Record Deliveries, a 1.4% Operating Margin: The Investment Cycle Has Arrived on the Income Statement

Results vs. Consensus

Quality of Beat/Miss

• Revenue: The +26% print is real and broad: record Q2 deliveries (+25% YoY), Services +50% on FSD subscriptions, insurance and supercharging, and Energy +13% on the second-best deployment quarter ever (13.5 GWh). Trailing-twelve-month revenue crossed $100B for the first time. The beat vs. Street was substantially pre-announced, though: the market has known the 480K delivery number since July 2, so print-day price discovery was about margins, and margins are where the quarter failed.
• Margins: Three separate compressions landed at once. Auto GM ex-credits fell to 16.3%, with the CFO's own bridge conceding the underlying trend is flat at that level, not the 17.5–18% recovery track the Q1 headline implied. Regulatory credits, nearly pure margin, fell 67% YoY to $146M. And energy GM halved sequentially to 20.4% on a $240M warranty true-up for vendor cell issues plus non-repeating Q1 tariff benefits, with management now guiding the business to normalize in the mid-to-low 20s. Services was the lone bright spot at a record 14.1%.
• EPS: The $0.17 miss vs. the ~$0.50 consensus is operational, not below-the-line: OpEx grew 47% YoY (R&D +49% on Semi/Optimus/Cybercab pre-production and AI compute depreciation, plus unquantified litigation charges) against gross profit growth of 23%. Management explicitly guided OpEx to keep growing "in 2026 and beyond." This is not a one-quarter timing artifact; it is the declared cost structure of the investment cycle.
• Cash: Operating cash flow of $4.7B (+85% YoY) was genuinely strong, helped by inventory release as deliveries outran production by ~28K units. The FCF swing to negative $1.1B is entirely capex, which more than doubled sequentially to $5.79B and is guided to increase further in H2.

Segment Performance

Automotive: The Demand Recovery Is Real; the Margin Recovery Was Not

Auto revenue of $20.5B (+23% YoY) on record Q2 deliveries of 480,126 fully validated the "demand resurgence" framing that Q1 left as an open question. Sequential delivery growth ran +60% in the Americas, +27% in APAC and +12% in EMEA; Model Y set records in the Netherlands, Australia and New Zealand, and the deck lists record deliveries across eleven markets. The Q1 inventory build cleared completely: deliveries exceeded production by ~28K units and days of supply fell from 27 to 15. Management says the order backlog is now the largest since 2023, and production, not demand, is the constraint, gated by battery pack capacity and, newly flagged, electronic components.

"Automotive margins, excluding regulatory credits, declined sequentially from 19.2% to 16.3%. As a reminder, we had highlighted in Q1 that we had a $230 million benefit from warranty true-downs and some tariff relief, which did not repeat in Q2. Controlling for the impact of those benefits from the prior quarter, our automotive gross margins, excluding credits would have been approximately flat." — Vaibhav Taneja, CFO

That paragraph is the most consequential disclosure of the print. In April, the 19.2% headline was presented with one-time caveats we sized at roughly 130–170bps, implying an underlying 17.5–18%. The Q2 bridge now says the clean number was approximately 16.3% in both quarters. Two new structural headwinds also surfaced: rising commodity prices, and interest-rate subvention costs that are recognized upfront as a revenue offset, so higher rates bite margins immediately. ASPs ex-FX declined on mix, and regulatory credits fell to $146M from $439M a year ago, a fading subsidy that was still worth roughly 60bps of GAAP auto margin this quarter versus 220bps a year ago.

Assessment: Auto is now a volume story with a flat margin base, not a margin-recovery story. The bull case's load-bearing wall (structural margin repair via mix, FSD attach and in-house packs) has to be rebuilt from a 16.3% base against subvention and commodity headwinds. The offset is genuine: a record backlog, supply-constrained volume and FSD attach above 55% of North American deliveries mean the revenue line compounds while the margin line idles. But at Tesla's multiple, margin trajectory is the thesis, and it just reset.

Energy: Volume Delivered, Normalization Arrived Early

Energy deployed 13.5 GWh (+41% YoY, +53% sequentially), the second-best quarter ever, with record EMEA deployments supported by Megafactory Shanghai's ramp. Revenue of $3.1B grew 13% YoY. The margin, though, halved sequentially from 39.5% to 20.4% on three stacked items: a $240M warranty true-up for vendor cell issues in legacy deployments, the non-repeat of Q1's $200M+ tariff benefits, and declining industrial storage ASPs amid growing competition. Management set the long-term normalized bar explicitly: mid-to-low 20s gross margin. Megafactory Texas (Megapack 3, Megablock) is nearing completion with production planned this year, and the demand framing shifted decisively toward AI data centers, including power-smoothing for training runs.

