Nvidia Corp (NASDAQ:NVDA, XETRA:NVD) is expected to unveil a broader suite of specialized artificial intelligence chips and networking technologies at its flagship developer conference next week, according to analysts at Bank of America.
The firm expects Nvidia to showcase a broadened AI computing stack, including systems designed for training, inference prefill, low-latency decoding and large-scale batching workloads.
Bank of America said the event will likely highlight the emergence of specialized AI inference engines, expanding Nvidia’s portfolio beyond training chips toward customized systems optimized for different stages of AI model processing.
Among the potential announcements are customized inference processors, including a CPX chip for inference prefill workloads and a low-latency decode processor known as an LPU, which analysts said could be integrated into future Nvidia rack-scale systems such as Rubin Ultra and the company’s next-generation platforms.
“These products represent a new wave of co-designed and disaggregated AI infrastructure,” Bank of America wrote, adding that such architectures could become increasingly important as AI workloads shift from training toward inference at scale.
The bank also expects Nvidia to outline its longer-term product roadmap through 2028, potentially providing updates on future GPU platforms including Feynman and further details on the ramp of the Rubin architecture.
Nvidia’s upcoming Rubin platform could deliver major efficiency gains, with analysts pointing to estimates that cost per token could fall roughly tenfold compared with the Grace Blackwell platform, improving economics for large AI deployments.
Another key theme at the conference could be high-speed networking and optics, including next-generation switches and optical technologies designed to support massive AI clusters.
Bank of America said Nvidia may provide updates on the Spectrum-6 Ethernet switch and the Quantum-X networking platform, as well as progress on co-packaged optics, a technology that integrates optical connectivity directly into switches to improve performance and energy efficiency in large data centers.
Investors will also be watching for signals on Nvidia’s supply chain and geopolitical risks amid heightened global tensions.
Analysts said the company could address the impact of the Middle East conflict on semiconductor supply chains, along with demand from sovereign AI projects and the pace of data center construction in the US and overseas.
Questions also remain around whether Nvidia can secure enough advanced wafers, memory, substrates and optical components to sustain its rapid annual product cycle.
Despite the uncertainty, Bank of America said Nvidia shares currently trade at about 17 times forward earnings, near a historical low for the company.
The bank noted the stock has been pressured following concerns around the pace of the AI investment cycle, even as Nvidia continues to benefit from strong demand for its Blackwell GPU platform, which analysts estimate could generate about $500 billion in cumulative sales over time.
Consensus forecasts already call for Nvidia’s data center business to approach $750 billion in cumulative revenue in 2026–2027 and roughly $1 trillion by 2027–2028, underscoring the scale of the AI infrastructure buildout.
Beyond new hardware, analysts will also be looking for updates on Nvidia’s software ecosystem, including inference frameworks built on its CUDA platform that could help maintain its competitive advantage in AI computing.
Bank of America said additional commentary on Nvidia’s capital allocation could also draw attention, as the company has already committed about $95 billion in supply agreements and invested heavily across the AI ecosystem, including stakes in companies such as OpenAI and Anthropic.
The GTC conference has increasingly become one of the most closely watched events in the technology sector, often setting the tone for the next phase of the AI computing cycle.