Nvidia's AI Plan & NemoClaw Will Shock You

NVIDIA's AI plan and Nemeclaw will shock you. So Jensen Huang is about to walk on to a stage tomorrow and change the game again. Not just a little bit, not incrementally. I mean, the kind of shift where you look back in two years and say, that was the moment everything changed.

Tomorrow is March 16th, GTC kicks off in San Jose, California, and if you do not know what GTC is, here's a short version. It's the biggest AI conference on the planet. 30,000 people flying in from over 190 different countries. The event is spread across 10 venues in downtown San Jose and Jensen Huang, the CEO of what is now one of the most valuable companies in the world that's ever existed on earth, is going to stand on stage and tell the world what comes next.

NVIDIA's market cap right now, $4.6 trillion. To put that in perspective, that's bigger than the entire GDP of Germany. That is more valuable than Apple was at its peak and almost all of it has been built on the back of AI, on the chips that power every major AI model you've ever used, on the systems that run chat GPT, Gemini Cloud, and every other AI tool you've ever typed into. But here's what most people watching this don't understand.

NVIDIA is not just a chip company anymore. Tomorrow, they're going to prove that in a way nobody saw coming. So if you're not familiar with this already, let me break down everything that's expected to drop at GTC 2026 and why every single one of these announcements matters to you personally. Not just to engineers, not just to big corporations, to you, to your job, to your business, your income, to the way work gets done for every person on earth.

Let's start with the thing that everyone is talking about, NemoClaw. If you watched my last video about OpenClaw and Nemo, Tron Super, you already know the setup. OpenClaw is a free open source AI agent that took off faster than anything in the history of software. Jensen Huang called it probably the most important software release ever.

It spreads to over 250,000 stars on GitHub in record time. And people are literally lining up outside tech buildings in Shenzhen in China, just to get help installing it on their laptops. Mac minis were selling out in multiple US cities because people were buying dedicated machines just to run 24, seven hours a day with OpenClaw. But OpenClaw had a problem.

Seriously, OpenClaw was built for individuals, regular people who want a personal AI assistant running on their own laptop. But what about a hospital? What about a bank with 10,000 employees? What about a defense contractor or a Fortune 500 company with serious legal and security requirements?

You can't just hand one of these organizations an open source projects and say, good luck. And that's exactly where NVIDIA steps in tomorrow. On the software side, NVIDIA is expected to debut its own platform for AI agents called NemoClaw. The service would allow companies to deploy agents across their systems.

Think of it like this. OpenClaw is a bicycle. It's free, it's fast, it's getting you where you need to go and pretty much anyone can ride it. NemoClaw is an armored vehicle built for serious work, built for organizations that can't afford a security breach and built for companies that run the world.

NemoClaw would give businesses a structured way to build AI agents, software that can carry out multi-step tasks autonomously and will position NVIDIA to mirror similar offerings from companies like OpenAI. Let me tell you why that last part is massive. OpenAI is the most famous AI company on the planet. They own the most widely used AI tools and NVIDIA is saying, we're going to compete with them in the software layer too.

Not just chips, software agents, the thing that actually touches your day-to-day work. Now here's something that almost nobody's talking about and it's one of the coolest things happening at GTC tomorrow. The conference itself, attendees will actually be able to build their own assistant customizing its personality and capabilities before deploying it on NVIDIA hardware. They're calling it BuilderClaw.

Basically, you walk up, you configure your own personal AI agent, you give it a personality, you give it capabilities and you walk out with something that's yours running on NVIDIA's chips. Think about what that means as a signal. NVIDIA is not just announcing a product, they are letting 30,000 of the smartest developers and business leaders in the world get their hands on it in real time. That's a masterclass in product launches.

You don't just tell people about the thing, you put the thing in their hands and let them feel it. Think about what that means as a signal. NVIDIA is not just announcing a product, they are letting 30,000 of the smartest developers and business leaders in the world get their hands on it in real time. This is a masterclass in product launches.

