TSLA Q3 2026 — A Tariff Refund Paid for the Quarter; the Banks Are Lined Up for the Build

The setup

Tesla’s Q3 2026 Update leads with strength: “Strength in our core business also continued, and we achieved our second consecutive quarter of TTM revenue in excess of $100B.” The filing underneath it tells a narrower story. Revenue was $28.05B, flat against $28.10B a year earlier. Income from operations was $718M, down 56% from $1.62B, an operating margin of 2.56%. And the 10-Q shows that the quarter carried a one-time benefit: Tesla recognized $786M of refunds of IEEPA tariffs it had paid, $526M against energy cost of revenues and $260M against automotive cost of sales, after the Supreme Court invalidated those tariffs in February, and gave $339M of that back to energy-storage customers as a reduction to revenue. The net benefit to operating income was $447M, which is 62.3% of the quarter’s operating income.

Take the refunds out and Tesla earned $271M from operations on $28.39B of revenue before the customer refunds, a 0.95% margin. Put that in its series. Across the nine quarters from Q3 2024, the operating margin ran 10.79%, 6.16%, 2.06%, 4.10%, 5.78%, 5.66%, 4.20%, 1.41% and 2.56%; the four quarters from Q2 2023 to Q2 2024 we can isolate from Tesla’s XBRL filings ran between 5.50% and 9.62%. Q2 2026, at 1.41%, was the prior low. Q3 2026 without the refunds is lower than that, and no quarter in the thirteen we can isolate comes close to 0.95%. What it is not: a collapse in the car business. Automotive gross margin was 18.5%, up from 17.0%, and deliveries fell only 2%. The money went somewhere else, and the somewhere is research and development, which rose 69% to $2.76B.

That is the frame for the rest of the print. Tesla has turned into a company that spends on Super Intelligence, robots, chips and batteries, and pays for it with a car business that is no longer growing. The bill shows up twice: once in R&D, which was 3.85 times operating income this quarter, and once in capital expenditures, which ran $4.73B against $2.25B a year earlier. Our call: Tesla reports negative free cash flow for Q4 2026, wrong if the January Update shows Q4 operating cash flow at or above Q4 capital expenditures. The 10-Q’s own floor, capex “in excess of $20 billion in 2026,” leaves at least $6.99B for Q4, more than Tesla has ever generated from operations in a single quarter.

TSLA closed at $386.47 on October 21 and traded at $379.23 at 10:22 a.m. PT on October 22, down 1.87%.

The numbers

Read the right-hand side of the Sankey before the left. Interest income on the cash pile was $445M, 62.0% of what the operating business earned. Net income attributable to common stockholders was $787M against $1.37B a year earlier. The non-GAAP figure, $1.70B, adds back $982M of after-tax stock-based compensation and the $108M after-tax loss on the SpaceX stake, subtracts a $224M after-tax unrealized gain on bitcoin, and adds $44M of tax items. Stock-based compensation was $1.17B before tax, up 75.7% from $663M, in large part because of the 2025 CEO Performance Award. Adjusted EBITDA was $3.90B, a 13.9% margin against 15.0% a year earlier.

What’s actually driving it

The refund paid for the quarter, and the release does not say so. The Update’s revenue bridge lists the refunds once, as a drag: energy revenue fell “partially due to tariff-related customer refunds.” The cost side of the same event, which is larger, appears only in the 10-Q. Decode it line by line. Energy generation and storage gross profit was $954M at a 33.1% margin, up from 31.4%. Remove the $526M cost refund and restore the $339M owed to customers, and energy earned $767M on $3.22B, a 23.8% margin, which is 7.6 points below a year ago. Automotive cost of sales of $16.57B includes $260M of refunds; without them automotive sales gross margin was 16.24% against 14.71% a year earlier, still an improvement, but a smaller one than the reported line. Total gross profit without the net benefit was $5.05B, against $5.05B in Q3 2025. The year-on-year gross-profit gain disappears. The 10-Q says amounts recognized in the first half were immaterial, so the refund is a Q3 event, not a run rate, and Q4 will be measured against a base that included it. This is the gap between the narrative and what the data supports in its plainest form: the Update calls the core business strong, and the core business, measured without a court ruling, earned about the same gross profit as a year ago and about a sixth of its operating income.