Assessment: We flagged after Q1 that the 39.5% record was one-time flattered and the 2026 energy story would be volume at compressed margins; Q2 delivered exactly that, plus a warranty charge that reminds investors legacy fleet liabilities are not zero. The volume trajectory (H1 22.3 GWh vs. FY25's 46.7 total) supports the "higher than 2025" guide, and the data-center demand thesis is credible and strategically aligned with the AI buildout. Model the segment at mid-20s margins; treat anything above as upside.

Services & Other: The Record Nobody Is Pricing

Services grew 50% YoY to $4.6B, now 16% of revenue, with gross profit up $302M sequentially to a record $648M at a record 14.1% margin. Growth came from all major sub-businesses: used cars, supercharging, service centers, insurance, and the FSD subscription tail. The segment also absorbs deliberate Robotaxi infrastructure investment, meaning the reported margin understates the mature-state economics.

Assessment: Two consecutive quarters of margin step-up (8.8% → 9.2% → 14.1%) make Services the cleanest structural-improvement line in the P&L and the natural landing zone for autonomy revenue as it materializes. Services now contributes $648M of quarterly gross profit, roughly a fifth of automotive's ex-credit gross profit, up from about 7% a year ago, and almost nobody models it as a driver.

Key KPIs

Key Topics & Management Commentary

Overall Management Tone: Management leaned all the way into the investment-cycle framing, presenting the collapsed operating margin as the deliberate price of "the most ambitious build-out of advanced infrastructure manufacturing capacity ever in history" rather than defending it as temporary. The CFO was notably disciplined: he pre-bridged the margin decline, quantified the one-times in both directions, named new headwinds (subvention, commodities) unprompted, and disclosed the $30B debt capacity without being asked. That candor is analytically useful and slightly ominous; companies arrange $30B of borrowing capacity when they expect to need it. Guidance language was expansive on products and silent on economics, continuing the pattern of shifting attention to operational metrics (miles, metros, attach rates) while financial disclosure on the new businesses stays thin. Tone was the most future-tilted of the past five quarters, on a print where the present-day P&L most needed defending.

The Margin Reset: 19.2% to 16.3%, and the "Approximately Flat" Admission

The single most important number in the print is 16.3% auto gross margin ex-credits, and the single most important sentence is the CFO's concession that stripping Q1's one-time benefits, margins were "approximately flat" sequentially. This retroactively re-bases the entire margin-recovery narrative: Q1's 19.2% included $230M of warranty and tariff one-times, so the clean level was already near 16.3% then, below Q4's reported 17.9%, and Q2 simply held that lower base. The sequential story was never a step-up. New headwinds are structural, not episodic: interest-rate subvention costs recognized upfront against revenue, rising commodity costs, and mix-driven ASP decline as the affordable trims and FSD-led buyers dominate the order book.

"Commodity price increases and interest rate changes all continue to add to our costs. Note that interest rate subvention costs are recognized upfront as revenue offset. So, as interest rates have risen this year, the cost of subvention has risen along with them, which had a negative impact on automotive margins." — Vaibhav Taneja, CFO

Assessment: Our April thesis carried auto margin recovery as a confirmed pillar with a 17%-clean floor as the downgrade trigger. The floor broke, not because Q2 deteriorated but because the floor was never where the headlines put it. Credit management for the transparency; debit the thesis. Until FSD monetization and pack-cost reduction visibly outrun subvention and commodities, the correct model is flat margins on rising volume.

Regulatory Credits Fall Off a Cliff

Regulatory credit revenue fell to $146M from $439M a year ago and $380M in Q1, the lowest quarterly figure in years. This is close to pure margin: the YoY decline alone is a ~$293M headwind to gross profit, worth more than a full point of auto margin, and reflects the post-IRA policy environment in which US emissions-credit demand has structurally shrunk. Management did not discuss it on the call; the deck lists "lower regulatory credit revenue" in both the revenue and profitability YoY bridges.

Assessment: The credit annuity that funded $1.5–2B a year of near-pure gross profit in recent years is ending. It hurts optically now, but there is a silver lining: forward margins are increasingly organic, and the quality of the remaining margin base is higher. Model credits at ~$100–150M/quarter declining, and stop treating them as recurring.

The Capex Cycle Extends: Over $25B This Year, Growing Two to Three More Years, $30B of Debt Capacity

Q2 capex of $5.79B more than doubled sequentially, H1 totals $8.28B, and the unchanged "more than $25B" FY guide therefore implies H2 spend above $16.7B, an $8B+ quarterly run-rate. The genuinely new disclosure is duration and funding: capex "will grow for the next two to three years," and Tesla has arranged debt facilities providing capacity to borrow up to $30B. PP&E grew $4.0B sequentially to $47.3B. The spend map spans the Robotaxi fleet, Optimus capacity, the semiconductor fab, solar manufacturing, AI compute and automotive expansion; on-site AI training compute more than doubled in H1 (Cortex 2 above 115 MW).

