NVIDIA's New AI Updates Just Made AI Agents Unstoppable

NVIDIA is about to change how AI agents run forever. So today is the day. Jensen Huang is about to walk onto the stage at SAP Center in San Jose, the same arena where the San Jose Sharks play hockey in front of 30,000 people from 190 countries. And the things he's about to announce or has already started announcing, depending on when you're watching this, are going to change how you work, how your company works and what AI can actually do for you right now.

I want to talk to you about all of it today. I want to break down what GTC 2026 actually means, which is NVIDIA's conference just about to drop. Not just for tech nerds, not just for investors, but for regular people who know how this changes their lives or want to know because it does change your life. And the people who understand this first are going to be way ahead.

So let's start with what drops. So GTC 2026 kicks off today, right? And this is happening in front of 30,000 people packed across 10 venues across downtown San Jose. This is not a small tech conference.

This is the biggest AI event on the planet right now. People are calling it the Woodstock of AI, the Super Bowl of tech. And that's not hype. That's because whatever Jensen Huang announced this week tends to become the foundation that the whole industry builds on for the next 12 months.

So what is he going to say? Three things, three huge things. And I'm going to walk you through every single one today. The first is Nemo Claw.

The second is Rubin Chips. And the third thing is Grok. And all three of them connect to the same idea that NVIDIA isn't just a chip company anymore. They're trying to become the company that powers every single AI agent on Earth.

Now, let's start with Nemo Claw, because this is the one that affects you the most directly. NVIDIA has been planning to launch an open source platform for AI agents called Nemo Claw. They've been building it quietly. And before the keynote even started, they were already pitching it to some of the biggest companies on the planet.

NVIDIA reached out to Salesforce, Cisco, Google, Adobe and CrowdStrike seeking partnerships ahead of the launch. Now, let me explain exactly what it is, because the name sounds technical, but the idea is really simple. You know how you can hire a real person to help you work, like an assistant who answers emails, schedules meetings, pulls data, writes reports? Well, Nemo Claw is the software that lets companies deploy AI to do all of that.

Not just for one person, for every single employee at the company at once. It's actually designed to be like an enterprise level software, kind of similar to something like OpenClaw, but it's designed for enterprise, right? So it's designed to enable companies to deploy AI agents that carry out tasks on behalf of their employees, processing data, managing workflows and implementing multi-step instructions with limited human oversight. Read that again, limited human oversight.

That means the AI doesn't just, for example, respond to you. It actually does the thing. You don't have to babysit anymore. You tell it what you need and it figures out the steps to actually get there.

Now, here's what makes this different from every other AI agent platform out there. It's free, it's open source, it'll work on any hardware, not just NVIDIA chips. Bear in mind, this is not 100% confirmed yet. This is what's expected to announce.

But Nemaclaw is designed to let enterprise companies send out agents to carry out tasks on behalf of their employees. So the platform comes with built in security and privacy tools. It's also hardware agnostic, meaning companies will be able to run it regardless of whether their products are built on NVIDIA chips, right? And that last part is massive.

NVIDIA has spent years locking developers into their CUDA software ecosystem. Once you build on NVIDIA's tools, switching is painful. That's been a huge part of how they dominate the market. Now, they're voluntarily giving that up with Nemaclaw.

They're saying, we don't care what chips you're running. Use our AI agent platform, build your AI workforce on us. And why would they do that? Because they're thinking bigger.

The logic is exactly the same as what Meta did with their LLAMA AI models, right? Meta gave away the models for free. And that sounds crazy for a company that wants to make money. But what happened?

Everybody started building on LLAMA. Everybody started running LLAMA workloads. And those workloads need chips to run on. Who makes the best chips?

Meta still mostly ends up spending money on NVIDIA hardware anyway. NVIDIA's open source approach mirrors Meta's playbook with LLAMA, which turbocharged open source AI development whilst boosting the demand for the hardware to run it. Give away the software, sell more chips, control the ecosystem. That's a game that Jensen Huang is playing, and he's been playing it for 30 years.