You don't just tell people about the thing, you put it in their hands and let them feel it. And for the people watching from home, this is the moment you need to pay attention because what gets built at GTC this week is going to be available to businesses everywhere in the coming months. The stuff that 30,000 people test drive tomorrow becomes the stuff that your employer starts using by the end of this year. Now, let me talk about the hardware side because this is where it gets genuinely mind-bending.

The centerpiece of GTC 2020 is the anticipated formal launch of the Vera Rubin platform. The successor to the highly successful Blackwell architecture. Blackwell redefined data center efficiency in 2024 and 2025. Rubin is expected to be a generational leap.

Let me translate that into plain English. Every time you use an AI tool, every time you ask Chachibity a question, every time you use an AI image generator, every time you use an AI coding assistant, there is a chip somewhere in a data center actually doing actual work. For the last year and a half, that chip has mostly been something called Blackwell. Blackwell was already incredible.

It was a massive leap forward from the generation before it. Vera Rubin is what comes after Blackwell. The Rubin platform harnesses extreme co-design across hardware and software to deliver up to a 10X reduction in inference token costs and a 4X reduction in the number of GPUs needed to train certain AI models compared with the previous Blackwell platform. 10 times cheaper.

Let that sink in. Cost of running AI is going to drop by 10 times compared to the current standard. That's not a small improvement. That's not a 10% discount.

That is the difference between something being a luxury and something being everywhere. Right now, the main reason companies don't deploy AI agents at massive scale is cost. Running a persistent AI agent that works 24 hours a day, making decisions, managing tasks, connecting to systems is very expensive. Very expensive.

Jensen Huang himself said that an agentic task uses roughly 1,000 times more computing than a simple chat conversation. And a persistent agent running all day uses roughly 1,000,000 times more computing than a simple query. 1,000,000 times more. Pretty insane.

And that's why this cost matters so much. Because if you make it 10 times cheaper, suddenly the maths changes for every company and almost every single person on the planet. And that's why the cost matters so much, right? Because if you make it 10 times cheaper, suddenly the math changes for every company and things that were too expensive to automate become affordable.

Agents that companies could only dream about become practical. And all of the demand flows right back to NVIDIA's chips. The Vera Rubin Superchip combines an ARM-based Vera CPU. Rubin's GPU is expected to be available in the second half of 2026.

Among the first cloud providers to deploy Vera Rubin-based systems will be AWS, Google Cloud, Microsoft Azure, Oracle, as well as cloud partners like CoreWeave, Lambda, Nebius, and Enscale. So when Vera Rubin ships later this year, it's not gonna be some niche corner of the market. It's gonna be directly into the biggest cloud platforms on earth. Every time you use AI through Amazon, Google, or Microsoft, you're gonna be running on Vera Rubin.

That's the scale we're talking about. But here's where the story gets even more interesting because Vera Rubin is not even the biggest announcement expected tomorrow. Jensen Wang has also teased new chips the world has never seen before. On his earnings call a few weeks ago, he said, and I'm paraphrasing here, I have some great ideas I want to share with you at GTC, which is Jensen's speak for brace yourselves.

Throughout the week, NVIDIA plans to share at least some of its vision for incorporating Grok's chip technology into its already dominant computing ecosystem. Now, let me explain why the Grok story is a big deal. Grok, spelled G-R-O-K, is a chip company that built something called an LPU, Language Processing Unit. Not a GPU, not a CPU, a chip designed from the ground up specifically for one job, running AI models fast, really fast.

We're talking hundreds or even thousands of tokens per second. When you use an AI tool and the text just streams out instantly with no lag, that's the kind of speed Grok chips were built for. NVIDIA reportedly paid $20 billion last year to license Grok's technology. John from Ross, Grok's founder, and Sonny Madra, Grok's president, both agreed to join NVIDIA to help advance the scale of licensed tech.

$20 billion for a license, not even a full acquisition. That tells you exactly how much NVIDIA values what Grok built. And tomorrow at GTC, we are expected to see how NVIDIA plans to combine its own GPU dominance with Grok's inference speed into a single unified system. Think about what that looks like.

Right now, training an AI model and running an AI model are two different things that require different hardware setups. Training is about brute force computing power over long periods of time. Running, or inferring, is about speed and efficiency in real time. NVIDIA already dominates training.

