Is the robot fleet — or the FSD roadmap — actually funded? Tesla's FY2020 annual report, filed February 8, 2021 and located via SEC filings against the SEC 10-K (period ended December 31, 2020), centers its capital plan on construction and ramp at Gigafactory Shanghai, Gigafactory Berlin and Gigafactory Texas. Autonomy spend is not a separate fundraise; it rides on the manufacturing engine. The 10-K's own figures show that engine turning over for the first time at scale: total revenues of $31,536 million (up from $24,578 million in 2019), income from operations of $1,994 million (against a $69 million operating loss in 2019), and net income attributable to common stockholders of $721 million — Tesla's first full year of GAAP profitability.

That coupling is the story. Tesla's Autopilot and FSD development, and the compute investment behind it, are financed by the operating cash a growing car business generates. The cash-flow statement makes the scale of that turn concrete: net cash provided by operating activities jumped to $5,943 million in 2020 from $2,405 million in 2019, while capital expenditures (purchases of property and equipment, net of sales) were $3,157 million, up from $1,327 million. The MD&A frames the relationship directly: "Sustained growth has allowed our business to generally fund itself, but we will continue a number of capital-intensive projects in upcoming periods." Operating cash of $5.9 billion against $3.2 billion of capex is the literal margin out of which the autonomy program is paid.

For a capex desk, that is a cleaner discipline than a pure-play AV startup burning venture cash. Tesla's self-driving program competes for capital against the next production line. And the filing's own risk language flags the dependency from the factory side:

“The contingencies inherent in the construction of and ramp at new facilities such as Gigafactory Shanghai, Gigafactory Berlin and Gigafactory Texas may be exacerbated by these challenges.”— Tesla, Inc. 10-K (FY2020) source

Execution risk on the factories is, transitively, risk on the autonomy timeline. If a gigafactory ramp slips — on supply, construction, or demand — the operating cash that funds Autopilot, FSD and the compute behind them is what gets squeezed first.

It also reframes the bull case. R&D expense in 2020 was $1,491 million, about 5% of revenues — a modest line for a company promising a self-driving fortune, precisely because the heavy spending is in tooling, equipment and factory build-out rather than in an isolated autonomy budget. If you believe FSD is worth a fortune, the FY2020 filing says the cheapest way to fund that bet was to sell more cars first. The vehicle business is both the product and the financing vehicle for the software ambition stacked on top of it.

The forward guidance shows how disciplined that funding was meant to stay. Tesla told investors it "currently expect[s] our capital expenditures to be $4.50 to $6.00 billion in 2021 and each of the next two fiscal years," adding that the business "has recently been consistently generating cash flow from operations in excess of" its level of capex. A guided capex range that future operating cash was expected to cover is the clearest possible statement that the autonomy roadmap was to be self-financed. On the balance sheet, the company also ended 2020 with cash and cash equivalents up $13.12 billion from the end of 2019, reflecting capital raises during the year — a cushion that further de-risked the factory build-out without making autonomy a separately financed line.

The filing also makes clear that the autonomy program is not just a budget line but an in-house hardware-and-software stack, which is why it lives inside R&D and capex rather than as a purchased input. Tesla describes "expertise in developing technologies, systems and software to enable self-driving vehicles using primarily vision and radar-based sensors," and states that "Our FSD Computer runs our neural networks in our vehicles, and we are also developing additional computer hardware to better enable the massive amounts of field data captured by our vehicles." Building the inference computer, the neural networks, and the data pipeline internally is capital- and engineering-intensive — and all of it is absorbed into the same consolidated R&D ($1,491 million) and capex ($3,157 million) lines that the factory cash funds. There is no separately financed "autonomy" entity to point a balance sheet at; the program is structurally fused to the car business that pays for it.

What to watch through 2021: whether capex stays inside that $4.5–6.0 billion band as Berlin and Texas come online, and whether the cash from Shanghai's ramp is sufficient to keep both the factory build-out and the AI/autonomy investment funded without leaning on external capital. The cash-flow statement, not the AI Day slide, will answer it.

The honest limit: a capex line tied to factories does not tell you how much is going specifically to autonomy compute versus tooling and equipment. Tesla does not break that out, and R&D at 5% of revenue is a blended figure. What the filing does establish is the dependency — the self-driving roadmap is only as funded as the factories that pay for it, and the FY2020 numbers ($5.9 billion operating cash, $3.2 billion capex, first-ever annual profit) say those factories had just begun to pay. For a reader sizing the autonomy bet from the outside, the discipline cuts both ways: the same fusion of programs that makes Tesla's self-driving effort cheaply funded also means the FY2020 filing offers no way to measure its return on its own terms — only the consolidated evidence that the machine paying for it had, for the first time, started to generate more cash than it consumed.