A man does not make moves without thinking 10 steps ahead. Now, let me tell you why Nemaclaw matters specifically right now, because of the timing right here. When Nemaclaw, the open source local agent framework that went viral in early 2026 before its creator, Peter Steinberger, was hired by OpenAI, it was found to have an unsecured database. So let anyone impersonate any agent on the platform.

Several large technology companies, including Meta, moved to ban it from corporate machines entirely. So there was this amazing agent platform. People loved it. OpenClaw was everywhere.

Even Jensen Huang reportedly called it the most important software release probably ever. And then it blew up because of a massive security hole. Big companies said no. IT departments said no.

It was too risky. Nemaclaw is NVIDIA's answer to that problem. Nemaclaw is being positioned as the enterprise safe answer to that chaos. Think of it this way.

OpenClaw was like a fast car with no seatbelts and no airbags. Great for the racetrack, but for your company's insurance, probably not going to cover it, right? Nemaclaw is the same car, the same speed, but now with a full safety cage, seatbelts, airbags and a black box. Now the corporate lawyers say yes.

The IT department says OK. And now it actually gets deployed at scale. That's what we're looking at with the release of Nemaclaw and its potential. Now, the agentic AR market is projected to reach $28 billion by 2027.

$20 billion in one year. And Nemaclaw is NVIDIA's play to own as much of that as possible. And here's where it gets even more interesting. OpenClaw itself was at GTC.

So Peter Seinberger, the creator of OpenClaw, is on stage at GTC discussing the rise of agentic systems. He's planning to be there. The guy whose platform blew up and got banned by Meta is now speaking at the world's biggest AI conference right next to the NVIDIA team that's building the safer version of this product. That's not awkward.

That's the AI industry in 2026. Everything moves fast. Everything's connected. And the best ideas, even the messy ones, end up building the next generation of tools.

What does this actually mean for you? Well, if you run a business, and I don't care if it's a small business or a huge company, you're going to be able to deploy agents that work for you within the next 12 months if you haven't already. So if you've not tested out OpenClaw, for example, this is a really powerful agent that you can set up once, tell it what it's supposed to do. Then it goes off and does it every day whilst you sleep.

And, you know, with stuff like the potential of NemoClaw, if it's free, if it's open source, you're not going to need a huge budget to access it. You're going to need the knowledge to set it up and then make it work for your specific situation. And that's why I keep saying this over and over. The people who learn how to work with AI agents right now are going to be untouchable in the next few years.

Not because technology will have passed them by, but because they'll have two years of actual practice whilst everyone else is still trying to figure out what an AI agent even is. So let's talk about the hardware, because if NemoClaw is a car, the Vera Rubin chip is the engine that makes the whole thing possible. Each Rubin's GPU delivers 50 PFOPS of inference performance using MVF P4, a 5x improvement of a Blackwell GB200. I need to translate that for you because PFOPS sounds like something a physicist made up, right?

One petaflop means the computer could do 1,000 trillion math operations per second. 50 petaflops means 50,000 trillion operations per second in a single chip. That's almost an incomprehensible amount of computing power sitting in a box you can install inside a data center, right? And the Rubin chip is five times faster than what NVIDIA released this year.

The Blackwell chip, which already seemed impossibly powerful when it came out, is now being outrun by a successor. And the thing that matters most to most people isn't the raw speed power number, it's what the speed translation and how it translates into, right? So the full NVL72 rack promises 10x lower cost per token inference compared to Blackwell. Now, bear in mind, this again needs to be announced.

It needs to be confirmed. But this is what we're expecting from the conference today. 10 times lower costs. That means running AI, asking it questions, having it do the work, deploying it in your products, costs 10 times less on Rubin hardware than it did on Blackwell.

And Blackwell was already cheaper than hot. The cost of AI keeps falling off a cliff every year. A year ago, running a sophisticated AI agent in your business would cost a lot of money, right? You needed serious infrastructure or you paid big cloud bills.