Grok dominated real-time inference. If NVIDIA combines them, they own both ends of the equation. By combining its GPU ecosystem with Grok's Dataflow architecture, NVIDIA could significantly improve both token generation speeds and cost efficiency, making AI inference faster and cheaper. Faster and cheaper.

Those are the two things that unlock mass adoption of anything. Every time something becomes faster and cheaper, it goes from being something only big companies can afford to something everyone uses. This is what happened with smartphones, it's what happened with streaming, and it's what's about to happen with AI agents. Now, let me talk about the roadmap, because NVIDIA is not just announcing what's available now, they are laying out where this goes over the next several years.

Jensen is expected to provide additional details related to Vera Rubin and Variant Ultra slated for the second half of 2027. He could also offer more information about NVIDIA's future Feynman GPU, scheduled for 2028. Feynman, named after the physicist Richard Feynman. And this one is the one that has researchers genuinely excited.

Feynman is the next generation designed specifically to handle the reasoning and long-term memory requirements of AI agents, not general AI workloads, specifically for agents, specifically the kind of persistent multi-step memory intensive work that AI agents do when they're running all-day managing real tasks. This is NVIDIA quietly but clearly saying we know the future is agents, we're building chips specifically for the future, we are not waiting to see how it plays out, we are designing the hardware that makes it possible. Rumors also persist that Huang will provide a preview of the Feynman architecture which is slated to be the first chip produced using Taiwan Semiconductor's advanced 1.6 nm process node, 1.6 nanometers. To give you a frame of reference, a human hair is about 80,000 nanometers wide.

We are putting computing power onto chips at a scale that is almost impossible to comprehend and at that scale the performance improvements are enormous, more transistors, more speed, less power, less cost. NVIDIA is not one chip cycle ahead of competition, they are multiple chip cycles ahead and the gap is growing not closing. Now let me step back and talk about the bigger picture here because I think most people watching this do not still fully understand what NVIDIA is building. It's easy to look at this and think they're a chip company that makes really good chips for AI and that's true but massively undersells what's actually happening.

Jensen Huang described what NVIDIA is doing in a way that I think sums it up perfectly. He said AI is no longer a single breakthrough or application, it's essential infrastructure. Every company will use it, every nation will build it. Essential infrastructures like electricity or roads or the internet.

Think about the company that provides electricity to an entire country. Every house, every business, every hospital, every school runs on what they provide. You can't opt out, you can't build an alternative from scratch, you just buy the power and plug in and that's the position NVIDIA is building toward with AI. They want to be the power grid, the thing that everything runs on and tomorrow at GTC they're going to show us the next set of upgrades to that power grid.

GTC will showcase every layer of AI spanning energy, chips, infrastructure, models and applications through keynotes, sessions and demos. Attendees will see how each layer has its own ecosystem of partners, technologies and skilled jobs and how the coordination of these layers is driving one of the largest infrastructure expansions in history. One of the largest infrastructure expansions in history. Those are NVIDIA's own words, not a blogger's take, not an analyst's opinion.

The company itself describing what is happening right now in those terms and here's the investment side of this that should make every person's jaw drop. In early 2026, Amazon shocked the market by announcing a planned 200 billion dollars in capital expenditures for the year followed closely by Alphabet and Microsoft who guided for 180 billion dollars and 155 billion dollars respectively. 200 billion dollars from Amazon alone in one year spent on AI infrastructure. That is more than most countries entire defense budgets.

That is Amazon essentially saying we believe AI is the most important thing happening on earth right now and we're betting hundreds of billions of dollars on it. And where does the money go? A massive chunk of it goes to NVIDIA chips because there's no other option at this scale. AMD is trying, Intel is trying, custom chips from Google and Amazon exist.

But when you need to move this fast and deploy this much AI this quickly, you use NVIDIA every single time. NVIDIA has actually invested in dozens of AI companies since last year and it's deployed billions of dollars across its ecosystem. This week in advance of GTC, the company announced it invested two billion dollars in AI cloud firm Nebius and it's also backing former OpenAI CTO Myra Murati's new startup Thinking Machines with over one gigawatt in NVIDIA chips. Myra Murati, she was the CTO of OpenAI, one of the people closest to the technology, the Bill Chakchi Petit.

