Analyze the changes in NVIDIA's FY26

As NVIDIA investors, we must often hear Jensen Huang talk about AI super factories, AI mega infrastructure, and how GPUs will become the most fundamental means of production in the AI era, just like water, electricity, and gas.

However, to judge a company, one cannot just listen to what the founder says; it's more important to see whether the financial statements have delivered on those promises.

So, compared to FY25, what exactly changed for NVIDIA in FY26? Do these changes truly indicate that the company is continuing to evolve towards an AI infrastructure platform?

Let's state the conclusion first: Yes, and this change has already begun to manifest very clearly in the financial statements.

But there's a premise to clarify here.

NVIDIA didn't suddenly transform from a "GPU company" into an "AI infrastructure company" only in FY26. In fact, in the FY25 annual report, the company had already defined Blackwell as a complete set of data center-scale infrastructure, including GPUs, CPUs, DPUs, interconnects, switch chips, systems, and network adapters. It was truly in FY26 that this platform logic began to be more centrally reflected in the revenue structure, cost structure, and risk structure.

From the results, NVIDIA remained exceptionally strong in FY26.

Full-year revenue was $215.938 billion, compared to $130.497 billion the previous year, a 65% year-over-year increase; gross margin was 71.1%, compared to 75.0% the previous year; operating expenses were $23.076 billion, a 41% year-over-year increase; operating profit was $130.387 billion, a 60% year-over-year increase; net profit was $120.067 billion, a 65% year-over-year increase; diluted earnings per share were $4.90, a 67% year-over-year increase.
This means the company did not weaken; instead, it continued to grow rapidly on a much larger scale.

But what's truly noteworthy is not just these aggregate figures, but how the growth came about.

In FY26, Data Center business revenue grew 68% year-over-year to $193.737 billion, having far and away become the company's core revenue engine. Breaking it down further, compute revenue within Data Center grew 59% year-over-year, while networking revenue grew 142% year-over-year. The compute portion was primarily driven by the Blackwell computing platform; the networking portion mainly came from the NVLink compute fabric of GB200/GB300, as well as growth in Ethernet and InfiniBand.

This actually illustrates a very important point:

What NVIDIA is selling now is increasingly not "single GPUs," but rather a whole suite of things:
• Computing platforms
• Networking platforms
• Interconnect platforms
• Rack-level / system-level solutions
• Supporting software and ecosystem

In other words, the biggest change in FY26 is not that NVIDIA started platformization, but that platformization began to truly materialize in the revenue structure.

The change in gross margin also illustrates this point.

The company's overall gross margin dropped from 75.0% to 71.1% in FY26. Management's explanation was clear: on one hand, the company's business model is shifting from Hopper HGX systems to Blackwell full-scale datacenter solutions; on the other hand, Q1 was also impacted by $4.5 billion in charges related to excess H20 inventory and purchase commitments.

Therefore, the key point here is not "competitiveness has declined," but rather:
The company is selling larger, more complex, complete solutions.

The benefits are, of course, also obvious:
Higher value per customer, deeper lock-in, stronger synergy between networking and computing, and a wider moat.

But the costs are equally real:
More complex cost structure, greater inventory pressure, more purchase commitments, and system-level products in the early ramp-up phase exert more pressure on gross margins than selling chips alone.

This is why I say that NVIDIA in FY26 is stronger than in FY25, but also "heavier."

This "becoming heavier" is reflected not only in gross margin but also in the risk structure.

Previously, you could understand NVIDIA more as a high-margin, high-growth, asset-light chip company, with its biggest risks being more about industry cycles, demand fluctuations, and changes in customer CapEx.

But by FY26, in addition to these risks, you must also start paying attention to:
• Supply chain forward commitment risks
• Inventory and purchase obligation risks
• Export restrictions and geopolitical policy risks
• Execution risks during the platform transition
• Bargaining power and volatility risks arising from increased customer concentration

The financial report has already given many signals.
For example, two direct customers accounted for 22% and 14% of total revenue in FY26, respectively, compared to 12%, 11%, and 11% in FY25; also, disclosed commitments reached $95.2 billion in FY26, compared to total future purchase commitments of approximately $45.1 billion disclosed in FY25. Cash, cash equivalents, and marketable securities on the balance sheet also increased from around $43.2 billion in FY25 to $62.6 billion in FY26, indicating the company is both scaling up and preparing for a heavier system.

Therefore, the H20 issue is essentially not simply "products not selling," but rather:
A business friction that occurred under the combined effects of policy environment + supply chain commitments + specific product roadmap.

Looking at FY25 and FY26 together, I think it can be summarized as follows:
FY25 was when NVIDIA proposed the idea of "platformization";
FY26 was when this idea truly began to develop a skeleton within the financial statements.

It no longer relies solely on a single GPU to conquer the world but is forming a more complete AI infrastructure platform:
GPUs, NVLink, Networking, CUDA, system integration, and software stacks are beginning to be packaged into a more irreplaceable whole.

This also means that when looking at NVIDIA in the future, we can no longer just focus on "whether next quarter's revenue exceeded expectations."
It's more important to simultaneously track four lines:
• The pace of AI infrastructure expansion
• Changes in export controls and geopolitics
• Gross margin and expense ratio trends
• Whether ecosystem control continues to strengthen

These four lines are what will truly determine NVIDIA's future valuation center.

As an investor, I believe AI infrastructure has just begun. NVIDIA's shift from selling cards to selling platforms essentially means it is increasingly firmly positioning itself as the "shovel seller" in this wave of AI mega infrastructure. In the future, whoever builds AI super factories or expands computing centers will most likely rely on NVIDIA's computing, networking, interconnect, and software ecosystem. Precisely because of this, I believe NVIDIA has the potential to become one of the primary beneficiaries of this AI infrastructure wave. As long as this trend does not fundamentally change, as long as the company's platform advantages and ecosystem control continue to strengthen, I will continue to firmly hold NVIDIA to share in the dividends brought by the AI era.
Personal thoughts, not investment advice.