Nvidia’s GTC Is Coming: Will Rubin, Groq, and AI Demand Hype Send NVDA Soaring Again?

Nvidia (NVDA) is preparing to host its annual GPU Technology Conference, or GPU , from March 16–19 in San Jose, an event widely viewed across the semiconductor and AI ecosystem as the “Super Bowl of AI.” What began years ago as a developer-focused gathering has evolved into one of the most closely watched events in technology, where CEO Jensen Huang lays out Nvidia’s roadmap for the next phase of artificial intelligence infrastructure. With Nvidia’s market capitalization approaching historic levels and the AI investment cycle entering a more scrutinized phase, this year’s conference carries heightened importance for investors, customers, and competitors alike.

The central theme heading into GTC is the evolution of AI infrastructure from a collection of high-performance chips into a fully integrated computing ecosystem. Nvidia has increasingly framed AI as a layered technology stack spanning energy, chips, infrastructure, models, and applications. Huang recently described this framework as a “five-layer cake,” arguing that all layers must scale together for AI to become foundational across the global economy. Nvidia sits at the center of that stack, supplying the GPUs, networking, software, and increasingly the full system architectures that power modern AI data centers.

The biggest product announcement expected at the conference is further detail on Nvidia’s next-generation Vera Rubin architecture, which analysts believe will begin shipping in volume in the second half of the year. Angelo Zino, senior semiconductor analyst at CFRA, says investors already have a broad understanding of the company’s near-term roadmap but will still be watching closely for confirmation of performance improvements and production timelines. “The Street is pretty well versed in what we’re going to see in the second half of this year,” Zino explained. “That’s going to be Vera Rubin. It’s basically an extension of Blackwell at this point in time that leverages the same architecture but with performance enhancements and a shift toward HBM4 memory.”

High-bandwidth memory is likely to be a major topic at GTC, particularly as Rubin transitions to the next generation of HBM4 chips. The AI boom has dramatically increased demand for memory bandwidth, since training and inference workloads require massive data movement between processors and memory. Rubin is expected to deliver significant improvements in memory bandwidth and efficiency, potentially exceeding 3 terabytes per second of bandwidth in high-end configurations. Those improvements could lower the cost per token for advanced AI models and reduce the number of GPUs required for large-scale training workloads.

Another area of intense interest will be Nvidia’s longer-term architecture roadmap. Zino notes that while Rubin and Rubin Ultra are already well understood, investors are eager for clarity on Nvidia’s next major platform beyond those products. “What we don’t know a lot about is thereafter,” he said. “The Street pretty much wants more clarity on what 2028 is going to look like. That’s going to be Feynman, and investors are going to want to see some specs and some details about that.” Feynman, which could be built on TSMC’s advanced A16 process node, represents the next major leap in Nvidia’s GPU roadmap and may incorporate new technologies such as silicon photonics.

Inference computing will likely be another dominant theme at the conference. While the early AI boom focused heavily on training models, investors and enterprises are increasingly focused on inference—the process of running AI models in production to generate real-world outputs. Nvidia is expected to unveil new systems optimized for inference workloads, potentially including hardware that integrates Groq’s specialized language processing units. These LPUs are designed to reduce latency in conversational AI and real-time inference applications. According to Zino, the Groq technology could play an important role as AI moves toward agent-based computing systems. “Custom silicon chips are going to play a bigger role in that environment,” he said. “One way Nvidia can answer that is with their acquisition of Groq and rolling out these LPUs.”

Nvidia may also introduce a new open-source AI agent platform known as NemoClaw. The platform is reportedly designed to allow enterprises to deploy autonomous AI agents capable of executing complex multi-step tasks across software environments. The strategy would expand Nvidia’s software ecosystem while encouraging developers to build AI tools that ultimately drive demand for Nvidia’s compute infrastructure. If confirmed, such a move would signal Nvidia’s growing focus on the software layer of the AI stack rather than simply selling chips.

Beyond hardware and software announcements, the conference will likely feature a steady stream of partnership announcements that highlight Nvidia’s central role in the AI ecosystem. The company recently announced a $2 billion investment in AI cloud firm Nebius and has entered a multiyear partnership with Thinking Machines Lab, a startup founded by former OpenAI executive Mira Murati. The collaboration involves deploying gigawatt-scale AI infrastructure built on Nvidia’s technology. Huang has previously estimated that a single gigawatt-scale AI data center could represent as much as $35 billion in spending on Nvidia hardware.

Those partnerships reinforce Huang’s repeated assertion that demand for AI infrastructure remains extraordinarily strong. At a recent conference appearance, Huang said the company’s demand outlook had moved “from being incredibly high to higher than that.” He also emphasized the company’s ability to scale production rapidly to meet hyperscale customer needs. “I got all the memories, I got all the wafers, I got all the CoWoS, I got all the packaging, I got all the connectors, I got all the cables,” Huang said, referring to Nvidia’s efforts to secure supply chains across the semiconductor ecosystem. “When Satya asks me to stand up a few gigawatts, the answer is no problem.”

Despite that confidence, investors remain focused on whether hyperscale spending on AI infrastructure can remain sustainable over the long term. Zino believes the current spending cycle still has several years of runway. “As long as the compute needs remain as strong as they are and we continue to see new use cases emerge, the Street will get more comfortable with the capex increases we’ve seen,” he said. “At least over the next two to three years, I don’t see a huge capex cliff.”

However, competition is intensifying. Companies such as AMD, Broadcom, and Marvell are expanding their presence in AI data center chips, particularly in custom silicon. Zino expects Nvidia’s market share to gradually decline from its current dominant position, though he believes the company can offset that through rising content within each data center. “It’s extremely difficult for any company to sustain a 90% share long term,” he said. “But the content growth inside the data center is still very significant with Rubin and Rubin Ultra.”

For investors, the biggest potential catalyst from GTC would be evidence that Nvidia can sustain strong demand beyond the current generation of AI infrastructure. Positive surprises could include early details about the Feynman architecture, stronger-than-expected performance gains from Rubin, or major new hyperscale partnerships. Confirmation that Rubin systems will ship in large volumes later this year would also likely reinforce confidence in Nvidia’s growth trajectory.

On the other hand, the conference could disappoint investors if Nvidia fails to address key concerns. Analysts will be looking for reassurance that supply constraints around advanced packaging and memory are easing. Investors will also want clarity on the pace of adoption for next-generation AI workloads such as agentic AI and real-time inference. If Huang’s keynote focuses primarily on incremental updates rather than breakthrough innovations, the event could fall short of the market’s elevated expectations.

Ultimately, Nvidia’s GTC conference has become a barometer for the broader AI industry. As Zino put it, the coming months represent an important stretch for the entire technology sector. “Over the next three months you’ve got GTC, then developer conferences from Alphabet, Meta, Amazon, and Microsoft,” he said. “It’s going to be a really important period to see what these companies roll out and how the AI ecosystem evolves.”