R&D is the cost line that moved, and it moved every quarter. The 10-Q puts the increase down to “costs related to SI and other programs, such as Optimus, silicon development, battery technologies, and new vehicle programs,” plus $178M more stock-based compensation. Research and development rose from $1.04B in Q3 2024 to $2.76B in Q3 2026, rising in each of the nine quarters, and it went from 5.80% of revenue a year ago to 9.84%. Total operating expenses rose 39.2% to $4.78B on revenue that did not grow. Income from operations went the other way: $2.72B, $1.58B, $399M, $923M, $1.62B, $1.41B, $941M, $398M, $718M. R&D has been larger than operating income in every quarter since Q1 2025, and in Q3 2026 it was 3.85 times it. A one-period change is a phase, not a decision, until you have the series; nine quarters of R&D rising while revenue goes sideways is a decision, and management has named it.

The name changed and the spending did not. A year ago the Q3 2025 Update said: “over time, we expect our hardware-related profits to be accompanied by an acceleration of AI, software and fleet-based profits.” The Q3 2026 Outlook repeats the sentence with one word changed: “an acceleration of SI, software and fleet-based profits.” Decode what a year of that sentence produced. Services and other gross profit was a record $734M, up from $366M, and the Update credits FSD subscriptions for higher automotive ancillary sales. Active FSD subscriptions reached 1.74M, up 67% from 1.04M, and the Q3 add of 0.26M was the largest of the six quarters disclosed. The Robotaxi service reached nearly 2M cumulative unsupervised miles and began carrying riders in Cybercabs in Austin on September 3. That is real progress on the software side. It is also small against the bill: the record services gross profit is 26.6% of the quarter’s R&D. Tesla does not disclose FSD revenue, so the reader cannot divide the subscriptions by the dollars. The acceleration the sentence promises has not yet shown up as profit in a line the filing reports.

Read the car business by its ingredients. Automotive sales revenue is units times price, and Tesla discloses enough to approximate both. Deliveries were 486,532, down 2.1% from 497,099; 6,911 of them went on operating leases. Dividing automotive sales revenue by the 479,621 vehicles sold outright gives about $41,896 a vehicle, against $41,816 a year earlier. Dividing automotive cost of sales the same way gives $34,548, against $35,667, or $35,090 before the $260M refund. So the margin gain on cars came mostly from cost and mix, not price, and about half of the cost gain was the refund. Regulatory credit revenue, which is almost pure profit, fell 73% to $113M; across the nine quarters from Q3 2024 it ran $739M, $692M, $595M, $439M, $417M, $542M, $380M, $146M and $113M. In Q1 2025 regulatory credits of $595M were larger than operating income of $399M; in Q3 2026 they were 15.7% of it. Read the ingredients, not the level: the automotive margin is up, but price is flat, units are down, and the free credit money is nearly gone.

The geography moved more than the total. The Q3 Update says: “We believe that FSD is driving demand in markets where it is available.” Decode it against the segment note. Revenue from the United States, the market where FSD (Supervised) has been available longest, fell 14.4% to $12.50B from $14.60B. China fell 7.9% to $5.21B. Other international markets rose 31.8% to $10.34B from $7.84B, helped by record deliveries in Canada, Australia, Japan and Taiwan and by nine EU countries approving FSD. Part of the US decline is likely the energy business, where the customer refunds sit. The filing does not split the US decline by segment, so we do not. What the geography does show is that the 2% delivery decline was a US problem offset abroad, not a stable global picture.

The energy business has the order book; the quarter does not show it. The Update calls the order rate and backlog the highest ever, and the 10-Q backs the claim with numbers. The transaction price allocated to unsatisfied energy performance obligations with terms over one year was $10.73B at September 30, up from $10.05B at June 30, with $4.80B expected within twelve months. Customer prepayments held as deferred revenue were $3.30B, against $2.31B at June 30 and $2.04B at December 31. Energy inventory was $4.45B against $2.71B at December 31, up 63.9%. Deployments were 13.7 GWh, up from 12.5 GWh. All of those are leading indicators pointing up, and the lagging line went the other way: energy revenue fell 16% because of the customer refunds, a lower average selling price per Megapack, fewer Powerwalls and fewer solar retrofits. Megafactory Texas, with 50 GWh of planned capacity, started production in the quarter. The next two quarters should show whether the backlog converts at the old margin or the new one; without the refunds, it is the new one, 23.8%.

