Nvidia CEO Jensen Huang defended the company’s dominance in artificial intelligence (AI) chips at the annual GPU Technology Conference (GTC) on March 18, amid growing concerns over competition and technological shifts. Huang emphasized Nvidia’s strong positioning as the AI industry transitions from training large models to inference, generating real-time responses for users.
Despite unveiling new hardware and software, including the Blackwell Ultra GPU and a next-generation Rubin AI chip integrating GPU, CPU, and networking capabilities, investor confidence wavered, with Nvidia’s stock declining 3.4% after the event. The company also faces competition from China’s DeepSeek, which claims to have developed competitive AI chatbots with fewer Nvidia chips, raising questions about future demand.
Huang announced that the Blackwell Ultra GPU would launch in late 2025, offering enhanced memory for handling larger AI models, while the Rubin chip, set for 2026, aims to accelerate AI processing. The CEO also introduced a new DGX Workstation, co-developed with Dell, Lenovo, and HP, as a high-performance alternative to Apple’s top-tier Macs.
Nvidia is ready to tackle the next chapter of artificial intelligence (AI). But what does that mean, exactly?
Let’s unpack it step-by-step, because there’s a lot to dig into here, from new chips to market pressures, and even a little investor skepticism.
Huang kicked things off by saying Nvidia is “well positioned” for a major shift in the AI world. Up until now, AI has been all about training, think of it like teaching a super-smart robot to learn from massive piles of data so it can chat like a human or spot patterns like a pro.
That’s where Nvidia’s pricey, powerful chips have been king. But now, the focus is swinging toward inference, which is when these AI models stop learning and start answering questions or solving problems in real time, like when you ask ChatGPT to write a poem or your car’s self-driving system figures out the fastest route.
Why does this matter? Well, inference is where AI gets practical, it’s the part that touches our lives every day. And Huang says Nvidia’s got the goods to stay ahead.
But here’s the twist: not everyone’s convinced, and the stock market took a little dip, Nvidia’s shares dropped 3.4% after his speech, while the broader chip index fell 1.6%. So, what’s going on?
One of the juiciest bits from Huang’s talk was his take on something called “agentic AI.” This isn’t just your average chatbot, it’s AI that can think, reason, and act on its own, like a digital assistant that doesn’t need hand-holding. Huang says these smart agents need 100 times more computing power than we thought a year ago. “Almost the entire world got it wrong,” he said, grinning from the stage.
To put that in perspective, imagine you’re building a house. Training an AI is like laying the foundation, hard work, lots of heavy lifting. Inference was supposed to be the easy part, like hanging curtains. But Huang’s saying that with agentic AI, it’s more like adding a skyscraper on top, way more complicated and resource-heavy.
Research backs this up: a 2024 study from MIT’s Computer Science and AI Lab found that advanced reasoning tasks in AI can spike computational demands by orders of magnitude, especially when the system has to plan or solve multi-step problems.
This is huge for Nvidia because it means their high-powered chips might stay in demand, even as the industry shifts. But it’s not all smooth sailing, more on that later.
Huang didn’t just talk big, he brought out some shiny new toys. First up, the Rubin AI chip, set to hit the market in late 2026. This bad boy combines a GPU (graphics processing unit), CPU (central processing unit), and networking chips into one package.
Think of it like a Swiss Army knife for AI, built to handle everything from training to inference at lightning speed. Before that, we’ve got the Blackwell Ultra, coming later this year with more memory to juggle bigger AI models, and way down the road, the Feynman chips in 2028.
There’s also a hiccup with the current Blackwell chip, it’s been delayed by a design flaw, which slowed production. Still, Nvidia’s not sweating it.
Huang said last month that orders for Blackwell are “amazing,” and the numbers back him up: in 2025, Nvidia shipped 3.6 million GPUs to top cloud providers like Amazon and Google, up from 1.3 million the year before, according to CNBC.
That’s a big leap, showing demand isn’t slowing down.
Then there’s the DGX Workstation, a beefy new personal computer powered by Blackwell chips, coming from partners like Dell and Lenovo.
Huang held up a motherboard and quipped, “This is what a PC should look like”, a jab at Apple’s high-end Macs.
Plus, they’re tossing in free software called Dynamo to speed up AI reasoning. Oh, and General Motors signed up to use Nvidia’s tech for self-driving cars. Busy day, right?
Nvidia’s been the undisputed champ of AI chips, raking in $130.5 billion in sales last year, mostly from data center chips that cost tens of thousands each.
Their stock’s soared over 400% in three years, powering tools like ChatGPT. But now, some AI startups, like China’s DeepSeek, are saying they’ve built competitive chatbots with fewer Nvidia chips, according to Reuters. That’s a curveball, fewer chips could mean less need for Nvidia’s pricey hardware.
Other players are jumping in too. Companies like AMD and Google are cooking up chips designed just for inference, often cheaper and more power-efficient.
A recent benchmark from IEEE Spectrum showed Nvidia’s Blackwell topping the charts for raw performance, but rivals are closing the gap in specific tasks. It’s like Nvidia’s still the fastest car on the track, but others are finding shortcuts.
If Huang’s right about agentic AI needing tons of power, Nvidia could keep its crown as the go-to chipmaker. But if startups and rivals prove you can do more with less, the game could change fast.
Stats tell part of the story: the AI chip market hit $45 billion in 2024 and is projected to double by 2028, per Statista. Nvidia’s got a hefty slice now, but the pie’s getting bigger, and everyone wants a bite.
Huang called the GTC “the Super Bowl of AI,” and it’s easy to see why, Nvidia’s playing a high-stakes game. They’ve got new chips, big partnerships, and a vision for where AI’s headed.
But they’ve also got delays, competition, and jittery investors to wrestle with. As Huang put it, “If you take too long to answer a question, the customer’s not coming back.” That’s true for AI, and maybe for Nvidia too.
The next few years will tell us if they can keep the lead or if the field catches up. For now, all eyes are on San Jose, and the chips that could shape tomorrow.