"CapEx will grow for the next two to three years as we expand our robotaxi fleet, expand our production capacity for Optimus, make investments for semiconductor fab, install solar manufacturing capacity and AI compute infrastructure… In addition to using our cash for such investments, we are being opportunistic in securing certain debt facilities that will give us the capacity to borrow up to $30 billion to help accelerate such investments." — Vaibhav Taneja, CFO

Assessment: In April the question was whether the trough was $20B or $25B for one year. The answer is now: $25B+ this year, rising for two to three years, partially debt-funded. Every line item still maps to a real product, and the balance sheet ($43.5B cash) can carry it. But the FCF inflection we originally dated to H2 2027 now credibly sits in 2028, and the equity story must survive two-plus years of negative-to-breakeven free cash flow. That is a different risk profile than the one we underwrote at Outperform, and the market's reaction (a broad AI-capex de-rate that also hit the hyperscalers) says we are not alone in re-pricing it.

Demand Resurgence Validated: Record Q2, Largest Backlog Since 2023, FSD as the Demand Driver

Q1's open question, whether the 50K-unit inventory build reflected soft demand, was answered emphatically: record Q2 deliveries, inventory cleared, days of supply nearly halved, and the largest order backlog since 2023. The demand driver management emphasized most is FSD itself: North America attach exceeded 55% of new deliveries, and elevated interest is visible in every market where FSD gets approved (four more European countries this quarter: Lithuania, Estonia, Denmark and Belgium, on top of the Netherlands).

"For a lot of people, they're actually buying Tesla full self-driving with a car attached as opposed to a car with FSD. They're coming into our stores in the U.S. and they want the full self-driving and with whatever car comes with essentially." — Elon Musk, CEO

Assessment: This is the strongest order-book position since 2023 and it decisively kills the structural-demand-decline bear case. The constraint has flipped to supply: battery pack capacity and now electronic components gate H2 volume. The FSD-pull dynamic matters strategically because it converts vehicle demand into a software-attach funnel, raising lifetime revenue per unit even while ASPs decline on mix. It is the one place where the margin story could re-inflect from here.

Robotaxi: 380,000 Unsupervised Miles, Seven Metros, Compounding More Than 10% a Week

The Robotaxi disclosure took a real step forward operationally: more than 380,000 cumulative unsupervised miles across six cities in two states (Austin, Dallas, Houston, plus Miami, Orlando and Tampa launched unsupervised in July), the Bay Area operating with safety drivers under a California TCP permit, and Phoenix and Las Vegas in preparation. Management characterized weekly unsupervised miles as compounding at more than 10% a week, with the growth expected to continue through year-end. The fleet runs early v15 builds with roughly 40% of the planned improvement tracks merged. The stated model for expansion: per-city launch effort trending toward zero, with the end-state being state-wide operation.

"We have driven more than 380,000 miles of unsupervised robotaxi now across six cities in two different states. We have had zero notable incidents… Historically, the so-called experts have always claimed that you need LiDARs, radars, HD maps and the entire kitchen sink to drive safely. Here, we show that such is not true." — Ashok Elluswamy, VP of AI

Assessment: The safety record and the compounding-miles disclosure are real progress, and the seven-metro footprint slightly exceeds the pace we modeled in April. But notice the metric migration: Q4's "25–50% of US population" became Q1's "dozen states by year-end" became Q2's cumulative-miles growth rate, and the fleet remains dozens of vehicles per city rather than hundreds. Miles-per-vehicle is genuinely high because robotaxis run near-continuously, but three quarters into commercial operation there is still no revenue, cost-per-mile or fleet-count disclosure. The operational execution earns full marks; the financial opacity is now the story.

Cybercab: In Production, With a Chassis-Data Gate on Scale

Cybercab production at Gigafactory Texas continued to ramp from its April start; engineering test drives of production units on public roads began in the quarter, and employee rides on the Texas campus started in July. Musk disclosed a scaling gate that had not been articulated before: because Cybercab is a new chassis, Tesla must accumulate chassis-specific driving data using units retrofitted with steering wheels and pedals before deploying volume, after which "the number of Cybercabs in cities will increase dramatically." Manufacturing targets are deliberately aligned to the projected growth of unsupervised miles.

Assessment: The April SOP commitment held and the deployment sequence (production, public-road validation, employee rides, fleet entry) is proceeding on a coherent path. The chassis-data gate is a sensible engineering reality that also conveniently defers the volume question. The >125K installed annual capacity is there; 2026 output will be a small fraction of it. Watch for first customer-facing Cybercab rides as the Q4 signpost.

Optimus: Timeline Slips Again as the Supply Chain Gets Built From Scratch

Optimus production slipped from Q1's "late July, August time frame" to "later this year": the Fremont S/X lines are decommissioned and first-generation Optimus lines are being installed, with initial builds earmarked for the internal Optimus Academy (training-data collection), not customers. Musk devoted his longest opening-remarks segment to recalibrating expectations, calling Optimus "the hardest product to scale manufacturing that we've ever made" because no humanoid supply chain exists, and framing the ramp as a long, flat initial S-curve. Further out, Optimus 4, to be built in Austin with far deeper vertical integration, carries an aspirational 10M units/year target versus 1M for Optimus 3. Ashok detailed the data strategy: observing factory workers, a dedicated demonstration team, internet-scale video, then reinforcement learning in the Academy.