Six months from now, when Rubin hardware starts running out to cloud providers, the cost of running AI agents is going to be so low the budget will stop being the main barrier. Now, let me tell you who's already lined up to get Rubin hardware. Among the first cloud providers to deploy their Rubin-based instances in 2026 will be AWS, Google Cloud, Microsoft and Oracle Cloud, as well as NVIDIA Cloud Partners, CoreWeave, Lambda, Nebius and Nscale. Microsoft will deploy NVIDIA, their Rubin and the L72 rack scale systems as part of next generation AI dent data centers.

And on the AI lab side, AI labs including Anthropic, Meta, OpenAI, Mistral and XAI all signed up on the trial to train on Rubin hardware. Every single major AI lab, every single major cloud provider, all committed to Vera Rubin. That means every AI model you use, Claude, ChatGPC, Gemini, Metaslama is going to be running on this hardware within the next year. And because the hardware is so much faster and cheaper, those models are going to get smarter and cost less to use.

That virtual cycle is what's driving this whole AI wave forward. Cheaper hardware enables smarter models. Smarter models creates new use cases. New use cases creates demand for more hardware.

NVIDIA and Thinking Labs announced a multi-year strategic partnership to deploy at least one gigabyte of next generation NVIDIA Vera Rubin systems to support Thinking Machines frontier model training. One gigawatt. That's the power consumption of a midsize city being dedicated to training AI models. We're building infrastructure at a scale that has no historical precedent.

This is not a trend. It's a new industrial era. As Jensen Huang himself said, AI is no longer a single breakthrough or application. It is essential infrastructure.

Every company will use it. Every nation will build it. Every company, every nation, not just the tech giants, not just startups. Every company.

If you sit near thinking that AI doesn't really apply to your industry, your job, your business, I'd push back on that hard because that's exactly what people said about the Internet in 1997. It's for tech people. It's for young people. My customers don't use it.

And then five years later, every business that didn't have a website was bleeding customers to the ones that did. AI is moving faster than the Internet did. And that's not me being dramatic. The data says.

Now, let's talk about Grok, because this is a part of the story that almost nobody is covering correctly. NVIDIA spent $20 billion in December to license Grok's intellectual property and hire key talent, including founder Jonathan Ross, who previously helped Google create TPUs. $20 billion, not to buy Grok outright, to license their technology. That's a very specific, very strategic move.

And here's why Grok matters so much. NVIDIA's regular GPUs, the ones that power everything from Chachibitty to your company's AR tools, are incredibly powerful for training models and for processing massive batch workloads. But they have a weak spot. When someone asks an AI a question and they need an answer right now, instantly in real time, that's called inference.

And doing inference super fast, at a very low cost, for millions of simultaneous users, is actually a different problem than what GPUs were originally designed for. Grok design chips, it calls language processing units, or LPUs, that are designed for inferencing or running AR models. The company claims its processors can run large language and other AR models up to 10 times more efficiently than GPUs. 10 times more efficiently for the specific use case of running AI in real time.

So NVIDIA looked at this and said, we need that because the whole AI industry is shifting from training models to running them. Every company that built a model is now trying to deploy it. That means inference is becoming the dominant workload. And Grok had cracked a piece of the code that NVIDIA hadn't.

So they spent $20 billion to bring that technology in-house, not to shut Grok down, to absorb what Grok made great and combine it with NVIDIA and what it already does. Jensen Huang has already said that the Grok agreement will pay and play a similar role to Mellanox, indicating that LPUs will help NVIDIA complement workload stages. The Mellanox acquisition is a great paradigm. NVIDIA basically bought Mellanox in 2020 for $7 billion to get their networking technology.

At the same time, people thought it was expensive and weird. Two years later, that networking technology was a core part of why NVIDIA could build the massive AI data centers that everybody needed. The Grok deal was the same play. You're going to understand it fully in 18 months when you see what it unlocks.

