When Excellence Becomes the Curse
Few events in modern markets carry the weight of what has come to be called the "Super Bowl of earnings." When the dominant AI chipmaker reports, the entire technology ecosystem holds its breath. And yet, despite delivering numbers that would once have been considered scarcely believable, the stock's reaction has been muted, even slightly negative. The leading executive himself called the response "one of the biggest mysteries of the universe."
The explanation is straightforward: expectations have become extraordinarily high. This is the curse of being the best execution story in the market since the ChatGPT moment three years ago. The company is no longer being judged as a normal corporation. The market wanted a massive guide-up, some kind of surprise China unlock in guidance, a major Rubin platform revelation, or some brand-new catalyst to push estimates meaningfully higher. What it received was still remarkable — but no longer surprising enough to satisfy a narrative already priced for perfection.
The Hidden Gem: A $200 Billion New Market
The most underappreciated revelation from the quarter was the disclosure of approximately $20 billion in CPU revenue for this year. On the surface, that sounds modest compared to the company's $400 billion annual run rate. But this CPU expansion shouldn't be glossed over. The total addressable market in this category is now described as $200 billion — and this is a market the company has never meaningfully addressed before.
This doesn't suddenly transform the GPU giant into a CPU story overnight. Rather, it offers proof that the platform can continue expanding as AI matures. It demonstrates that the company has graduated from being merely a GPU merchant into something far more consequential: a full AI factory supplier. And the next major catalyst — physical AI — has not even begun to play out yet. That horizon alone could justify the next leg up in the broader story.
From Training to Inference: A New Era Has Arrived
The first stage of AI was all about training. It was GPUs, GPUs, GPUs, and the foundries that built them, with one company owning the lane decisively. But AI has now graduated into the inference era, and inference is expanding opportunity in ways that training never did.
Market attention has shifted to something like instant gratification: where is the next bottleneck? Power, networking, CPUs, memory — observers are constantly searching for the next unlock, sometimes forgetting who is actually laying down the groundwork for the entire AI economy. Data center demand continues to accelerate, producing a remarkable 92% year-over-year growth number. The question of whether hyperscaler capital expenditure can sustain this pace persists, but there are no signs of slowing down.
The Networking Revolution
If there was a single standout metric in the recent quarter, it was networking. While compute revenue stands at roughly $60 billion growing at 77% year-over-year, the networking business has reached $15 billion and is growing nearly 200% year-over-year. This is the kind of disparity that quietly rewrites a company's identity.
As time progresses, networking will become a bigger and bigger component of the overall business. This shift will reinforce the message that the company is not just selling accelerators — it is selling the full AI factory stack. AI clusters are becoming larger and larger, which means the bottleneck moves beyond raw GPU compute. GPUs must communicate with other GPUs, with CPUs, memory, storage, and networking systems at massive scale. If networking slows, the entire system underperforms.
The Prisoner's Dilemma
This brings us to perhaps the most compelling framing of the current moment: a prisoner's dilemma is unfolding among the biggest companies in the world, the ones with the deepest pockets. They cannot afford to have any system underperform, because they are locked in an arms race. The market can keep debating whether the leading chipmaker is "just a GPU company," but the business is moving deeper into every layer of the AI system.
The Vera Rubin platform looks to be a critical inflection point. If it remains on track for launch in the second half of this year, it extends the road map and pushes out the so-called "peak earnings" debate that the market keeps signaling. A clean Rubin launch could very well coincide with new all-time highs once hyperscalers release their next round of updated capital expenditure projections.
China: The Free Call Option
The China situation represents pure upside optionality. Current guidance assumes essentially zero data center compute revenue from the country. If that market reopens, it would be a massive unlock — China is the number two market in the world. But leadership has clearly recognized how little control any single company has in opening that market. There is perhaps too much investor attention on something so uncertain when so much else in the organization is humming.
If China opens, fantastic — that becomes the catalyst before hyperscalers report their next capex numbers. If it does not, the business is still growing at a scale almost no company in history has ever delivered.
"Demand Has Gone Parabolic"
The most revealing line from the recent earnings call was the description of demand as "parabolic" because generative AI has arrived, AI can do productive work, and tokens are becoming profitable. That last point is the most important. If tokens are profitable, then model builders have a reason to produce more of them. More tokens means more compute. More agents means more inference. More inference means more networking, memory, CPUs, power, cooling, and data center infrastructure.
This is the feedback loop that should put to rest any premature talk of a peak in the AI infrastructure trade. The cycle is broadening, not narrowing. Demand is insatiable, and there is no evidence it is going away.
The Real Risk: Politics, Not Competition
The most underdiscussed risk to the AI infrastructure thesis is not competition or execution — it is policy. The growing perception among Americans is that AI represents "the rich getting richer," with deep displacement risks for both blue-collar and white-collar workers. This sentiment is unlikely to favor a smooth political environment for AI buildout. The upcoming midterm elections could deliver a net-negative narrative for data centers, potentially leading to a tap on the brakes for the broader AI infrastructure story.
The phenomenon known as NIMBY-ism — "not in my backyard" opposition — has become increasingly relevant in conversations about where massive data centers can actually be built. Local resistance, regulatory friction, and political headwinds could create temporary disruptions.
Competition Is Finally Appearing — But Monopoly Remains
For the first time, there are real signs of competition. The market no longer requires the most advanced chips for every workload. Inference doesn't always need the bleeding edge — 70% performance is often "good enough." Custom silicon programs have shown serious momentum. One major hyperscaler's custom chip effort took significant attention away from the most advanced GPU ramp-up, sucking oxygen out of the room.
But context matters. Even if market share slips from roughly 95% to 85%, that still represents an overwhelming majority of the pie. The market is fixated on a slight decrease, but 85% is still a monopoly.
The Path to $10 Trillion
Even acknowledging temporary bumps along the way — whether from policy, competition, or sentiment — the trajectory remains intact. This is everything one would want to be in as an AI winner. It may well become the first ten-trillion-dollar company the world has ever seen.
The narrative shouldn't follow the price action with this stock. This remains an exceptionally well-run company that is doing everything one would want from the leader of the AI economy. The market will reward it eventually, because the data center demand keeps accelerating, the networking business keeps outpacing compute, and the inference era is just getting started. Time will prove out what the market is currently struggling to price.