"The production scaling challenge is very substantial. This is going to be the hardest product to scale manufacturing that we've ever made at Tesla because everything on the robot is new. And the difficulty of scaling the production ramp is proportionate to the newness of the parts in the robot." — Elon Musk, CEO

Assessment: Two consecutive quarters of slippage (V3 reveal pushed in Q1, production start pushed in Q2) now separate the Optimus narrative from the Optimus schedule. The honest difficulty framing is welcome and probably accurate, but it means Optimus contributes essentially nothing to 2026-27 economics while absorbing a growing share of the capex and R&D lines. We trim our FY27 unit range again and treat Optimus as a free option, not a modeled business.

Energy and the AI Power Thesis: Megapacks, Megapods and Doubling US Output

Beyond the quarter's numbers, management repositioned the energy business as AI infrastructure. Musk's demand framing: hyperscalers cannot smooth the violent power cycling of training runs (consumption can drop 70% for 100 milliseconds) without fast-acting battery power electronics, which is why SpaceX has bought large volumes of Megapacks for its data centers. He also floated a striking arithmetic claim: US generation capacity of roughly 1.2–1.3 TW against average usage of ~0.5 TW means batteries alone could nearly double effective US energy output. A new product concept, Megapods (AI4 compute paired with x86 in Megapack-style enclosures, deployable at Superchargers with their ~7 GW of grid connections), extends the same logic to distributed inference compute.

Assessment: The data-center demand thesis for storage is the most under-appreciated element of the print: it gives the energy business a growth vector uncorrelated with EV demand, and Megafactory Texas's Megapack 3 ramp is the near-term catalyst. But note who the marquee customer keeps being: SpaceX. The related-party density of the growth narrative keeps rising.

The SpaceX Entanglement Deepens

The SpaceX relationship now touches nearly every layer of the story: the $2.0B equity stake bought in Q1 produced a $1.0B unrealized gain that is the difference between a good-looking and bad-looking GAAP net income line; a framework agreement (disclosed by the General Counsel) governs collaboration on Terafab and Digital Optimus; Grok orchestrates Digital Optimus; Starlink is being integrated into Cybercab and eventually all vehicles for Robotaxi connectivity; and SpaceX is a major Megapack buyer. Asked directly about the persistent combination chatter, Musk declined to engage on the call, deferring to "the appropriate process." Terafab details, including its location, were explicitly deferred to a standalone announcement event.

Assessment: Each individual link has industrial logic. Collectively, they create a valuation and governance problem: Tesla's GAAP earnings now move with the private marks of its CEO's other company, its growth narrative depends on related-party demand, and the most consequential capital project (Terafab) will be structured across both entities. The refusal to engage the combination question, while procedurally correct, confirms it is live. We treat related-party concentration as a bear pillar that is materializing, not merely emerging.

The Semiconductor Flank: Samsung, TSMC, Micron and the Development Fab

Tesla's chip supply chain got unusually specific airtime: Samsung's Texas fab and TSMC's Arizona fab are investing "tens of billions" substantially in service of Tesla AI compute; Micron granted Tesla a significant memory allocation "on reasonable terms given the pretty insane pricing of memory these days"; Panasonic added billions in cell capacity. Equipment orders are placed for the Austin development fab, designed to house lithography mask production, logic, memory, packaging and test under one roof for fast chip iteration. On silicon: an incremental AI4+ (moderate improvement over AI4) and AI5 both target production around mid-2027, with AI5 going first into Optimus, and Musk committed to upgrading all camera-equipped cars below Hardware 4 to a next-generation AI board when financially sensible.

Assessment: The memory-allocation disclosure is the sleeper item: in a DRAM/HBM-scarce market, guaranteed Micron supply is a real competitive asset for scaling Optimus and Robotaxi compute. The HW3 answer also quietly firmed: the retrofit program announced in Q1 is now framed as upgrading the entire sub-HW4 camera fleet to next-gen boards, which converts a liability into a monetizable upgrade cycle if executed. AI5 volume at mid-2027 keeps the chip roadmap credible but pushes its P&L relevance past our rating horizon.

Guidance & Outlook

Implied Q-over-Q ramp: With H1 deliveries of 838K and the largest backlog since 2023 against explicit supply constraints (battery packs, electronic components), we model Q3 deliveries of 490–520K and FY26 at roughly 1.90–1.98M (+9–13% YoY). The margin math is harsher: H1 non-GAAP EPS of $0.74, plus an H2 that adds capex-driven depreciation, near-zero credits and energy margins in the low 20s, points to FY26 non-GAAP EPS of roughly $1.40–1.60 even with strong volume. FCF: H1 net +$0.35B; H2 capex above $16.7B implies FY26 FCF of approximately $(7)B–$(5)B unless operating cash flow accelerates dramatically.

