Tesla Q2 2026 earnings: revenue hits a record $28.2B, EPS misses estimates

Tesla's Q2 results, released right after Wednesday's close, confirmed the exact tension flagged in the prior day's preview: how fast — and in which direction — the company is deploying capital. Revenue beat, but profitability fell well short of Wall Street's bar, turning the "spending pace" question into a margin problem.

Profitability metrics broadly missed. GAAP operating income fell 57% year-over-year to $398M (vs. $1.5B expected), with operating margin compressing to 1.4%, while adjusted EBITDA of $3.27B also came in below the $4.0B analysts expected. Tesla pointed to lower average selling prices, reduced regulatory-credit revenue, and higher AI, R&D, and stock-compensation expenses. Trailing-twelve-month revenue topped $100B for the first time.

Cash flow was the bright spot. Free cash flow came in at -$1.09B, a burn — but far smaller than the -$3.64B Wall Street had feared. CFO Vaibhav Taneja reaffirmed on the call that full-year capex will exceed $25B, covering Optimus, Cybercab, AI infrastructure, and the TeraFab chip plant being built with SpaceX and Intel. WSJ noted this quarter alone saw $5.8B deployed into AI and robotaxi infrastructure — a large chunk of the full-year guidance already spent by Q2.

Elon Musk again framed Tesla as a company looking "beyond the car business." Robotaxi launched unsupervised (no safety driver) rides in Dallas and Houston during the quarter and expanded into Miami, while Cybercab's steering-wheel-less public-road testing in Austin moved into its next validation phase. FSD subscribers grew 56% to 1.48 million. Optimus is "installing first-generation production lines" and will "start production soon," Musk said, though early units are for training-data collection and feature development — not customer deliveries. He called it "the hardest product to scale manufacturing we've ever made." Musk also addressed the AI hardware supply chain, thanking Micron for a "very significant" memory-chip allocation on reasonable terms despite what he called "pretty insane" current memory pricing — reassurance that AI-infrastructure buildout won't be memory-supply-constrained.