One of the energy customers is SpaceX, and the receivable is the size of a quarter’s sales to it. Tesla recognized $221M of Megapack revenue from SpaceX in Q3, 7.7% of energy revenue, and $626M across nine months, at $171M and $478M of cost. At September 30 the receivable from SpaceX was $729M, of which $588M was billed for Megapacks not yet delivered, matched by deferred revenue; net of that, the receivable would be $141M and the deferred revenue $60M. Tesla also owns SpaceX stock: it invested $2.00B in March, converted from a preferred investment in xAI, for under 1% of the company. The stake was worth $2.87B at September 30 under the fair-value option, a $863M unrealized gain over nine months that reversed by $142M in Q3. Our SpaceX Q3 dive covers the other side of that ledger. The 10-Q adds one sentence that matters more than any of these figures: the Tesla board has formed a special committee of disinterested directors “to explore deepening Tesla’s relationship with SpaceX, which could be through certain arrangements including without limitation a business combination.” The filing says there is no assurance a transaction happens and that any deal would need the committee’s recommendation and possibly a shareholder vote. It gives no size, structure or timetable.

The operating cash flow was a record, and the payables made it. Q3 operating cash flow was $6.26B, the highest in the fifteen quarters the Update tabulates, edging Q3 2024’s $6,255M. Decode what produced it. Accounts payable, accrued and other liabilities added $3.99B of cash in the quarter, and deferred revenue added $1.26B; against them, prepaid expenses and other assets absorbed $1.82B, receivables $565M and inventory $313M. Working capital in total contributed $2.59B, so operating cash flow before working capital was about $3.67B. Accounts payable rose to $17.24B from $15.32B at June 30, and days payable outstanding rose to 66 from 58. Part of that is the build: suppliers and contractors for the Texas factories and the compute clusters are owed money at quarter end. Part of the deferred revenue is customer prepayment for Megapacks. Either way, the payable is the loan you don’t see: the record cash quarter was funded in part by Tesla’s suppliers, and payables give the cash back when they are paid.

Depreciation is the cost the capex will write. Depreciation, amortization and impairment was $1.85B in Q3, up 14.0% from $1.63B and up from $1.62B in Q2, the largest quarterly step in the fifteen-quarter table. Property, plant and equipment, net, was $50.59B, up 28.4% from $39.41B a year earlier; long-lived assets in the United States rose to $45.24B from $35.85B at December 31. Capex ran 2.56 times depreciation in Q3. The Supporting Infrastructure page explains where it goes: Cortex 1 and Cortex 2 training clusters at more than 225 MW combined, a plan to reach 400 MW of training compute by year-end, 4680 cell lines, LFP cells in Nevada, cathode and lithium refining in Texas, and the Austin Fab, where Tesla plans to start commissioning equipment in 2027. The carrying cost is senior to the shareholder: every dollar of that build comes back through cost of revenues and operating expense over its life, ahead of any profit the products it enables might earn.

The capex guide moved down while the bank lines moved up. Here is the guide in the order Tesla filed it. The 10-K in January: “in excess of $20 billion in 2026.” The Q1 10-Q in April and the Q2 10-Q in July: “in excess of $25 billion.” The Q3 10-Q: “in excess of $20 billion,” with “AI” replaced by “SI.” Decode the arithmetic. Tesla spent $13.01B in the first nine months. Applied to nine-month spending, the July floor would have required at least $11.99B in Q4; the October floor requires at least $6.99B, after $4.73B in Q3. Some of the cut is spending that slipped into 2027, some may be spending Tesla decided not to do; the filing does not say which. In the same quarter, on September 29, Tesla signed $30.0B of new unsecured bank facilities: a $20.0B three-year delayed-draw term loan led by Citibank, an $8.0B five-year revolver and a $2.0B 364-day revolver led by Wells Fargo, with room to add $4.0B more to the revolvers. They replaced a $5.0B revolver that was never drawn. Tesla “does not currently plan to draw on the facilities in 2026.”

A delayed-draw term loan is a plan with a date on it. The structure tells you when Tesla expects to need the money. The term loan can be drawn up to ten times in eighteen months. Undrawn commitments fall to $10.0B on the first anniversary and to $5.0B after fifteen months, and expire after eighteen. A company that wanted only a liquidity backstop would sign the revolvers, which it did, and stop there. A $20.0B commitment that shrinks if unused is sized and timed for spending in late 2026 and 2027. The Q2 10-Q introduced a phrase that the Q3 10-Q repeats: Tesla intends to keep a strong balance sheet and sufficient liquidity, “which may include additional funding.” Decode it with the cash: $45.45B of cash and investments, $9.52B of debt and finance leases, almost all non-recourse, and a covenant requiring at least $5.0B of liquidity. Tesla does not need to borrow to survive. It is arranging to borrow so the SI build does not have to wait for the car business to fund it. On who ultimately pays for the capex, Tesla’s answer has changed in two quarters from “the cash flow” to “the cash flow, then the banks.” The filing does not connect the facilities to the SpaceX committee, and neither do we; both are new this quarter, and both are big enough to change the balance sheet.