Street at: Consensus entered the print at ~$2.30+ FY26 EPS and will have to come down 30%+. The revenue line, conversely, gets marked up: $28B quarters with a record backlog support $110B+ FY26 revenue against prior marks near $100B.

Guidance style: Consistent with pattern: no volume guide, no EPS guide, no FCF quantification, but unusually candid margin bridging and an unprompted debt-capacity disclosure. Management is guiding the spend precisely and the returns not at all. The asymmetry is the message: they want the market underwriting the buildout, not the near-term earnings.

Analyst Q&A Highlights

Robotaxi Fleet Size vs. City Count

The sharpest exchange of the call pressed the gap between the expanding city map and the small number of vehicles actually deployed, reportedly dozens per city rather than hundreds, and asked why Tesla does not simply saturate one market first. Management's answer came in four layers: the expansion pattern is deliberately breadth-first to prove the software stack generalizes without per-city engineering; robotaxis run near-continuously so a small fleet generates outsized miles; operational kinks are being worked out at controlled scale before mass deployment; and city-by-city regulatory requirements force sequential market entry anyway.

Q: "If I look at the launches, it's been — you've been adding cities, but the number of units is still — it looks like based on media reports, sort of in the dozens as opposed to hundreds. Why not just sort of scale up Austin or one or two cities before adding cities? And sort of what do you need to sort of get that higher volume numbers in a major city?"
— Colin Langan, Wells Fargo

A: "The reason we have been expanding across different cities instead of just doubling down on a single city is that we want to make sure that our stack is a very general one… since these vehicles operating in the robotaxi fleet drive basically continuously as opposed to human drivers who use vehicles for maybe a couple of hours a day or something like that, these vehicles are in mostly continuous operation, which means that even for a few vehicles, you can get a lot of miles out of them."
— Ashok Elluswamy, VP of AI

Assessment: The generalization argument is technically coherent and the miles-per-vehicle point is fair, but the exchange confirmed the fleet remains very small three quarters into commercial operation. The honest summary: Tesla is still in validation, not scaling, and the "exponential" framing describes a growth rate off a tiny base. The Cybercab chassis-data gate disclosed later in the same thread reinforces that volume deployment is a 2027 event.

The SpaceX Combination Question

For the first time on an earnings call, the persistent market chatter about combining Tesla and SpaceX was put to the CEO directly. He acknowledged the growing overlap, declined to discuss a combination on the call, and handed the question to the General Counsel, who disclosed that the two companies deepened their relationship "earlier this year" through the equity investment and a framework agreement governing joint projects.

Q: "There's been a lot of chatter about SpaceX and Tesla combining in some way. I mean, a lot of collaboration already today. I mean, Elon, do you see synergies eventually from combining the companies?"
— Colin Langan, Wells Fargo

A: "As you can tell from the many collaborations on so many fronts with SpaceX… there's more and more overlap, especially with Terafab, that's really going to be a gigantic project. So — but obviously, we can't talk about combining companies and that kind of thing on an earnings call. It's got to be done with the appropriate process."
— Elon Musk, CEO

Assessment: A non-denial with process language is the most newsworthy possible answer short of confirmation. Combined with the framework agreement, the $2B stake, the $1B mark now flowing through Tesla's GAAP earnings, Starlink integration and SpaceX Megapack purchases, the boundary between the two companies is visibly dissolving. Investors should assume some form of deeper structural linkage is under consideration and price the governance risk accordingly.

Capex Pace and Capital Efficiency

A well-framed question asked whether the pace of capex is effectively capped by the efficiency of deployment, and whether spending is now the binding unlock on Tesla's supply constraints. The CEO's answer was unusually candid about the trade-off: Tesla is deliberately accepting lower capital efficiency in exchange for speed, on NPV grounds. The CFO added that Tesla acts as its own general contractor on nearly all construction and that every dollar targets productive assets.

Q: "To what extent right now in this CapEx cycle is your pace of CapEx effectively capped simply because spending any more would be done in an inefficient manner? And to what extent is the pace of CapEx really the primary determinant right now in unlocking the different supply constraints that you have?"
— Dan Levy, Barclays

A: "We should be spending on CapEx as fast as we can… without it being too wasteful. So we're not trying to aim for like some extremely high efficiency capital spend because that would slow things down… it's okay to be a little less capital efficient if we get things done sooner because that's actually going to be the higher NPV outcome for the company… I think probably this is the fastest industrial scale-up since World War II in America."
— Elon Musk, CEO

Assessment: "Speed over capital efficiency, justified by NPV" is the entire bull-bear divide in one sentence. If the terminal businesses (Robotaxi, Optimus, energy-for-AI) are as large as management believes, the trade is obviously right; if any of them disappoint, Tesla will have built the world's most expensive optionality portfolio partly on borrowed money. The answer removed any remaining ambiguity about whether spending discipline will moderate the cycle: it will not.