The reason all of this matters for regular people is something that I want you to sit with for a minute. When AI inference gets 10 times faster and 10 times cheaper, which is what the combination of Vera Rubin and Grok technology is pointing towards, the kinds of things AI agents can do changes dramatically. Right now, AI agents are a little slow sometimes. You know, you give them a complex task and they take a minute.

They process things in steps. That's fine for a lot of use cases, but for the most powerful use cases, think, for example, AI that's monitoring your business in real time, catching problems before they happen, automatically responding to customer issues in milliseconds. You need speed. You need inference that's as fast as a snap.

And that's what the combination of Vera Rubin hardware and Grok's LPU technology is building toward. Jensen Huang teased this at GTC, teased several new chips the world has never seen before, with rumors pointing to an inference focused processor incorporating technology from Grok, targeting 10 times lower inference costs and specialized agentic workloads. World surprising chips. Jensen Huang's words.

He doesn't say things like that lightly. Now, let me zoom out for a second, because I want to connect all three of these pieces together. Nemeclaw, Vera Rubin and Grok, because they're not separate stories, they're one story. NVIDIA has spent the last decade dominating AI hardware, right?

They made the chips to train our models. They make hundreds of billions of dollars doing that. But the hardware game is getting more competitive. AMD is catching up.

Google, Amazon and Meta are all building their own custom chips. Analysts think that NVIDIA will begin to see share loss starting in 2027 once in-house ASIC programs gain some scale, especially in the inference market. NVIDIA sees this coming. They've been seeing it coming for years.

So the move they're going to make is to move up the stack, right? Don't just make the chips, make the software platform that everyone builds on. Make the agent framework that every company deploys. Make the inference technology that every AI runs on.

Make yourself so embedded in the full stack that even if someone uses different hardware, no problem, they're still in NVIDIA's ecosystem. This is vertical integration through open source. Give away the software, open the ecosystem, sell more chips. That's a play, and it's a brilliant play, because even if AMD takes 10% off the chip market, if Nemeclaw becomes the default way that enterprises deploy agents, NVIDIA still wins.

If the Grok inference chip becomes the standard for real-time AI, then NVIDIA still wins. They're not just playing one game, they're playing every game simultaneously. Now, let me come back to what this means for you, the real-world impact, because I can talk all day about chips and software frameworks, but the thing I really want you to understand is how this lands in your actual life. The AI Profit Boarding, which is a community I run and where I help people actually learn how to apply this stuff, exists because there's a massive gap between what's being announced at conferences like GTC and what regular people know how to do, and how to implement it.

So, most people watching the GTC keynote are developers and investors. They speak the language, they know what a petaflop means, they understand what open source implies for enterprise adoption. But the majority of business owners, creators and marketers, and professionals watching this have no idea how to translate what Jensen Huang just announces into actual changes in how they work. That's the gap I feel.

And this week, that gap is particularly wide because what's happening at GTC isn't incremental. It's not a slightly better version of last year, it's a whole new layer of the stack being built. If you're a business owner, here's the direct implication. Within the next six to 12 months, every major software tool you use, your CRM, your email platform, your project management tool, your customer service software, is going to have an AI agent capability built in.

That's what NemoClause partnerships with Salesforce, Cisco, Google and Adobe potentially mean. Those companies will potentially bake agent capabilities directly into products you already pay for. The question isn't whether AI agents are coming to your business, they are. The question is whether you're the person who sets them up and makes them work for your goals, or whether you're the person who gets handed a half-configured tool and doesn't know what to do with it.

And the difference in outcome between those two people is enormous. If you're a manager or team leader, here's your direct implication. Your team is about to get a lot smaller in headcount, a lot more powerful in output. Not because people will be fired overnight, but because the tools available to each individual on your team are going to multiply their productivity so dramatically that you'll be able to do more with the same number of people or the same output with fewer people.

