NVDA

What was called "Super Bowl of AI", unfortunately, ended in a draw, leaving both bull and bear teams feeling equally proud, but not entirely satisfied. The game moved into overtime as it has not yet surpassed what was actually expected of the event in the wildest dreams of the excited crowd. We are talking, of course, about Nvidia CEO Jensen Huang’s keynote speech at the GTC conference this week. The after-party mood is no apparent disappointment but rather a kind of lasting satiety with self-repetitions, as it usually happens when the next book of your favourite author has almost the same matter and a very similar plot, isn't imbued with novelty, and so you have a feeling it's still another self-portrait of the Master, even though a deuced talented copy of the previous one.

Jensen Huang knew everybody was expecting, and he decided to proclaimed the four "noble truths" he already broadcasted before when talking about his vision for AI prospects in the form of four waves, which are 1) Perception AI at the initial stage, focusing on speech recognition and other simple tasks; 2) Generative AI, with the focus of the past 5 years, involving text and image creation through predictive patterns; 3) Agentic AI, which is the current phase where AI interacts digitally and performs tasks autonomously to result in "reasoning models"; 4) Physical AI to represent "the future of AI", powering humanoid robots and real-world applications.

Nvidia's father probably tried to emphasize the multiplying AI monetization potential in stages #3 and #4 of this revolutionary era, stating that all computations take 100 times more tokens and resources than was originally expected at the current point of Agentic AI, but the market had already heard this philosophy and now wanted more precise sales numbers. Huang expected consuming companies' AI-driven capex (capital expenditures) of over $1 trillion globally by the end of 2028. Among other notable details, it was only said that in Nvidia GPU Hopper’s peaking year, they shipped around 1.3 million units, while for Nvidia’s Blackwell in 2025, 3.6 million units have been already ordered, and so traders perfectly knew how to respond.

In particular, they didn't buy the discourse and even briefly dropped the stock price from $122.9 at the peak on March 17 to a much lower range of the next day's regular session, between $114.5 and $115.5. However, the bulls were not averse to buying back these dips again, expanding the trading range to $120+ again within the rest of the week. A simple conclusion is that Nvidia may have mixed dynamics, and even with a possible downside bias in the short term, but will outperform a portfolio of other tech assets to slap into its role of the AI flagship in the months ahead once the current correction phase is exhausted. In short, a temporary decline somewhere close to $100 per share could not be ruled out, but the attractiveness of much higher targets above $180 is calling even stronger than ever.

Why are we so optimistic in the longer-term? We agree with the concept of Nvidia that each large product manufacturer will need two separate facilities, one is for manufacturing the product itself, and the second one is for production of the AI assistance on the surrounding information to sell this product better and in a more effective way with higher marginality. Nvidia's collaboration with General Motors to develop AI for next-generation vehicles, factories, and robots is a good example of it. They also work on AI-integrated wireless networks for 6G with T-Mobile and Cisco. Nvidia's open-source Dynamo Library inference software to double performance for Llama models and boost token generation by over 30x for providers including Amazon Web Services (AWS), Google Cloud, Meta and Microsoft Azure is another sign of Nvidia's prevailing power.

“While much of what was announced had been somewhat anticipated, we think Nvidia’s continued full stack/platform innovation was once again showcased; NVDA is solidly in a league of its own,” Wells Fargo analysts commented on the event. “The rate of innovation on all fronts continues to impress and suggests a growing moat vs peers... We were hoping for more proof points for total addressable market (TAM) expansion and Total Cost of Ownership (TCO) advantages but what is clear is the scale and strength of NVDA’s offerings across hardware/software and vertical domains”, Jefferies said. “Nothing hugely surprising given all the pre-event speculation, but we still thought it sounded good. The roadmap looks really solid, and their capability gap vs competitors across their entire massive stack continues to widen,” according to Bernstein investment house.