Hardware Upgrade Path and the AI Chip Timeline

A two-part question probed Tesla's propensity to make its own silicon for Optimus and whether the Hardware 3 upgrade program remains the plan for running v15-class FSD. Management confirmed the upgrade intent in its broadest form yet: all camera-equipped cars below Hardware 4 become upgrade candidates to a next-generation AI board, with the incremental AI4+ chip and AI5 both targeting production around mid-2027.

Q: "The plan to upgrade Hardware 3 to Hardware 4 in order to be able to process, I suppose it's version 15 of the — and future versions of FSD. Is that still the plan to do hardware upgrades on the cars?"
— William Stein, Truist Securities

A: "I think it's going to make sense to upgrade all cars that have less than hardware 4, almost any cars that have cameras basically… anything that's set up for cameras, it will be financially sensible at some point to upgrade them. And I think we want to upgrade them to the next generation of AI board."
— Elon Musk, CEO

Assessment: The HW3 liability we flagged in April is being reframed as a fleet-wide upgrade cycle, skipping HW4 as retrofit target in favor of next-gen boards arriving mid-2027. The economics remain undisclosed (who pays, at what price), but the v14 lite release to AI3 cars in the US and South Korea this quarter shows the interim mitigation shipping. The class-action tail risk persists; the operational answer is now visibly in motion.

Optimus Supply Chain and Partner Investment

The opening question of Q&A asked whether suppliers are investing alongside Tesla to scale Optimus domestically, given the absence of an existing humanoid supply chain. The answer surfaced the quarter's most concrete ecosystem detail: Samsung and TSMC building US fabs with substantial Tesla-dedicated capacity, Micron granting a significant memory allocation, Panasonic expanding cells, and specialized component investment (metal injection molding, flexible printed circuits) forming around the robot, with Tesla in-sourcing wherever no capable partner exists.

Q: "Are you seeing any suppliers willing to invest alongside you to set up domestic manufacturing to potentially help you scale more quickly in the U.S.?… from an investment cycle standpoint, potentially more capital efficient to have some partners help you along the way."
— Andrew Percoco, Morgan Stanley

A: "Our suppliers have been great… Samsung and TSMC, in particular, are building fabs at TSMC in Arizona and Samsung in Texas and putting in tens of billions of dollars to build AI compute for Optimus and robotaxi. Panasonic has also invested many billions in increasing battery cell production."
— Elon Musk, CEO

Assessment: The supplier ecosystem forming around Optimus partially offsets the capex-intensity concern: tens of billions of partner capital is being deployed that does not sit on Tesla's balance sheet. The Micron allocation, volunteered unprompted, is the notable specific: memory scarcity is the binding constraint on edge-AI scaling, and Tesla just told the market it has secured supply.

Federal Regulatory Unlocks for Cybercab

With the NHTSA administrator publicly signaling openness to removing steering-wheel and pedal requirements, management was asked what else is needed at the federal level to unlock the Cybercab ramp. The vehicle-engineering lead's answer was flatly that nothing is: the FMVSS modernization is progressing, the NHTSA relationship is collaborative, and the friction sits at the state level, with New Jersey singled out as "disheartening."

Q: "The NHTSA administrator, Jonathan Morrison, was just on CNBC… it seemed like he wants to get rid of the steering wheel. It seemed like that would be a gating factor. But is there other things that you need at the federal level to really kind of unlock your ability to ramp Cybercab?"
— Walter Piecyk, LightShed Partners

A: "The short answer is no. I mean I think we have a great relationship with NHTSA and administrator Morrison, especially… We've been open on this with them for the last couple of years about what our plans were and what we're doing. And I don't want to say we're in lockstep, but I feel like we have a partner there, and we're working together on it."
— Lars Moravy, VP of Vehicle Engineering

Assessment: Federal regulatory risk to the Cybercab thesis is lower than commonly modeled; the real map is 50 state-level negotiations, which is why the Robotaxi rollout tracks state boundaries (Texas, Florida, Arizona, Nevada) where frameworks are permissive. The unprompted New Jersey comment flags the emerging counter-risk: states writing prescriptive sensor requirements that would functionally exclude Tesla's camera-only stack.

Energy Storage: Data Centers vs. the Grid

The closing exchange asked how energy storage demand splits between data centers and utility peak-shaving, and whether supply stays constrained. The CEO's answer reframed the segment around AI: training-run power cycling requires battery smoothing that hyperscalers cannot get elsewhere, SpaceX buys Megapacks for exactly this, and batteries could in principle nearly double effective US power output by arbitraging the gap between installed generation (~1.2–1.3 TW) and average usage (~0.5 TW).

