R&D capitalized at scale answers the funding question — disclosure answers nothing about mix. Nvidia's FY2025 annual report, filed February 26, 2025 and surfaced via SEC filings against the SEC 10-K (period ended January 26, 2025), reports research-and-development expense of $12,914 million for the fiscal year — up 49% from $8,675 million, and equal to 9.9% of net revenue. The same filing describes automotive platforms, autonomous and electric-vehicle solutions, and Jetson for robotics among its offerings. The R&D is enormous; the catch for analysts is that the autonomy-and-robotics slice stays blended into a much larger whole.

Start with the funded-list read. The scale of the surrounding business is almost hard to overstate from the filing: total revenue of $130,497 million, up 114% year over year; gross margin of 75.0%; operating income of $81,453 million, up 147%; and net cash provided by operating activities of $64,089 million. As of January 26, 2025 the company held $43.2 billion in cash, cash equivalents and marketable securities. At $12.9 billion of annual R&D against $64 billion of operating cash flow, Nvidia can finance an autonomy and robotics roadmap out of operating spend many times over. Unlike a pure-play AV company budgeting against a venture runway, Nvidia's physical-AI ambitions are a rounding error against its own research budget. The roadmap is funded; that is not in question.

What is in question is how much of it is autonomy and robotics at all — and here the filing gives one rare, telling number. Automotive revenue was $1,694 million in FY2025, up from $1,091 million in FY2024, an increase the 10-K says was "up 55% from a year ago, driven by sales of our self-driving platforms." That growth rate is striking; the absolute size is the point. At $1.69 billion, the entire Automotive line is roughly 1.3% of Nvidia's $130.5 billion in revenue. The self-driving and robotics business that anchors the "arms-dealer of physical AI" thesis is, in the financial statements, a small fraction of a company overwhelmingly defined by Data Center.

“The Compute & Networking segment includes our Data Center accelerated computing platforms and AI solutions and software; networking; automotive platforms and autonomous and electric vehicle solutions; Jetson for robotics and other embedded platforms; and DGX Cloud computing services.”— NVIDIA Corporation 10-K (FY2025) source

That single sentence is the disclosure problem in miniature. Nvidia reports two operating segments — Compute & Networking and Graphics — and folds automotive, autonomous-vehicle solutions, and Jetson robotics into the same Compute & Networking segment as the AI-accelerator data-center business that produces the vast majority of revenue. An outside analyst can see the $1.69 billion Automotive product line, but cannot isolate autonomy-and-robotics R&D, gross margin, or operating profit from the much larger AI-accelerator business they are bundled with.

To see how lopsided the bundle is, put the numbers side by side. Data Center revenue was $115,186 million in FY2025, up from $47,525 million a year earlier — on its own, roughly 68 times the size of the $1,694 million Automotive line it shares a segment with. The 10-K even attributes the company's gross-margin expansion to "a higher mix of Data Center revenue," which is to say the headline profitability that funds everything is being set by the AI-accelerator business, not by autonomy. NVIDIA does disclose the autonomy effort qualitatively: the filing describes a "full self-driving capability as well as an open, modular DRIVE software platform for autonomous driving, mapping, and parking services," and warns that automotive "design wins" "involve a lengthy process" and "do not guarantee" future revenue. That is real product and real risk — just not real segment-level disclosure.

That blending is strategically coherent — Nvidia genuinely treats autonomy, robotics, and data-center AI as one compute story, and its history bears that out: the filing notes the company shipped its first autonomous-driving system-on-chip in 2018 and that the 2020 Mellanox acquisition expanded its canvas into networking. But coherence frustrates valuation. The physical-AI thesis rests on a business the filing largely does not size beyond a single product-line revenue figure, so investors are asked to take the robotics-and-autonomy option on faith in the platform rather than on a disclosed segment with its own margins.

It is worth being precise about what "funded many times over" does and does not mean here. It means NVIDIA faces no financing constraint on its autonomy ambitions — $64 billion of operating cash flow and $43.2 billion of liquidity dwarf any plausible robotics R&D bill. It does not mean the autonomy business is large, profitable, or even separately measured; on the numbers the filing provides, it is none of those things yet. The funding question and the size question have opposite answers, and conflating them is the most common error in valuing the "physical AI" story.

What to track: any move toward separate disclosure of robotics or autonomy as those businesses scale, and whether Automotive's 55% growth rate stays high enough for long enough to eventually demand its own detail. A line growing 55% off a $1.7 billion base can stay buried; a line compounding toward $10 billion-plus eventually cannot. Until then, the $12.9 billion R&D figure and $64 billion of operating cash flow tell you the bet is funded, not how big the autonomy piece of it is.

The honest limit: aggregate R&D, a single Automotive revenue figure, and a blended platform description establish capacity, intent and direction — not the size or economics of the autonomy-and-robotics business. FY2025 confirms Nvidia can fund the bet many times over, that the self-driving-driven Automotive line is growing fast, and that the company has chosen, so far, to keep robotics and autonomy inside a segment dominated by something else entirely. For the capex desk, the practical consequence is that any valuation of NVIDIA's "physical AI" optionality has to be built on the $1.69 billion Automotive disclosure, its 55% growth, and the qualitative DRIVE-platform language — not on a margin or a return on the autonomy R&D, because the filing simply does not provide one.