She left, started her own company and NVIDIA immediately backed her with more than one gigawatt of computing power. That's not a small bet. That is NVIDIA saying we believe in you and we want your company running on our infrastructure. This is the strategy.

Invest in the companies building on top of you. Make sure the best AI builders in the world are using your chips and then when those companies become enormous your chips are baked into everything they build. It's brilliant and it's working. Now let me bring this back down to earth because I can feel some of you thinking, okay this is all fascinating but what does it mean for me?

I'm not a chip designer, I don't run a data center, I just have a job and I'm trying to figure out how AI affects my life. Let me answer that directly. Every single announcement at GTC tomorrow makes AI agents more capable, more affordable and easier to deploy. Nemeclaw means companies can now run AI agents safely across their entire organization.

Vera Rubin means the cost of running those agents drops dramatically. The Grok Inference chip means those agents respond faster. The Feynman Preview means even bigger capabilities are coming in 2028. Now string those together.

Cheaper agents, safer agents, faster agents, capable agents running across every major company on earth. You send emails at work, well AI agents are going to build that. You schedule meetings, they'll build that too. Do you write reports?

AI agents are already beginning to build and draft those. Do you answer customer service questions? Do you do research or organize files? Well every single one of those tasks is in the crosshairs of what gets built on top of the chips Jensen Huang is announcing tomorrow.

And I'm not saying this to scare you, I'm saying it because the people who understand what's coming and start learning how to work with these tools instead of against them are going to have a massive advantage and not just today but for years to come. Let me talk for a minute about what this means for smaller businesses and regular people, not just the giants. NVIDIA is reportedly investing up to 26 billion dollars in open source models. Open source means free, means anyone can use it, means like a small business owner in Bangkok or a freelancer in London or a solo founder in Texas can access the same underlying AI technology that Amazon and Microsoft are using.

That has never happened before in the history of technology. Usually the big companies get the expensive powerful tools and everyone gets a watered down version. AI is flipping that. The open source models that NVIDIA is backing are genuinely world class and they're free to download and run.

The gap between what a 10 person company can do with AI and what a 10,000 person company can do is smaller than at any point in history. After tomorrow's announcements that gap gets even smaller. Now I want to talk about the conference sessions themselves because this is not just a keynote. GTC spans topics from physical AI and AI factories to agentic AI and inference.

More than 700 sessions provide all the details. This year's GTC spans topics from physical AI and AI factories to agentic AI and inference. Several hundred sessions. That is a week's worth of deep dives into every corner of AI.

Robotics, healthcare, autonomous vehicles, climate science and at the center of all of it the threads connecting everything is agents. On Wednesday March the 18th Huang will moderate a panel on open models with Harrison Chase, co-founder and CEO of Langchain and leaders from A16Z, A12, Cursor and Thinking Machines Lab. The conversation will be about where open models stand across frontier closed ones and what it means for everyone being on top of them. This panel matters.

The big question AI right now is whether open source models can compete with the closed models and the proprietary ones that come from companies like OpenAI and Anthropic because if open models get good enough no single company controls the best AI. Anyone can build on top of it, anyone can customize, anyone can run on their own hardware without paying per query fees to a big company forever. NVIDIA's interest is obviously in open models because open models need to run on hardware and NVIDIA sells hardware but there's a broader point here too. Open models mean more competition, more competition means faster improvement and faster improvement means better tools for everyone.

The panel Jensen is moderating includes the CEO of Cursor which is one of the most exciting companies in AI right now. Cursor built an AI coding tool so good that it now generates over 300 million dollars a year in revenue with only 20 people. 20 people with 300 million dollars in revenue. That's what AI first companies look like.

It's not a distant prediction, it's happening right now and Myra Murati's Thinking Machines lab is going to be represented there too. This is a woman who helped build Chachipiti and NVIDIA just partnered with the company in a huge way. NVIDIA and Thinking Machines labs announced a multi-year strategic partnership to deploy at least one gigawatt of next generation NVIDIA via Rubin systems to support Thinking Machines frontier model training. One gigawatt to train frontier AI models, that's an insane amount of computing power dedicated to one company's AI research and it signals where the frontier of AI capability is heading.