Q: "To what extent is energy storage going to be supply constrained for the foreseeable future?… what is that demand curve right now between data centers versus just broader peak shaving for utilities?"
— Dan Levy, Barclays

A: "During a training run, the power consumption can drop by 70% for 100 milliseconds. And you really need fast-acting advanced power electronics to be able to smooth out these massive changes in power… you could potentially double the energy output in the United States just with batteries. So really, we see demand for Megapack being very, very high in the future."
— Elon Musk, CEO

Assessment: The AI-power framing gives energy a secular demand narrative detached from EV cycles and positions Megapack 3 as an AI-infrastructure product. The margin guide (mid-to-low 20s) caps the profitability upside, but volume visibility improved. Note again the circularity: the flagship data-center customer cited is SpaceX.

What They're NOT Saying

1. Robotaxi unit economics, third consecutive quarter: Cumulative paid miles got a chart; revenue, cost per mile and fleet count got nothing. Each quarter the operational disclosure improves while the financial disclosure stays at zero. A business claimed to be compounding at 10%+ weekly should have a revenue line by Q4.
2. The "dozen states by year-end" target: April's headline Robotaxi guide was not reiterated. The current footprint is seven metros in three states with two more in preparation. The quiet successor metric is unsupervised-miles growth, which is the third metric migration in three quarters.
3. China FSD: The Q1 call set a Q3 2026 target for broader China approval. The word "China" appeared on this call only in the context of Megafactory Shanghai. Silence on a previously dated regulatory milestone is rarely good news.
4. Roadster: April's "may be able to debut in a month or so" produced no demo; the deck lists the program as design development. A small item, but a clean data point on promise-to-delivery conversion.
5. FY26 FCF quantification: "Negative for the rest of the year" survives, but with H2 capex above $16.7B implied, the plausible FY26 FCF range is $(7)B–$(5)B and management declined to size it. The $30B debt capacity disclosure is the indirect answer.
6. Litigation charges: Flagged as an OpEx driver, never sized. Given the docket (FSD marketing claims, the HW3 base, autonomy incidents), an unquantified legal accrual deserves a number.
7. Optimus V3 reveal: Pushed from Q1 2026 to mid-2026 in April; mid-2026 has now passed with no reveal and no new date. Production "later this year" is doing unsupported work.
8. Terafab economics and location: Deferred to a standalone launch event. Deliberate, and defensible as communications strategy, but the largest capital commitment in the plan remains entirely un-underwritable.
9. FY26 volume guide: Fifth consecutive quarter without one, even with a record backlog that would make a guide easy to clear. "Focused on maximum capacity utilization" is the closest offered.

Market Reaction

• Pre-print setup: TSLA closed at $374.01 on July 22, down 16.8% YTD (vs. S&P 500 +9.5%), down 7.7% over the trailing 30 days, and up 12.6% over the trailing twelve months. The 52-week closing range is $302.63–$489.88; the stock entered the print roughly mid-range, having already de-rated through July on capex anxiety.
• After-hours move: Shares fell roughly 5.6% in after-hours trading, from $374.01 into the mid-$350s. The decline built in two legs: the 4:05 PM ET release landed the EPS miss and the 1.4% operating margin; the 5:30 PM ET call added the two-to-three-year capex horizon and the $30B debt-capacity disclosure.
• At publication (Thursday session, live): The sell-off extended into Thursday's open, with the decline running to low-double-digit percentages intraday amid a broader AI-capex de-rate that also hit megacap AI spenders; the session had not closed at publication time.

The reaction is rational and, notably, not idiosyncratic. Tesla's print landed the same evening the market began repricing the entire AI-capex complex, and Tesla is now underwritten as an AI-capex story: the quarter's message was record demand, vanishing near-term margin, and a spend curve that rises for two to three more years partly on debt. The delivery beat was three weeks old; the margin base reset and the capex-duration extension were new. A stock carrying a ~215x trailing multiple into a 30%+ forward-estimate reset does not need a company-specific catastrophe to fall hard; it needs exactly what it got, confirmation that the earnings trough is deeper and longer than the consensus model assumed.

Street Perspective

Debate: Is the 2–3 Year Capex Cycle Value Creation or Value Destruction?

Bull view: Every dollar maps to a named factory or capability with a visible product behind it, partner capital (Samsung, TSMC, Micron, Panasonic) is co-investing tens of billions off Tesla's balance sheet, and the CEO's speed-over-efficiency logic is how platform monopolies get built. The $30B debt capacity at Tesla's balance-sheet quality is cheap optionality, not distress.

Bear view: Operating margin is 1.4% today, the FCF inflection has moved out to 2028, and the returns on this spend are asserted, never quantified: no Robotaxi economics, no Optimus date that has held, no Terafab underwriting. Funding a negative-FCF buildout with debt while GAAP earnings lean on a related-party private mark is precisely the pattern that ends badly when any one bet slips.

Our take: Both are right about different horizons, which is what makes this a Hold rather than a directional call. The strategic logic is coherent and the balance sheet can carry it, but the equity must now traverse eight-plus quarters where reported earnings are structurally suppressed and every valuation support depends on disclosure management has so far withheld. Until the returns become measurable, the spend is a faith instrument.

