIDIA's new AI plan will shock you. Jensen Huang just walked onto a stage in San Jose in front of thousands of people and he said something that stopped the room cold. He said computing demand has increased by 1 million times in the last two years. 1 million times, not double, not 10x, 1 million.

And he said he sees at least a trillion dollars in orders for NVIDIA's chips through 2027. Last year he said 500 billion, he just doubled it in 12 months. That's what NVIDIA's GTC 2026 conference looked like. And if you run any kind of business, or you're building something with AI right now, or you're still on the fence about whether this stuff is real, this event changes the picture completely.

And I'm going to talk you through it today. So let me walk you through what actually happened, and most importantly, what it means for you. So here's the thing that most people get wrong about NVIDIA. They think it's a chip company.

It sells graphic cards, right? Not anymore. Jensen Huang stood on that stage and made one argument for three hours straight. NVIDIA is now the operating system for the AI economy.

Not just the chips, the whole thing, the software, the agency infrastructure, all of it. He called today's data centers, AI token factories. Not servers, factories, industrial machines that produce tokens, the way a power plant produces electricity. And his pitch was simple.

Companies that run this infrastructure will print money. Companies that don't will fall behind. Huang said computing demand has risen 1 million times over the last few years. And he celebrated CUDA's 20th anniversary and the flywheel is just getting started.

And then he got specific. Let's talk about NemoClaw first, because this is the one that affects you directly. Jensen Huang urged companies to adopt an open core strategy, calling AI agents the next computing platform. NVIDIA introduced NemoClaw, a secure enterprise version designed to address privacy and governance challenges.

Here's a simple version. OpenClaw is the fastest growing open source AI agent project in history. It lets you build AI agents that work on your computer with your files, with your apps, without sending everything to the cloud. NemoClaw is NVIDIA's version of that.

With security built in, one command to install it. It runs on any hardware NVIDIA sells, right? Your laptop, a workstation, a cloud server. Jensen called OpenClaw the operating system for personal AI and the fastest growing open source project in history.

Launching NemoClaw to give anyone a one command install for secure, always on AI agents. He compared it to Linux, to Kubernetes, to HTML. His exact message to every CEO in that room was, what's your OpenClaw strategy? Think about that for a second.

The most valuable company on earth is telling every business on the planet to get an AI agent strategy now. A freelance social media manager could use this to run an agent that monitors trends, drafts content, and schedules posts all locally on their laptop. No subscription costs, no data leaving their machine. That's what NemoClaw makes possible.

Now, let's talk about Nemetron 3 Super because this is the AI brain that powers these agents. NVIDIA launched Nemetron 3 Super, a 120 billion parameter open model with 12 billion active parameters designed to run complex, agentic AI systems at scale. Open source, free to download, free to use. But here's the part that matters.

This isn't a chatbot. It's built to be the brain of a multi-agent system. The model delivers over 5x throughput compared to the previous Nemetron Super. It tackles the context explosion with a native 1 million token context window that gives agents long-term memory for aligned, high-accuracy reasoning.

What does a 1 million token context window actually mean? Imagine an employee who can read your entire business history, every email, every document, every file, and remember all of it at once. Never forgets, never loses a thread. That's what this model can hold in its head.

Nemetron 3 Super powered NVIDIA's AIQ research agent to the number one position on Deep Research Bench and Deep Research Bench 2 leaderboards. Benchmarks that measure an AI system's ability to conduct thorough, multi-step research across large document sets whilst maintaining reasoning coherence. The best open research agent in the world, which is free and available right now. Perplexity already integrated it.

CodeRabbit, Factory, and Greptile are all using it. Palantir and Siemens are deploying it. And you can download it right now from Hugging Face or build.nvidia.com. Before we go further, if you want to actually learn how to use all of this stuff in your business, not just watch from the sidelines, the AI Profit Boardroom is where that happens for live coaching course per week.

Daily step-by-step tutorials, 30-day roadmaps so you know exactly what to do and when. A community of 2,600 business owners already automating with AI prompts for everything. A local map so you can find and meet other members near you. And support around the clock because someone is always online.

Go to the AI Profit Boardroom, link in the comments description, or go to the AI Profit Boardroom.com if you want in. Now, the hardware. This is where it gets almost hard to believe. The centerpiece was the Vera Rubin platform.

A system of seven new chips and five rack types designed to function as one massive AI supercomputer. It pairs Rubin GPUs and Vera CPUs with the new Grok 3.1 Inference Accelerator, which NVIDIA claims delivers up to 35 times higher inference throughout per megawatt. That's throughput per megawatt. 35 times more output per watt of energy used.

The central NVL72 rack is expected to deliver four times training performance and 10x inference performance per watt compared to Blackwell. And Blackwell chips were already the most powerful AI chip on earth. NVIDIA didn't just upgrade the chip, they redesigned how inference works at the architectural level. They noticed the speed and cost are enemies of each other in AI compute.