The award sets the scoreboard. The 2025 CEO Performance Award, described again in the 10-Q, ties tranches to operational milestones that include 20 million vehicles delivered, 10 million active FSD subscriptions, 1 million bots delivered, 1 million Robotaxis in commercial operation, and Adjusted EBITDA milestones starting at $50 billion. Against them: cumulative deliveries are 10.2M, FSD subscriptions 1.74M, and trailing Adjusted EBITDA $15.00B, down from $16.63B at its Q4 2023 high. The first EBITDA milestone is 3.33 times the current trailing figure. The award is why stock-based compensation rose 75.7%, and it is also the clearest statement of what management is paid to deliver: not the car business alone, which on trailing deliveries of 1,742,908 would take about 5.6 more years to reach 20 million, but the four milestones that depend on software and robots.

Here is the bull case at full strength. Tesla produced a record $6.26B of operating cash flow and positive free cash flow of $1.53B while spending $2.76B on R&D and $4.73B on capex, and ended the quarter with $45.45B of cash and investments against $9.52B of mostly non-recourse debt. The car business is not deteriorating: automotive gross margin rose to 18.5%, days of supply were 11, and Tesla says it ended the quarter with the largest passenger-vehicle backlog in four years, limited by battery supply rather than demand. FSD subscriptions grew 67%, the Robotaxi fleet is carrying riders in purpose-built Cybercabs, Optimus production started in Fremont, and the energy business has $10.73B of contracted backlog and $3.30B of customer prepayments. The capex guide went down, not up. On that reading, a margin of 2.56% is what a company looks like at the bottom of an investment cycle that is starting to ship products, and the $30.0B of bank lines are insurance. We land the other way for one reason: the products that ship are not yet in a reported profit line, and the line that pays for them, operating income from cars and energy, was $271M in Q3 without a court refund. A company can carry that for a long time on $45.45B of cash. It cannot do it while also funding at least $6.99B of capex in a single quarter out of operations.

The guide and the Q4 bar

Tesla does not give revenue, margin or delivery guidance. The Outlook says it is “focused on maximum capacity utilization at our factories” and that volumes depend on demand, supply readiness and how many vehicles it keeps for its own fleet. The one number it gives is the capex floor, so that is the bar.

Compute what Q4 has to deliver. Capex of at least $6.99B means operating cash flow of at least $6.99B for free cash flow not to be negative. The best quarter in the fifteen-quarter table is Q3 2026’s $6.26B, so Q4 would need a new record by 11.5%. The fourth quarters in the table produced $4.37B in 2023, $4.81B in 2024 and $3.81B in 2025. To reach $6.99B from Q4 2025’s $3.81B, operating cash flow would have to rise 83.2% on the year. Two things could do it: a second large working-capital inflow like Q3’s $2.59B, or capex landing exactly on the floor while customers prepay heavily for Megapacks. Neither is impossible. Both would have to happen at once, in a quarter that loses the tariff refund and carries another step up in R&D. On Q3’s numbers, operating cash flow before working capital of about $3.67B against a $6.99B capex floor leaves a gap of about $3.31B for working capital to fill.

The full-year view says the same thing in another form. Trailing free cash flow was $3.30B at Q3, down from $6.83B a year earlier and from $7.00B at Q1 2026, on trailing operating cash flow of $18.71B and capex of $15.41B. As a rough check, a 2026 capex total of $20.00B set against that trailing operating cash flow would put annual free cash flow at about −$1.29B. The 2026 figure Tesla publishes in January is the receipt for this call: a negative Q4 is the most likely outcome on the company’s own floor.

How peers read this

Tesla now looks less like a carmaker and more like a smaller version of the hyperscaler problem we wrote up this morning in Meta Q3 2026. In both, a mature core business (ads there, cars here) is being asked to fund a compute and infrastructure build whose revenue is not yet in a filed line, and in both the company has started arranging outside capital in the same quarter its free cash flow came under pressure. The differences run in Tesla’s favor on the balance sheet and against it on the core. Meta’s ad engine grew 26% and earned 35.4% before its legal charge; Tesla’s revenue was flat and earned 0.95% before its refund. Meta signed $810.25B of commitments it cannot cancel; Tesla’s spending is mostly its own capex, which it can slow, and it carries $35.92B of net cash rather than net debt.

