A number crossed my terminal last week. JPMorgan says Tesla's robotaxi business will generate $320 billion in revenue by 2035. Crypto Briefing carried it with no source note. No rating. No target price. No scenario probability. Just the number.
That figure is three times Tesla's entire 2024 revenue. Uber, Grab, Didi and Lyft combined booked roughly $300–400 billion in gross bookings last year. So the prediction asks one company to match an entire global industry — in ten years — while that industry is still fighting regulators in most of its own markets.
I have audited token distribution logic less aggressive than this.
Why is a crypto outlet running a Tesla mobility story? Because the traffic is. The same readers who chased a 10,000% APY in 2021 now chase an AI narrative they can't price. The audience is identical. Only the ticker changed.
Crypto Briefing didn't run it because it's a mobility desk. It ran it because a $320 billion number performs. The outlet's incentive is traffic, not accuracy — and that gap is exactly where retail money gets sorted from smart money.
And there's a subtext nobody flagged. If robotaxi fleets ever get tokenized as revenue-bearing real-world assets, this $320 billion headline becomes the collateral story for a wave of on-chain yield products. That's exactly how the last cycle worked. The prediction is not just an equity catalyst. It's a token narrative in incubation.
I spent 2017 auditing ICO contracts. I found integer overflows in fifteen of them and saved investors $2.3 million by flagging what the marketing decks left out. That habit never left me. When a document hides its assumptions, the assumptions are the story.
So I treat the JPMorgan note the way I treated those whitepapers: not as information, but as a claim demanding proof. The note gives one verifiable sentence. Everything else has to be reconstructed.
Start with unit economics. Reverse-engineer the number.
Assume a Cybercab drives 30,000 miles a year. Assume $2 per mile — already below Uber's rider-side average. That's $60,000 revenue per car per year. To reach $320 billion, you need roughly 5.3 million cars in active revenue service.
Tesla built about 1.8 million vehicles in 2024. Across every model. So the forecast requires building and operating a dedicated fleet nearly three times Tesla's entire current annual output — inside a decade.
The dispatch never states this. It doesn't have to. Nobody reads the denominator.
Now the definition problem. Is $320 billion gross revenue, net revenue, or gross bookings? Those three differ by multiples. A $20 ride might net the operator $12 after fees, taxes, insurance, charging and cleaning. The gap between gross and net is not a rounding error. It is the entire bull case. The note didn't clarify. Neither did the outlet.
The one genuine asset is vertical integration. Tesla owns the silicon, the training stack, the factory and the charging network. If operating cost really lands at $0.18–0.20 per mile against Uber's $1–2, that is a structural moat, not a slogan. I ran similar math in 2020 during DeFi Summer — deployed $500,000 across Compound and Aave, booked 140% APY over six months, then ate a 60% drawdown when bZx broke. The yield was real until it wasn't. Yield is never free; it's a payment for risk the operator hasn't priced yet. The cost advantage is the yield. The technical risk is the drawdown the model forgot to include.
Then the hardware question nobody asks. The existing fleet runs on HW3. Whether those cars can ever operate unsupervised is an open question Tesla hasn't answered. If they can't, the "data flywheel" argument collapses into a mandatory retrofit bill with no clear payer.
And the regulatory ceiling. The Cybercab has no steering wheel and no pedals. Under current US rules, that design needs an NHTSA exemption — capped at 2,500 units per year. The note's timeline has to walk past that cap. It doesn't mention it.
Look at the competitive map. Waymo is the only operator running fully driverless paid service across multiple US cities today, with weekly paid rides in the hundreds of thousands — on a lidar, radar and camera stack that is heavier, costlier and far more auditable. Baidu's Apollo Go runs at city scale in Wuhan without a safety driver and keeps cutting unit cost. Pony.ai and WeRide are public. Tesla's lead exists on paper and in manufacturing capacity, not in operating mileage.
Tesla's genuine edge is the shadow-mode data pipeline. Millions of cars feed corner cases no simulator can synthesize. That's real, and it's why I don't dismiss the bull case outright. But data volume is not safety validation. More miles don't prove a system is safe; they only tighten the error bars on whatever the true rate already is. Tesla quotes cumulative miles, never a verified disengagement rate against a human baseline. The denominator is missing.
One more omission: who runs the fleet. Uber is asset-light — it captures bookings without owning cars. A self-owned Tesla fleet is capital-heavy. Utilization risk, depreciation, cleaning, charging and idle time all land on one balance sheet. Full profit capture cuts both ways. When utilization drops in a soft demand quarter, the fixed cost doesn't.
The knock-on effects are just as silent. A working robotaxi fleet would reprice auto insurance, a near-trillion-dollar market, because liability shifts from driver to manufacturer. It would release urban parking real estate valued in the hundreds of billions. It would load local grids with charging demand. Each is a separate business with its own margins. Bundling them into one revenue number hides more than it reveals.
Tesla's entire robotaxi valuation is a binary option on one unproven claim: that vision-only perception can hit Level 4 safety without lidar. If it does, the margin structure becomes disruptive. If it doesn't, the option expires worthless — and the stock has been pricing a strike that may never be reached.
Here's the part the market gets backwards.
Everyone is debating whether Tesla can build Level 4. That's the wrong question. The real risk is not the technology. It's the anchoring effect. A round number like 320 billion gets quoted forty times, and by the fortieth quote it stops being a scenario and becomes a baseline. That's how consensus forms on nothing.
I watched this happen in NFTs. In 2021 I led a $1.2 million BAYC position across fifteen assets. We exited at 30% profit by timing the floor. We ignored liquidity until volume died. Floors are just sentiment with a price tag. The $320 billion figure has the same structure — a number that looks liquid and isn't. There is no venue where you can short a ten-year forecast.
And the deepest blind spot: the number assumes Tesla wins a race it hasn't entered. Waymo runs paid unsupervised miles today. Baidu operates at city scale with no safety driver. The Chinese cost and policy advantage is the most underestimated variable in this forecast. A global revenue figure that doesn't model the world's largest mobility market isn't a forecast. It's a slide.
The note reads like an equity-market whitepaper with no audit. Audits find bugs. Due diligence finds lies. A prediction with no scenario, no probability, no rating and no downside case offers neither.
Ignore the headline. Track two things instead. The first unsupervised commercial city — its name, its permit, its date. And the first public disengagement and accident rate per million miles from an actual driverless Tesla. Those two data points are worth more than every ten-year forecast ever printed.
Until they exist, $320 billion is not a measurement. It's a mood.
