Pay Attention to Tesla: Why EV Sales Miss the Autonomy and Energy Story

Headlines say Tesla is losing. Deliveries fell. BYD took the EV crown. Panic is the product. That framing is wrong, and it is wrong for a reason you can measure.

Tesla delivered about 1.64 million vehicles in 2025, down from 1.79 million in 2024—roughly an 8.5% year-over-year drop. Q4 landed near 418,000, about 15.6% below the same quarter a year earlier, and short of the Street’s ~440,000 expectation. BYD hit 2.26 million pure-electric sales, up 28%, with overseas volume crossing 1 million for the first time. Those numbers are real. Pretending they are not is useless.

Now ask the question the doom pieces skip. If Tesla is “dying,” why did the equity finish 2025 near a $1.5 trillion valuation and up roughly 11% while deliveries fell? Why sit at ~279 times earnings while Apple runs near 30, Microsoft near 35, and Nvidia—actually printing record AI-chip revenue—trades nearer 50? Either every pension, hedge fund, and long-horizon holder has gone collectively insane, or the market is pricing something that is not quarterly hatchback volume. Farzad’s read is the second one.

Cars are the data machine, not the endgame

Tesla is not competing to win a pure unit-sales trophy. It sells cars to fund AI, energy, and robotics—and every car on the road feeds the Full Self-Driving stack. Owners drive. Interventions, clean miles, and ugly edge cases stream back. Construction zones that rearrange overnight. Scooters blowing red lights in Austin. Merges where everyone cuts everyone else. You cannot hand-code that catalog. You accumulate it.

The fleet has logged more than 7 billion FSD miles. On the current path, that stack approaches about 10 billion by around mid-2026—the scale Elon has pointed to as the unsupervised threshold. Timelines will slip. They always have. The miles do not pause while headlines argue. Waymo, Cruise (effectively gone), Zoox, Pony, Mobileye, and the Chinese autonomy names are not collecting customer-fleet data at anything close to that volume. In machine learning, algorithms and silicon without data lose. Data is the moat.

Supervised FSD already does something almost no other production automaker offers at scale: parking-spot-to-parking-spot driving that people actually use. That alone is a product advantage while the unsupervised bet matures. Even if robotaxi ramps slowly and Optimus takes longer than promised—which it will—the cars keep generating cash and training signal.

Energy is the quiet growth engine

While every front page obsesses over car deliveries, Tesla Energy just posted its best year. Q4 deployed about 14.2 GWh. Full-year storage hit roughly 46.7 GWh—more than double the prior year. Gross margin on energy in Q4 printed about 31.4%. Automotive sat near 16.1%. The business everyone doomscrolls runs half the margin of the business almost nobody covers.

Capacity is stacking: a Houston facility aimed at adding ~50 GWh, Lathrop and Shanghai in the mix, a path toward roughly 133 GWh of annual manufacturing capacity when those pieces come online, plus a ~10 GWh cell factory in Nevada. Mega Block is pitched at ~23% faster installs and ~40% lower cost versus the current Megapack generation. Wright’s law—costs fall as cumulative production doubles—is already familiar from solar, batteries, and cars. It is showing up again in grid storage.

That matters because AI data centers are starving for power. Demand is not linear. Microsoft is working to restart Three Mile Island—the plant that scared the country in 1979—just to feed compute. Whoever can ship utility-scale batteries at falling cost sits in the middle of that scramble. Energy growing at something like ~84% year over year is not a side quest. It is a second P&L that can carry a large slice of today’s valuation while automotive absorbs the narrative heat.

BYD won units. That is not the same race.

BYD outsold Tesla on pure electric units. Fact. Also fact: BYD’s gross margin fell to around 16% in Q2 2025, and profit dropped about 30% year over year—the first profit decline in three years. More cars, less money. Flood the market, take share, hope the math works later. Classic volume playbook, often backstopped by opaque subsidy structures that make clean apples-to-apples comparison hard.

BYD also spans cheap cars, premium cars, buses, trucks, commercial fleet, even odd industrial products. Stacking their total units against Tesla’s is like stacking all of Android against the iPhone and declaring Apple lost. Apple still takes the profit pool. Different segments, different economics.

Tesla could chase a $12–15k down-market car in China, India, or Southeast Asia. It has the manufacturing and battery stack. It chooses not to—for now—because the prize it is optimizing for is autonomy, robotics, and energy, not an EV sales crown that commoditizes as margins compress industry-wide. If Tesla ever decided to torch margin for units in markets it enters, the fight would look very different. That is not the game on the board.

Robotaxi: forty cars vs a factory that can build millions

Go back to December 2025. Tesla began unsupervised robotaxi testing in Austin—no safety driver, empty front seat, city streets, left turns, pedestrians, highway merges. Elon and AI lead Ashok Elluswamy posted about riding those cars around Christmas. Footage from downtown Austin showed passengers in the back like an Uber with nobody ready to grab a wheel. About forty robotaxis are operating in Austin now, with 2026 expansion talk for cities such as LA, San Francisco, Miami, and Phoenix. Cybercab mass-production targets have pointed at April 2026; expect slips, but the hardware path is visible.

Waymo is real. Hundreds of thousands of driverless rides per week across multiple cities is not vapor. Its fleet is still under ~3,000 vehicles, with annual capacity targets on the order of ~10,000. Tesla’s manufacturing base can still turn out on the order of 2 million cars a year even after a soft delivery year. That is roughly a 200x production gap. Waymo needs HD maps and geofences, expensive sensor suites, and city-by-city rollout. Tesla’s bet is cameras plus a neural net that improves from every customer mile worldwide—Austin feeding Toronto feeding Berlin—without remapping every neighborhood first. Which stack scales to Buenos Aires, Jakarta, or Lagos without a mapping army? That is the actual debate. Unit EV crowns are a distraction from it.

Optimus working on the factory floor is the other physical AI signal investors watched in 2025. Proof of concept is becoming floor reality. Slow is still directionally different from vaporware press releases.

Innovator’s dilemma—running in reverse

Clayton Christensen described how winners die: protect the cash cow, ignore the inferior tech that becomes superior, optimize yesterday’s metric until the next wave owns the market. Kodak invented the digital camera and buried it to defend film. Blockbuster passed on Netflix. Nokia owned phones and missed smartphones.

Tesla is doing the opposite on purpose. It is letting BYD wear the EV sales crown while it spends attention and capital on unsupervised autonomy, humanoids, and storage margins. That looks like spin if you only score 2025 deliveries. Strategically, the EV market is already squeezing—BYD’s own profit drop is the receipt. Autonomy, robotics, and grid storage are the trillion-scale arenas. Legacy auto cannot easily make that pivot; boards, analysts, and quarterly delivery scoreboards trap them. Elon has said out loud, repeatedly, that he does not care about hitting Street delivery prints. That is either a feature for long-horizon holders or a disaster. It is not the boring playbook that killed the last generation of innovators.

What to watch instead of the rearview mirror

Car-sales headlines will keep coming. The stock can sell off hard on a soft automotive print. That is how the tape works: headline, panic, questions later. The questions that actually move the long-term story are different. Can automotive margins hold while energy keeps scaling? Where is Cybercab production really? How many FSD miles pile up per day? What is the Optimus factory deployment timeline? When does unsupervised driving leave the Austin sandbox?

In five years, almost nobody will care who sold the most EVs in 2025. They will care who owns the most capable autonomous fleet, who has useful humanoids in volume, and who has the most deployed storage. On those axes, Tesla is building while the media scores a different contest.

Pay attention to the right scoreboard. The scarce assets are fleet learning and energy margins—not a quarterly delivery trophy.

Check the video here.