The Number That Matters Was Not the Revenue

Nvidia posted $68.1 billion in Q4 revenue against expectations of $66.2 billion and earned $1.62 per share against estimates of $1.53. The data center business grew 75% from a year ago. None of that is the real story. The real story is that the company guided Q1 to $78 billion, when analysts expected $72.6 billion. That $5.4 billion gap above the Street is where the structural signal lives. And it was built entirely without assuming a single dollar of revenue from China.

There is a pattern to how Nvidia's quarterly reports get processed. The market shows up with a number, Nvidia beats it, and then the conversation immediately moves to the guidance. This quarter was no different, but the guidance beat was wide enough to reframe the question. Analysts had walked into tonight expecting Nvidia to comfortably top Q4 estimates and guide to something in the low-to-mid $70 billion range. What they got instead was $78 billion, with nothing from China baked in.

The data center unit drove $62.3 billion of the quarter's $68.1 billion in total revenue. That is not a division of a company anymore. That is a company. Hyperscalers, meaning the four major cloud providers, accounted for just over half of that total. The rest came from sovereign AI projects, enterprises, and smaller cloud operators all trying to get their hands on compute before supply tightens further. The Blackwell architecture absorbed that demand without margin degradation. Gross margins came in at 75.2% on a non-GAAP basis, which is above where many on the Street modeled the Blackwell ramp would land given how aggressive the product transition has been.

The $78 billion guidance figure is the most structurally important piece of this report. Nvidia is not assuming any data center revenue from China in that number. Export restrictions on the H20 chip remain in effect, and the company did not assume they ease. What that means practically is that $78 billion is closer to a floor than a midpoint. Any movement on China policy becomes upside or downside applied on top of a baseline that is already printing massive numbers without it.

The combined capital expenditure plans from Alphabet, Amazon, Meta, and Microsoft are tracking toward roughly $700 billion for 2026. Nvidia captures a meaningful share of every dollar that flows into AI data center construction. The pipeline is not contracting. It is accelerating even as the stock has spent months trading sideways while the broader tech index has slipped into the red for the year.

The other dynamic worth understanding is the product transition already in motion. Vera Rubin, the successor to Blackwell, is currently in production with first systems expected to reach customers in the second half of 2026. Jensen Huang described it as delivering roughly ten times lower inference cost per token compared to Blackwell. That matters because the AI workload mix is shifting. Training was the dominant use case through 2024 and into 2025. Running those models at scale, what the industry calls inference, is becoming the growth driver now. Vera Rubin was designed specifically for that workload. The transition extends the runway on premium pricing and keeps Nvidia ahead of the architectural curve before any custom silicon efforts from the hyperscalers can compete in a meaningful way.

What is being priced incorrectly is time. The market knows Nvidia is dominant today. The debate is always about how long that lasts. Tonight's numbers suggest the answer is longer than the consensus model assumes. Blackwell is effectively sold out through mid-year, which means supply is the binding constraint on revenue, not demand. The next architecture is already in production and begins shipping before the year ends. The customer commitments already on the books, multi-year agreements with hyperscalers that now account for more than half of data center revenue, mean that pricing does not reset the moment new competition arrives.

The stock was trading around 24 times forward earnings going into this report. That multiple implies fairly rapid normalization. The mechanics of a sold-out architecture transitioning into a next-generation platform with pre-committed buyers do not point toward normalization in the next two quarters. They point toward extension. Whether that resolves as multiple expansion or simply as earnings growing into the current valuation is a separate question, but the structural direction is not ambiguous from tonight's numbers.

The constraints on this picture are real. Custom silicon from Google, Amazon, and Microsoft is maturing. These companies have strong financial incentives to reduce dependence on Nvidia and they are spending billions to get there. That competition does not threaten the next two quarters. It is a story that plays out over years. The more immediate risk is execution on the Rubin ramp itself. Any delay in yield or production creates a gap between peak Blackwell demand ending and Rubin revenue beginning. Gross margins could compress in that window, and the market would reprice quickly.

Geopolitical risk is also not resolved. China is a wildcard in both directions. A policy shift that reopens the market adds tens of billions to the revenue outlook with no incremental cost. Further tightening would not materially damage the Q1 guidance, which already excludes China, but it would affect the long-term ceiling on how large the business can grow.

The quarter confirmed three things for investors paying attention to structure rather than headlines. Demand from hyperscalers is not decelerating. Gross margins held above 75% through a major product transition. And the forward guide was set at a level that assumes a known headwind remains fully in place. That combination is not consistent with the rapid normalization the current multiple implies. The cycle is not over. It has a new product, a new workload category, and a pipeline of committed buyers who just told the market they plan to spend more in 2026 than anyone expected at the start of the year.

—

This analysis is for educational purposes. It does not constitute investment advice or a recommendation to buy or sell any security. Investors should conduct their own due diligence and consult financial advisors.