Now before I get to what you should actually do with all this, I want to touch on the robotics side because it's going to be a big part of GTC this week and it connects directly to the AI agent story. Expect NVIDIA to showcase new developments around its Isaac Robotics platform and Groot robotic models which are designed to power humanoid robots and industrial automation systems. NVIDIA's Omniverse simulation platform will likely play a major role too. The technology basically allows developers to build digital twins of real-world environments where robots can train and simulate tasks before being deployed in the real world.

Digital twins, let me explain what that means. So a digital twin is an exact virtual copy of a real physical space. It could be a factory floor, a warehouse, a hospital, a city. You build the virtual copy, you train your robots and your AI agents inside the simulation and then you deploy them in the real world.

The benefit is that AI can fail a thousand times in the simulation, learn from every failure and then show up to the real world basically already trained. And this is how you get robots to actually work in a messy real world environment. Not robots that only function in perfect lab conditions. Real world robots can handle surprises and adapt on the fly.

And here's the connection to agents and matters. The same technology that teaches a robot to navigate a factory floor also teaches an AI agent to navigate a complex software environment. The principles are the same. Simulate, fail, learn, improve, deploy.

NVIDIA is building these simulation tools not just for physical robots but for the AI agents who run software. Nemo claw agents in simulated environments before they get handed the keys to your company's email system. This is where it goes. Now let me talk about what this all means for the regular person who's sitting there watching this thinking I still don't know where to start with any of this.

Here's the honest truth. You don't need to understand chips to benefit from what's happening. You don't need to know what HBM4 memory is. You don't need to understand LPUs or Inference training.

What you need to understand is this. The tools built on top of all of this are about to get dramatically better, faster and cheaper and the people who learn how to use those tools right now, before the majority of people catch up, are going to have an advantage of compounds over time. I've been saying this for a long time. The AI learning curve is real and the people who climbed it early are already miles ahead at this point.

Not because they're smarter but because they learned and started sooner. The gap I'm talking about is not theoretical. I see it every single day. I talk to people who set up AI agents in their business and they're doing in two hours what used to take two days.

They are producing content at a scale that they never could before. They're automating the boring repetitive work and spending their time on the things that actually grow their business. And then I talk to people who still haven't started yet, who are waiting until they feel ready, who think they'll catch up later and the honest thing I can tell you is that later gets harder every month, not easier. The gap between people who understand this stuff and people who don't is growing.

Every new chip announcement, every new model release, every new agent framework that drops makes that gap wider. That's why I built the AI Profit Boardroom. It's a community where we cut through all the noise and focus on one thing. Real AI automation that actually makes you money and saves you time.

Not theory, not speculation. Actual workflows, actual tools, actual results. People inside the boardroom are already building AI systems that run parts of their business whilst they sleep. They're automating customer research, content creation, research for reporting and scheduling and a hundred other things that used to eat up their entire week.

If you want to stop watching from the sidelines and start actually building, the links in the description and the comments or you can go to the AIProfitBoardroom.com and join us especially right now whilst most people are still trying to figure out what GTC even is or what it's about. You can be the person in your industry who already knows how to use these tools. So let me bring everything together. Here's what I expect to happen tomorrow and what to watch out for.

Jensen Huang walks on stage at 11 a.m. Pacific Time at the SAP Center in San Jose. It streams free on NVIDIA's website. There's no registration required.

The keynote runs about two hours. Expect Nemaclaw to be officially announced. The enterprise AI agent platform. The thing that brings safe secure AI agents to every major company on earth.

This is huge for the agent ecosystem. Every company that was watching from the sidelines because OpenClaw had security issues now has a path forward with NVIDIA's backing. Expect Vera Rubin details. The new chip architecture that cuts AI costs by up to 10x.