Debate: Does the Margin Reset Change the Terminal Story?

Bull view: The 16.3% clean base is a trough artifact of subvention, commodities and dying credits; FSD attach above 55%, in-house pack ramps, and eventual Robotaxi mix carry gross margin structurally higher from here. Services' record 14.1% shows the software-and-fleet margin engine already working.

Bear view: The recovery narrative has now been re-based twice (17.9% to 17.5–18% clean to 16.3% clean), each time revealed by the CFO's own bridge rather than projected in advance. Vehicles are being sold increasingly on financing subsidies and an FSD promise, while the high-margin credit annuity disappears. Flat-at-16% is the new normal until proven otherwise.

Our take: The bear has the evidence and the bull has the mechanism. We side with modeling discipline: hold auto GM ex-credits at 16.0–16.8% through FY26 and let FSD attach prove the re-inflection before crediting it. The honest unknown is how much of the current attach surge converts to durable subscription revenue as the fleet ages.

Debate: Is Robotaxi Progress Fast Enough to Justify the Multiple?

Bull view: 380K unsupervised miles, zero notable incidents, seven metros, 10%+ weekly compounding and per-city launch cost trending to zero is exactly what an early exponential looks like; the camera-only vindication collapses the cost curve versus lidar competitors, and Cybercab gives Tesla the only purpose-built robotaxi vehicle at scale.

Bear view: Three quarters of commercial operation, a fleet measured in dozens of vehicles per city, three successive headline-metric migrations, and zero financial disclosure. Waymo-scale competitors publish safety and utilization data; Tesla publishes growth rates off undisclosed bases. Extraordinary claims with curated evidence deserve a discount, not a premium.

Our take: Operationally we score Robotaxi ahead of our April model (seven metros vs. three). Financially, the disclosure vacuum is now the single biggest analytical obstacle to underwriting the stock: it forces the market to price the business on faith exactly when the capex bill for it is arriving. First economics disclosure is the most important potential catalyst in the next two prints, in either direction.

Debate: The SpaceX Question

Bull view: The deepening integration (chips, connectivity, compute orchestration, energy offtake) is a unique structural advantage no competitor can replicate, and a formal combination would consolidate the Musk ecosystem's cash flows behind the capex cycle.

Bear view: Tesla shareholders are already carrying SpaceX exposure they did not buy: a $2B stake whose $1B quarterly mark is the difference between GAAP profit and roughly breakeven, related-party demand propping the energy growth story, and a Terafab structure that will split economics across entities on terms set by overlapping boards. A combination discussion conducted "with the appropriate process" is still a governance overhang measured in hundreds of billions.

Our take: The industrial synergies are real, but from a minority shareholder's seat the direction of value flow is unauditable, and this quarter demonstrated it can dominate the GAAP income statement. We treat related-party concentration as a materializing risk and would treat any combination proposal as a rating-relevant event on announcement.

Model Update Needed

Valuation impact: Trailing-four-quarter non-GAAP EPS is now $1.74, putting the pre-print $374.01 close at ~215x trailing and roughly 250x our revised FY26 midpoint of $1.50. The multiple only resolves on FY27–28 numbers, which depend on the three disclosures Tesla has not made: Robotaxi economics, Optimus schedule reliability, and Terafab structure. We cut our 12-month price target range to $350–$420 (from $385–$455), anchored to the $374.01 pre-print close; against the mid-$350s after-hours level the midpoint implies a roughly market-level return, consistent with a Hold. The cut reflects the EPS re-base and FCF deepening; we deliberately did not cut further for the capex-duration extension because the offsetting demand and autonomy execution was genuinely strong. Upside trigger: Robotaxi economics disclosure or clean auto GM back above 17%. Downside trigger: FY26 capex drifting toward $30B, drawn debt, or a China FSD failure.

Thesis Scorecard Post-Earnings

Overall: Thesis weakened on net. The operational half of the story is the strongest it has been in two years: demand, autonomy execution, FSD monetization and energy volume all confirmed or exceeded. But the financial half broke: the margin-recovery pillar that anchored the Outperform case is not recovering (it was never where the headlines placed it), the earnings base has to be rebuilt ~35% lower, the capex trough extended by two years and acquired a debt component, and GAAP earnings quality degraded via related-party marks. When the pillars that confirmed are the ones already priced and the pillar that broke is the one carrying the multiple, the risk/reward is no longer favorable.

Action: Downgrade to Hold from Outperform, honoring the trigger we set in April (auto GM ex-credits below 17% clean; it printed 16.3% with a flat clean bridge). We would return to Outperform on any of: first Robotaxi economics disclosure showing a credible path to contribution, clean auto GM ex-credits back above 17% for two quarters, or a valuation reset toward the low $300s that pre-pays the trough. We would consider Underperform if FY26 capex drifts toward $30B with debt drawn, if the China FSD silence extends into 2027, or if a SpaceX combination proposal emerges on terms that subordinate Tesla minority holders.