Summary: Tesla reported Q3 2026 results on October 21, 2026, with revenue essentially in line at $28.046B and a slight beat on non-GAAP EPS ($0.48 vs $0.45 expected). The more interesting signals for builders are in the shareholder letter details: FSD (Supervised) attach above 75% of new North American deliveries, nearly two million cumulative unsupervised robotaxi miles, and a plan to reach 400 MW of in-house training compute by year-end.
The headline numbers
Per Teslarati’s earnings report, the quarter landed close to expectations:
- Revenue: $28.046B vs $28.34B expected
- Non-GAAP EPS: $0.48 vs $0.45 expected
- Operating margin: 2.6% vs 2.5% expected
- Operating income: $718M vs $721M expected
- Free cash flow: $1.531B vs roughly -$4.51B expected
The free cash flow figure is the outlier. Expectations implied a large cash burn; Tesla instead delivered positive FCF. That suggests the capex cycle assumed by consensus did not hit as hard this quarter, though the shareholder letter does note continued construction of the Austin fab.
FSD attach is the quiet story
The number that matters most for autonomy watchers is buried in the letter: FSD (Supervised) reached over 75% of new vehicle deliveries in North America, a nearly twenty percentage point sequential increase. Tesla says it believes FSD is driving demand in markets where it’s available.
From a product standpoint, that’s the shift from optional feature to default purchase driver. When three out of four buyers pay for the software, the pricing lever on autonomy becomes the core business question, not a side bet.
Robotaxi: supply and orchestration next
Tesla deployed Cybercab into its Austin robotaxi fleet during the quarter and reported nearly two million cumulative unsupervised miles by quarter-end. Operating hours run 6am to midnight, and the company says it expects to cut wait times through more fleet supply and improved orchestration algorithms.
One notable product decision: Grok is available in Cybercab with Bot and Connectors, so riders can manage inbox, calendar, files, and tasks hands-free. Tesla is framing the cabin as a productivity space, not just a ride. Whether anyone actually works in a moving car is untested, but it’s a real attempt at differentiating on in-cabin software.
It’s also worth watching the regulatory side separately. NHTSA has previously contacted Tesla after robotaxi incident videos, so the pace of fleet expansion will partly depend on regulators, not just orchestration code.
Compute plans and the robotics angle
Tesla plans to reach 400 MW of “SI training compute” by end of year, which the letter says supports development of both vehicle autonomy and humanoid robot software. The Austin fab is under construction with equipment commissioning targeted for 2027.
That last line connects the earnings to a larger bet. The compute buildout serves Optimus as much as the cars — a bet we’ve been tracking since Tesla’s Optimus chief Milan Kovac departed. Owning the training stack is the strategic claim here; the 2027 fab timeline is the practical constraint on it.
Takeaway
This was an in-line quarter, so the durable signals are structural: software attach rates climbing toward saturation in the most mature market, and a capital plan that treats training compute as core infrastructure. The limitation: most of what comes next — robotaxi scale, Optimus, the fab — lives beyond this quarter’s income statement.
Sources
View the original post on X