You need to understand this now so that you can be the leader who builds the high output team, rather than the leader who asks the same number of people to produce the same output and wonders why your competitors are running circles around you. If you're an individual contributor, someone who works in a company doing marketing, sales, operations, customer service, writing or analysis, here's your direct implication. The most secure version of your career for the next 10 years is not ignoring AI and hoping your specific skills stay relevant. The most secure version is becoming the person on your team who actually knows how to use these agent tools to do more in less time.

The people who add AI agent skills to whatever they already do become twice as valuable. The people who don't are going to find out their role is being redefined around them. And I'm not saying that to scare you. I'm saying it because it's real and because the people who act on it now get a headstart that compounds over time.

Now, let me get back to what else is happening at GTC because there's a lot more. The pre-game show will feature the CEOs of Perplexy, Langchain, Mistral, Skilled AI and Open Evidence. And think about who those people are for a second. Perplexy is the AI search engine that millions of people use instead of Google.

Langchain is a framework that most AI developers use to build agent applications. Mistral is a European open-source AI lab that's been putting out some incredible models. All of them were there and will be there to kick off Jensen Huang's biggest week of the year. That tells you something about where the power is concentrated in AI right now.

And it tells you that NVIDIA is the gravitational center of it. Every major AI company needs to be in that room. That's just where the industry is. GTC sessions this week will dig into how Nemetron Super integrates with Nemacore and how it runs on Vera Rubin software.

The Nemetron 3 Super is NVIDIA's own AI model and they launched it just before GTC and it's specifically built for agentic AI workloads. Now, let me explain what Nemetron 3 Super is because it matters way more than people realize, right? Nemetron Super 3 is a 120 billion parameter hybrid member transformer mixture of experts model with 12 billion parameters active at inference time and a 1 million token context window. That sounds like a lot of jargon.

Let me break it down simply, right? 120 billion parameters is the total size of the model, right? Now, you want to think of parameters like the total amount of stuff that AI has learned. The more parameters, the more it knows.

But here's the smart part. Only 12 billion of those are active at any given time. The model routes each question to the most relevant specialized experts inside it rather than activating all 120 billion for every single request. So imagine if you had a company with thousands of specialized employees.

Instead of asking all of them to work on every single project, you have a smart routing system that figures out which three or four people are best suited for each specific task and only activates them, right? You get the combined knowledge of thousands of experts but the cost and speed of just using three or four. That's what a mixture of expert architecture does. And it's why Nemetron 3 Super achieves up to five times higher throughput and twice the accuracy of the previous generation Nemetron Super model.

Five times faster, twice as accurate and it connects directly to other potential tools like Nemeclaw to power the agent it deploys. NVIDIA isn't just selling you the hardware. They're not just giving you the agent platform. They're also giving you the AI brain that powers the agents.

It's a complete package from the chip to the software to the model. That's the full stack that Jensen Huang has been talking about for years. And GCC 2026 is a moment where it all comes together in one coherent vision. And then there's the stuff at GCC that people don't talk about as much but it's maybe just as mind bending if you think about the long-term game, right?

So GCC 2026 includes two dedicated physical AI days covering robotics, autonomous vehicles, industrial AI and digital twins. Physical AI, that's a term NVIDIA uses for AI that operates in the physical world. Not just answering questions on a screen, actually controlling robots, driving cars, managing industrial equipment, simulating real-world physics to train AI systems. The highlight on the robotic side is Isaac Groot N1.6, NVIDIA's vision language action model for humanoid robots.

Like its autonomous vehicle counterpoint, Alpameo 1 Groot N1.6 applies chain of thought reasoning to physical control, allowing robots to reason through novel situations step-by-step rather than pattern match against training data. Chain of thought reasoning in a robot. Let me sit with that for a second. Until recently, robots were not very smart in a very specific way.

They could do repetitive tasks incredibly. Pick this up, move it there, weld this seam. But the moment something unexpected happened, the part was in a slightly different position, for example, or the bolt was the wrong size or the thing that was supposed to be there wasn't, the robot failed. It just couldn't adapt.