AWS, Google, Microsoft and Oracle already lining up to deploy in the second half of this year and every AI tool will eventually run on these chips. Expect the GROK inference announcement, the combination of NVIDIA's training dominance and GROK's inference speed in a single product. This is what makes agents faster and cheaper to run in real time and expect Feynman to get the 2028 chip designed specifically for the agent era. Built from the ground up for the kind of AI that runs all day, maintains memory and takes complex multi-step actions.

Expect Robotics, Isaac, Groot, Omniverse, Digital Twins. The physical AI story is going to be a major part of tomorrow's keynote and expect the Open Models panel on Wednesday with Myra Morati, the Cursor CEO, people from ACE16Z and Sequoia discussing where OpenAI stands relative to closed AI. This conversation shapes the next two years of AI development. Here's the bigger picture I want you to leave with.

What NVIDIA is doing right now is one of the most ambitious things any company has ever attempted in the history of business and they're not just building better chips, they're building the entire stack. The chips, the models, the agent platforms, the simulation environments, the networking, the data center and architecture and every layer from raw silicon to the AI agent that answers your emails in the morning. From energy and chips to infrastructure, models and applications, every layer of the stack is advancing at once. Every layer at once.

That's not something you can catch up to quickly. When one company controls that much of the infrastructure and they're advancing every part of it simultaneously, the lead they have compounds with every single chip generation. And here's the thing that connects back to you sitting here watching this. All of that infrastructure, all those chips, all that computing power exists for one reason.

It exists to run AI models and those models exist to do work, real work. The work that right now gets done by people just like you. That's not a threat, it's a fact. And the question is not whether it's going to happen, it's already happening.

The question is whether you're going to be the one of those people who figures out how to use these tools to become ten times more productive and valuable or one of the people who gets blindsided by them. I know which side I want you to be on. Based on the fact that you're here watching this, I think you know which side you want to be on too. So here's what I want you to do.

Tomorrow if you can, watch Jensen Huang's keynote. It streams free, starts at 11 a.m. Pacific. Watch it the same way you would watch a keynote from a company that just changed the rules of the game because that's exactly what this is.

Pay attention specifically to NemoClaw. Watch how Jensen explains the agent vision. Watch the BuilderClaw demonstrations and understand what those 30,000 developers are building tomorrow in San Jose. You'll probably be landing in your industry within months.

If you're a developer, look at the Vera Rubin specs and start thinking about what you can build when AI costs ten times less to run. If you run a business, start making a list right now of every repetitive task your team does every week. Email, data entry, scheduling, reporting, research, customer routing. Every item on that list is a candidate for an AI agent on this platform that Jensen is announcing tomorrow.

And if you're in a corporate job, start paying attention to which companies in your industry are moving on AI and which ones are not. Because the ones that move first are going to have a structural advantage that gets harder to close every quarter. If you want to learn exactly how to build these systems, how to set up AI agents that actually work in the real world, how to automate the parts of your business, of your job, that are eating up your time every day, come join us in the AI Puffer Board. Link in the comments description or go to the AIPufferboard.com.

This is where people go from watching the news to actually using AI to build better businesses and better careers. Come join us. One last thing, there's a pre-show happening tomorrow before Jensen even takes the stage. The game show features notable speakers including Plexi CEO Aravind Sinharadas, LangChain CEO Harrison Chase, Skilled AI CEO Deepak Pathak, Open Evidence CEO Daniel Nadler and Mistral AI CEO Arthur Mensch.

The CEOs of some of the most important AI companies in the world are going to be warming up the room before Jensen even walks out. That pre-show starts at 8 a.m. Pacific. If you can catch any of it, it's definitely worth catching.

Tomorrow is not just a product announcement, tomorrow is the beginning of what NVIDIA is calling the agentic AI era. The shift from AI that answers questions to AI that does work. The shift from chat to action, from asking to building, from assistant to agent. The infrastructure for that era is being formally launched tomorrow on stage in San Jose.

The only question left is whether you're going to understand what's happening whilst it's happening or try to piece it together after the fact. I will see you on the other side of tomorrow's keynote because we're gonna cover everything that drops, break it all down in plain English and tell you exactly what it means for your business in your life. That's what this channel is for. That's why we do this every single day.

See you tomorrow.