Connect the dots and you get a quietly useful pattern. NVIDIA's description of its own business has stayed remarkably stable across three annual reports, even as the financials underneath it moved a lot. That stability is the story.
In FY2024 (filed 2024-02-21), the sec.gov filing describes accelerated computing for "computationally intensive workloads such as artificial intelligence, or AI, model training and inference, data." The FY2025 report repeats the "training and inference" framing almost verbatim, and the FY2026 report does too — all three surfaced together via EdgarBeast.
Three documents, one sentence, barely changed. When a company keeps describing itself the same way under oath, that's the framing to trust over any quarter's press release. It tells you the core thesis — sell general-purpose acceleration, with AI as the largest workload — is not a marketing pivot but a durable self-understanding.
What did change is the scale. NVIDIA's reported revenue climbed across these same filing years, while the workload sentence held still. That gap — stable description, rising numbers — is the cleanest way to see that the AI boom showed up as volume, not as a redefinition of the product. You can read the FY2026 framing in the sec.gov filing and compare it line for line.
The lesson for any reader trying to cut through AI noise: track the self-description, not just the headline number. A company that quietly rewrites what it claims to do is telling you something; a company that holds its definition steady while it grows is telling you something else. NVIDIA, for three years running, has been the second kind.