Now, chain of thought reasoning changes that. The robot can now think through the problem. It can assess the situation, reason through what's different and figure out a new approach, not just pattern matching, actually thinking. Bind that with the speed improvements from Vera Rubin's software, the, sorry, hardware, the agent deployment framework from NemoClaw and the model improvements from Nemotron 3, and you can start to see where this is all going.

The AI agent that manages your business software in 2025 starts to look a lot like the robotics system that manages physical operations in 2027. And the line between digital AI agents and physical AI agents is going to blur over the next few years. And NVIDIA is positioning itself to own both sides of that line. Disney is presenting a session at GTC called Disney's Robotic Characters from the Screen to Reality via Physical AI.

And Disney is at a chip conference now talking about robots. That's a headline in itself. That's how fast physical AI is moving into mainstream applications. If Disney is putting AI-powered physical characters in their theme parks, and I believe that's exactly what that potentially could be about, then the public's acceptance and understanding of physical AI is going to accelerate dramatically.

Because the best way to normalize something is put it in a place where 30 million people a year visit. Now, let me talk about the money side of this because some of you are investors, some of you are business owners making capital allocation decisions, and all of you should understand the economic scale of what's happening. NVIDIA's Q4 fiscal 2026 results showed total revenue of $68.13 billion, a 73.2% increase from the same quarter a year earlier. Adjusted EPS climbed from 82% from the year ago value.

The company's data center business surged to 62.3 billion, rising 75% year over year. $68 billion in one quarter. The data center business, which means AI chips, was $62 billion of that. So $68 billion was the increase in one quarter, and AI chips is $62 billion of that.

So 91% of NVIDIA's revenue comes from selling AI hardware, and that hardware is growing at 75% per year. For context, there are very few companies in the history of business that have sustained 75% annual revenue growth at that scale. Most companies would love to grow 10% a year at that scale, right? NVIDIA is growing at 75%, at $68 billion a quarter.

So the main thing to understand this, the underlying business story, more AI workloads, cheaper hardware, driving more demand, software layer expansion, capturing more in the value chain, is about as clear as a business gets, right? And this is really changing the way everything is going, and this is how things are changing. You can see how NVIDIA is going to dominate, because we're at a sort of electricity moment for AI right now. You know, the conference is one of the marker points where people look back and say, that's when it became obvious, and I don't want you to miss this moment.

That's why I do this every day, right? That's why the AI Profit Board exists, because there's a real gap between people who understand this enough to act on it, and people who are watching from the sidelines waiting for it to become more obvious. And the gap in results between those two groups is growing every single month. You don't need to understand petaflops to benefit from this.

You don't need to know what a mixture of experts model is. You don't need to understand everything about chips. What you need is a clear plan for how you're going to start using AI agents in your work, in your business, in your creative process, and a community of people who are doing the same thing and sharing what's working. And that's what the AI Profit Board is, right?

If you've been on the fence about joining, the week of GTC 2026 is as good a moment as any to make the call, because the world Jensen Huang is describing, the one where AI is infrastructure, where every company has AI agents, where physical AI is operating in the real world, that world doesn't appear instantly on a Tuesday afternoon. It gets built, one announcement, one deployment, one business at a time, and the people who start learning how to navigate it now are the ones positioned to succeed when it arrives in form. So the fact that you're here watching this, learning this, thinking about this, that really puts you ahead of most people. The next step is applying it, starting somewhere, picking one process, one workflow, one part of your work, and figuring out how an AI agent could do it better or faster.

I would start there, then the next one, then the next one. By the time NemoClaw is widely deployed and Vera Rubin hardware is running your AI tools, you'll have a year of practice that nobody can take away from you. That's the game, and this week at GTC 2026, you got significantly more real. I'll see you on the next one.

If you haven't already, check out the AI Profit Boarding, link in the comments description, or just go to the AIProfitBoarding.com to check it out.