With just days left until Nvidia’s flagship GPU Technology Conference (GTC), Bank of America is telling investors exactly what to watch and why the stock’s current valuation may not reflect what’s to come.
Ahead of CEO Jensen Huang’s GTC keynote on March 16, analyst Vivek Arya reiterated a buy rating on Nvidia (NVDA) with a $300 price target. The note points to three specific areas of focus that Arya sees as perhaps the clearest signal of Nvidia’s trajectory through 2027 and 2028.
Timing is important. Nvidia shares are currently trading at what Arya described as a historically depressed 17 times forward price-to-earnings ratio, which he described as a trough level following a significant increase in the Blackwell architecture’s $500 billion in cumulative sales.
The bank sees the GTC keynote as a catalyst that could start closing the valuation gap.
Arya’s notes lay out exactly what investors should be paying attention to when Huang takes over on March 16. These three areas are more than just product announcements. They show how far ahead of the competition Nvidia’s roadmap actually is.
-
Product roadmap to 2028: Bank of America expects Nvidia to outline a complete product line from its current Vera Rubin platform to Feynman GPUs in 2028, and Arya said these three generations will be well ahead of the competition in terms of developer and enterprise commitment.
-
Co-designed reasoning portfolio: The bank is expected to release a range of new custom products, including CPX chips for inference pre-population workloads and language processing units (LPUs) for low-latency decoding, which may be integrated into Nvidia’s next-generation rack systems.
-
Proprietary optics in amplified networks: Bank of America is eyeing details on Nvidia’s next-generation 102.4T Spectrum-6 switches and 115T Quantum-X with co-packaged optics, saying these technologies could become critical infrastructure for large-scale artificial intelligence cluster deployments.
Much of the discussion about AI infrastructure over the past two years has focused on training, the process of building large language models. Nvidia dominates this market.
But the bank’s note suggests that the next battleground is inference, the process of running these models at scale for end users.
More NVIDIA:
Bank of America describes the new CPX and LPU products as “a new wave of co-engineered and disaggregated AI infrastructure,” adding that these architectures are likely to become increasingly important as AI workloads shift from large-scale training to inference.
Arya specifically highlighted LPU, developed in partnership with Groq, as a potential strategic addition. The unit uses large amounts of fast on-chip SRAM memory to handle low-latency token generation, a capability that directly addresses one of the most demanding requirements in real-time AI applications.
Bank of America compared the potential long-term significance of the Groq deal to Nvidia’s 2020 acquisition of Mellanox, which became the foundation for its expanded networking and artificial intelligence business.
RELATED: Nvidia’s History: Company Timeline and Facts
In addition to its product roadmap, Bank of America’s report also makes a straightforward valuation argument. Nvidia trades at 17 times forward earnings, currently at the low end of its historical range of 25 to 56 times.
Arya’s $300 target is based on 28 times 2027 earnings, which he believes is reasonable given Nvidia’s dominance in AI computing, networking and software.
Data center revenue forecasts that support this goal are important. Wall Street consensus estimates that cumulative revenue from Nvidia’s data center business will reach about $750 billion in 2026 and 2027, rising to about $1 trillion in 2027 and 2028.
Bank of America believes any color Huang provides on GTC’s Rubin ramp could push those forecasts further.
Nvidia has committed approximately $95 billion in supply agreements through fiscal 2027, including investments in the artificial intelligence ecosystem from companies such as OpenAI and Anthropic. This level of supply chain commitment gives Bank of America confidence that the revenue channel is real and not speculative.
Arya’s comments are optimistic, but not without caveats. The bank also sought clarification from GTC on several unresolved issues that have made some investors wary.
-
Impact of Middle East conflict on supply chains, sovereign demand and energy cost inflation for U.S. and non-U.S. data center construction
-
Nvidia has the ability to procure sufficient wafers, memory, substrates and optics based on its annual product cadence
-
The timing of introducing co-packaged optics and whether it is optional or required in Nvidia clusters
-
How Vera Rubin’s cost per token (estimated to be ~10x cheaper than Grace Blackwell’s) translates into real-world customer economics
Bank of America’s report comes as Nvidia’s stock price has been under pressure, although the bank said its fundamentals are intact. The 17 times forward earnings, which Arya calls a historic low, is the same multiple that preceded some of Nvidia’s strongest multi-year rallies.
The GTC keynote on March 16 was more than just a product event. For investors following Nvidia (NVDA), now is the moment when the market gets its clearest picture of whether the $1 trillion data center consensus in 2027 and 2028 is a ceiling or floor.
Related: Bank of America earmarks U.S. auto stocks for holding
This article was originally published by TheStreet on March 11, 2026, and first appeared in the Investment section. Click here to add TheStreet as your preferred source.