Nvidia Corp (NASDAQ:NVDA, XETRA:NVD)'s upcoming GPU Technology Conference (GTC), being held from March 16 to 19, is expected to highlight the company’s evolving approach to artificial intelligence infrastructure, with analysts at UBS anticipating updates on system architecture, networking leadership, and the durability of AI spending.
The analysts wrote that the event could help address several ongoing investor debates around Nvidia’s technology roadmap.
While they see potential for a positive reaction in the stock, they do not expect the conference to significantly alter the broader investment narrative.
“We do see an upside bias for the stock on the event, although it is hard to see NVDA being able to provide thesis-altering commentary that creates a breakout for the stock,” the analysts wrote.
However, they added that the company may provide “some more confidence around system scalability, networking leadership, and AI capex durability.”
UBS expects the conference to underscore Nvidia’s push toward system-level optimization of AI workloads, building on the SuperPod architecture the company highlighted earlier this year. Rather than focusing primarily on successive GPU generations, Nvidia has increasingly framed AI infrastructure as a tightly integrated system spanning multiple computing elements, networking technologies and memory components.
The analysts wrote that Nvidia is “increasingly expanding the definition of an AI system beyond a simple compute plus networking rack to include heterogeneous compute configurations, dedicated networking, and memory/cache offload via BlueField DPUs.”
In their view, this reframing “shifts performance discussions away from standalone GPU generations and toward how workloads are decomposed, orchestrated, and scaled across the full system,” adding another layer to Nvidia’s software stack and strengthening the company’s broader platform strategy.
Networking is also expected to play a prominent role at the event as Nvidia continues to expand its position in the data-center interconnect market. UBS noted that Nvidia has reported strong share gains and now measures itself as the largest networking semiconductor vendor by revenue.
“We expect networking to remain front and center at the event,” the analysts wrote, adding they anticipate additional details on scaling technologies such as co-packaged optics and potential applications in future chip generations.
The UBS team also addressed the ongoing debate around memory architectures in AI systems, particularly as some emerging chip designs emphasize SRAM-based approaches.
While such designs could raise concerns about demand for traditional DRAM, especially high-bandwidth memory, the analysts said they continue to see DRAM as a critical performance component in large-scale AI systems.
“DRAM will remain the fundamental differentiating factor in AI hardware performance,” the analysts wrote, noting that SRAM-based architectures face scaling limitations, particularly around memory capacity.
Even the most advanced SRAM implementations currently provide significantly less capacity than HBM, which can offer as much as five times more capacity in current systems, they noted.
Hybrid memory approaches are already appearing in the market, they added, and memory suppliers are continuing to develop technologies aimed at improving performance metrics such as time-to-first-token by expanding cache capabilities.
UBS maintains a Buy rating on the stock with a 12-month price target of $245, implying potential upside of roughly 38% at their time of writing.