The most important takeaway from the keynote was the demand outlook. Jensen’s updated framing around Rubin points to CY27 revenue that is about 7% above where the Street was at, which would move growth from roughly 30% to closer to 40% or better. Three days following the event, the Street increased expectations up to 36% growth, underscoring analysts willingness to buy into the guide.

What stood out just as much was the stock reaction. Even with a stronger-than-expected demand implication, the shares struggled to fully reflect the improvement. That disconnect continues to highlight the wall of worry around Nvidia. Investors and analysts hear the upside, but many still do not believe the company can sustain this level of growth off an already massive base.

A second important take away from GTC was that inference economics are not headed toward a race to the bottom. Jensen’s argument, is that the market will increasingly reward the smartest tokens with the highest pricing. There will still be lower-cost tokens for high-throughput, low-latency use cases, but larger and more capable models should command meaningfully better pricing.

That is where the Groq-related discussion becomes important. The idea is that Nvidia can help power the premium end of the inference market, where application providers are paying for better reasoning, more context, and higher-quality output.

The broader takeaway is that not all tokens are created equal, and over time the pricing spread between basic and premium inference should widen. This view pushes back against the bear case that inference becomes purely commoditized and Nvidia’s customers will never make any real money from the massive AI buildout they’re underwriting.

Jensen spent a significant part of the keynote reinforcing that Nvidia is not simply a chip company. It is CUDA (Nvidia’s priority software layer that allows customers to build accelerated applications faster), compute, and networking. By customers standardize on the Nvidia platform they get better performance and lower cost per token out of their investment.

The final topic of the keynote was Physical AI which should become a more meaningful growth driver beginning in CY28. Last year, Automotive and Robotics is still only around 3% of revenue. We believe the segment represents another mega growth driver for the company, and can grow from from $2.3B last year to more than $70B by CY30 (Deepwater estimate) based largely on robotic manufacturing, autonomous vehicles and humanoid robots.

Jensen has weighed in on the topic that has wiped out between 10–30% of the value of many software companies this year. His message was that this shift is not the end of software, but a business model reset.

His view is that every software company must evolve into an agentic one. The implication goes beyond simply using more AI. Software companies will need to rethink what they sell and how they charge, moving toward a usage-based model.

This shift will move the industry from seats to outcomes, from SaaS to task completion. It is likely to unfold over a 3–5 year period and will create both disruption and opportunity across software. Some companies will struggle to adapt their products and pricing models, while others will be well positioned to benefit.

Disclaimer