So they split the job. Rubin GPUs handle the heavy thinking. Grok LPUs handle the last output and the fast output. During GTC 2026, NVIDIA demonstrated it live and inference providers running the same infrastructure saw tokens and token generation speeds jump from 700 to nearly 5000 per second after NVIDIA updated their software stack.

Same hardware, seven times the output, same machines, just a software update, seven times faster. That's not a hardware announcement, that's a cost announcement and AI just got dramatically cheaper to run without anyone buying new chips. Then there's Dynamo 1.0. NVIDIA announced Dynamo 1.0 open source software for generative and agentic inference at scale.

Just as a computer's operating system coordinates hardware and applications, Dynamo 1.0 functions as a distributed operating system of AI factories, seamlessly orchestrating GPU and memory resources across the cluster to power complex AI workloads. It's already in production. AstraZeneca, ByteDance, Callweave, Pinterest, SoftBank, Tencent Cloud together AI and many more have already deployed Dynamo in production. These aren't experiments, these are live systems at scale.

Here's where I want you to sit with something for a second. A year ago, the conversation was about whether AI was a trend. Six months ago, the conversation was about which chatbot is the best. Today, NVIDIA, the most valuable company on earth, just told the world that the era of AI agents running your business infrastructure is here.

Not coming, it's here now. Jensen Huang said every SaaS company will soon become agentic as a service. Every tool you pay for monthly, every software subscription, it's all becoming agent-based. Meaning instead of you clicking around in software, the software does the work for you.

And NVIDIA is building the entire stack that makes that happen. Cisco, CrowdStrike, Google and Microsoft have already signed up and on to Nemeclaw. The dominant open source agent platform now defaults to NVIDIA hardware at every single layer. The big players are already in, right?

The infrastructure is being built right now. The only question is whether you're building on top of it or just watching this happen. Let me give you the forward view. As mass AI adaption shifts from chatbots to agentic apps that spawn off other agents to accomplish tasks, the number of tokens being generated has exploded, creating even greater need for running inference at faster speeds.

Now, more agents means more tokens and more tokens means more compute needed. And more compute needed means the companies that get good at deploying agents now have a massive head start over everyone who waits. Jensen says he now sees through 2027 at least $1 trillion in demand for NVIDIA hardware, right? And that's not just a prediction from him.

That's a purchase order number. Companies are already committed. The infrastructure is being built. The agents are being deployed.

In fact, Uber is actually putting NVIDIA-powered robo-taxis in 28 cities by 2028. Starting in Los Angeles and San Francisco, they're deploying more than 3,500 NVIDIA Blackwell GPUs across its worldwide operations, scaling drug discovery and manufacturing. So just to put this in perspective, Uber are using this across their robo-taxis, and Roche, the drugs company, are actually implementing this across their drug discovery and manufacturing business. So this is already in production, in hospitals, in cars, in enterprise software.

And the agencies, creators, the solopreneurs, and e-commerce business that figure out how to plug these systems and this AI into agent systems early, they're going to do more work with fewer people at higher margins. It's just math. So here's what this means if you're a normal person running a business or building something. Number one, OpenClaw and NemoClaw are free.

You can download them, run them, start building an agent to handle something repetitive in your business. An e-commerce store could, for example, have an AI agent monitoring competitor pricing and updating their own listings automatically. Number two, Nemetron 3 Super is free. It's a hugging face right now.

It's the best open-source AI agent brain available, according to NVIDIA, and you don't have to pay for it. You can also check this out on OLAMA. Bear in mind, if you want to use this, it requires a lot of power. If you don't have the power, you can actually use APIs from NVIDIA directly instead.

Number three, the next six months matter intensely. The infrastructure is being built, the tools are landing, and the companies getting fluent in AI agents right now are setting themselves up for a very different 2027 than the ones who don't. Number four, watch what Jensen Huang says next. Not because he's always right, but because when the most powerful company in AI tells you where they're putting a trillion dollars, well, that's a signal worth taking seriously.

And the thing that gets me about this whole GTC event isn't the hardware. It's actually the language shift. Jensen Huang didn't just talk about AIs at all. He talked about it as an economy, as infrastructure, as a factory.

He talked about token budgets, the way companies talk about headcount budgets today. That's a different world than the one most people are still operating and living in. And the gap between the people who get that shift early and the people who figure it out late, that gap is going to be absolutely enormous. You don't need to be a developer.

You don't need to understand how chips work. You just need to understand that the AI agent arc is not a future thing anymore. And this era is coming fast, and video just made it present tense. The only real question is, what are you going to do with this?

If you're ready to actually do something with it, come join us at the AI Profit Boardroom. You get four AI automation coaching calls a week, daily AI automation tutorials, 2,600 members already building, and a community that shows you exactly how to use this stuff in a real business. Check it out at the AIprofitboardroom.com. Don't wait for this to feel obvious, then it's too late.

Thanks for watching, and I'll see you on the next one.