Against SpaceX, the read is direct rather than analogous. SpaceX is a Tesla customer, a Tesla investment and now the subject of a Tesla board committee. A combination would put a company whose cars no longer fund its R&D next to one whose launch business funds its satellite and AI build. Until the committee reports, the only numbers that bind are the ones in the 10-Q: $221M of Q3 Megapack sales, a $729M receivable, and a $2.87B stake.

For suppliers, the read is mixed. Battery availability, Tesla says, is still the limit on vehicle volume, which is constructive for cell makers and the lithium chain. But the energy-storage tariff refunds are a reminder that the price of imported cells is a policy variable: the $526M Tesla recovered in Q3 is cost it carried for more than a year. And the compute build at 400 MW by year-end is a buyer of accelerators, memory and power gear on the same order as a mid-sized neocloud.

Valuation reality check

At $379.23 on 3,952,757,946 shares outstanding at October 15, the cover-page count, market value is about $1.50T. The cover count is above the 3,541M diluted weighted average used for EPS; we use it because it is the figure that prices the whole company today. Cash and investments of $45.45B less $9.52B of debt and finance leases leaves $35.92B of net cash, so enterprise value is about $1.46T. On the last four quarters, that is 416.7 times GAAP diluted EPS of $0.91 and 220.5 times non-GAAP EPS of $1.72. Enterprise value is 97.6 times trailing Adjusted EBITDA of $15.00B and 14.1 times trailing revenue of $103.57B. Market value is 453.8 times trailing free cash flow of $3.30B.

None of those multiples describes a car company, and none was meant to. The market is pricing the four award milestones that are not about cars: FSD at scale, Robotaxi, Optimus and an Adjusted EBITDA line more than three times today’s. The filing gives the reader three ways to track that: FSD subscriptions (1.74M), cumulative unsupervised Robotaxi miles (nearly 2M), and services and other gross profit ($734M). None is yet large enough to be seen in operating income. The valuation question this print raises is narrower than the usual one: not whether the multiple is too high, but how long the car business can keep paying for the bet out of its own cash, and the Q3 answer is about one more quarter before outside money arrives on the balance sheet or capex slows again.

What we’d watch from here

• Q4 2026 free cash flow in the January Update. That is the call: negative. Positive would mean operating cash flow of at least $6.99B against the capex floor, a record by 11.5%.
• The first draw on the $20.0B term loan. Tesla says no draw in 2026. A draw in 2027, before commitments step down to $10.0B in September 2027, confirms the build is outrunning operating cash.
• Operating income without one-time items against $271M. The tariff refund will not repeat. Q4 operating income below Q3’s $718M on comparable revenue says the refund was the margin.
• R&D against $2.76B and as a share of revenue against 9.84%. A first quarterly decline in the series would be the first sign of a ceiling on the SI spend.
• Energy gross margin against 23.8% excluding refunds, with the $10.73B backlog and $3.30B of prepayments as the test of whether Megafactory Texas ships at the old margin.
• Days payable against 66 and payables against $17.24B. A fall gives back part of Q3’s $2.59B working-capital inflow.
• The SpaceX special committee. Any filed terms, and any change in the $729M receivable or the $2.87B stake.

The trade

Evidence-led, not a chase. A 1.87% move on the reaction day says the market priced the print as roughly neutral, which is reasonable for a quarter whose headline gain was a refund and whose news was a board committee and a bank line. The event that tells you the most is not the next delivery number but the January Update’s cash-flow table: a negative Q4 free cash flow alongside a 2027 capex figure larger than 2026’s would confirm that the build has moved to outside funding. A holder who wants exposure to the SI and robotics optionality and is uneasy about the funding could size so that a term-loan draw or an equity-linked financing is not a forced decision. We have not priced strikes here.

The risk sits in Tesla’s own notes: a 2.56% operating margin with a non-recurring refund inside it, a capex floor above any quarter’s operating cash flow, $30.0B of new bank lines with a draw window that closes in March 2028, a related-party customer with a $729M receivable, and a board committee exploring a combination with that customer.

Bottom line

The Q3 2026 Update says the core business stayed strong. The 10-Q shows a quarter in which a court-ordered tariff refund supplied 62.3% of operating income, R&D was 3.85 times operating income, and the record operating cash flow leaned on $2.59B of working capital. Tesla cut its capex floor to $20B and signed $30.0B of bank facilities in the same quarter, and its board began exploring a deal with SpaceX. The car business still earns money. It no longer earns enough to pay for what Tesla is building. Our call: Tesla reports negative free cash flow for Q4 2026. Wrong if the January Update shows Q4 operating cash flow at or above Q4 capital expenditures.