This past week, analyst James Schneider took a seat in a Goldman Sachs research office and made what may prove to be one of the year’s most significant decisions. Nvidia is the “most important stock” of 2026, according to him. Not the most captivating. Not the most popular. the most significant. It’s worth taking a moment to consider that particular word choice.
The timing is important. The setup leading up to Nvidia’s first-quarter earnings report, which is scheduled for May 20, is peculiar. The stock is up about 15% so far this year, which sounds good until you contrast it with AMD’s 90% gain or Intel’s astounding 197% gain over the same period. The names that were supposed to be pursuing the AI hardware wave are outpacing Nvidia, the company that essentially defined it. That has been noticed on the Street, and there’s something genuinely peculiar about it.
Schneider of Goldman focused on the valuation, and the results are startling. Currently, Nvidia’s price-to-earnings ratio is roughly ten times lower than its three-year median of thirty-two times. That compression is difficult to fully explain for a company that continues to report data center revenue of $193.7 billion in a single fiscal year, or almost 90% of its total. The market may be pricing in a protracted, gradual decline. Goldman doesn’t believe that narrative. Schneider increased his estimates of earnings per share by an average of about 12%, making his forecasts for the calendar years 2026 and 2027 14% and 34% higher than the Wall Street consensus, respectively. Those figures are aggressive. Clearly, the company isn’t hedging.
A few particular wagers form the basis of the main thesis. One is Nvidia’s current flagship chip architecture, Blackwell, which Goldman claims has already shown a ten-fold cost improvement over its predecessor. Another is the Vera Rubin platform, which is anticipated to launch in the second half of this year at a time when agentic AI and the move to inference workloads are driving up demand for CPUs. Additionally, hyperscaler capital expenditures across Amazon, Google, Microsoft, and other companies are close to $700 billion. This is part of the larger data center narrative. That figure does not indicate that the cycle of AI spending is coming to an end.
Schneider takes care to recognize the tension in the setup, though. In his own words, the standard for stock outperformance on earnings day is quite high. Despite exceeding expectations in the past, Nvidia’s stock continued to decline. The market has obviously already factored in some degree of positive news, and May 20 could go either way. A beat-and-raise quarter with reliable guidance on gross margins is what the Street wants, and Goldman appears to think it is possible, especially as input costs increase and Rubin intensifies.
As this situation develops, there’s a sense that the true question isn’t whether Nvidia can outperform consensus. Most likely, it can. The more difficult question is whether Jensen Huang can change the focus of the earnings call to the competitive threat posed by custom silicon. Amazon’s Trainium, Microsoft’s Maia, and Google’s TPUs are no longer theoretical substitutes. The same clients who once drove Nvidia’s fastest growth are now using them, deploying them, and making improvements. According to Goldman, Nvidia continues to lead in terms of inference cost effectiveness. They contend that this is supported by the Blackwell data.
Nvidia is thinking far beyond selling chips to big tech, as evidenced by the company’s recent partnership with IREN Limited and its plan to deploy up to 5 gigawatts of Nvidia AI infrastructure across IREN’s global data center pipeline. The non-hyperscaler growth story is found in sovereign AI, AI factories, and smaller enterprise deployments. According to Goldman’s analysis, investors need greater insight into how quickly this aspect of the business is progressing. Whether that visibility occurs on May 20 is still up in the air. However, when it does, Nvidia’s valuation might finally match its true standing in the market.