This is a sample of CapRelay's research. CapRelay provides similar research and more for every US-listed company.
Start free trialTesla is in the early innings of a genuine business model transition — from a hardware-centric automaker into a company monetizing real-world AI at scale. Two catalysts are converging: an accelerating Robotaxi business that is growing at compounding weekly rates across seven U.S. markets, and a vehicle demand recovery that has produced record Q2 2026 deliveries and the largest order backlog since 2023. Meanwhile, FSD attach rates in North America have crossed 55% of new deliveries, a level that suggests buyers are increasingly purchasing Tesla primarily for the software, with the hardware as a bonus.
The near-term financial pain from a $25B+ CapEx cycle is real, but the investments are in concrete productive assets — Cybercab, Semi, Optimus, LFP cells, Megapack 3 — all of which are either in production or beginning ramp in 2026. The bearish case requires believing that this capital cycle destroys value; the bullish case requires believing that even a partial success across these programs generates returns that dwarf the investment.
Key thesis points:
Tesla launched its Robotaxi service in Austin in June 2025 with paying customers and no safety driver. As of FY26Q2, the company has accumulated 380,000 unsupervised miles across six cities with zero notable incidents. The service has expanded to seven total U.S. markets, including Florida cities launched in July 2026, and Cybercab production commenced at Gigafactory Texas in FY26Q2.
The growth trajectory is notable. Weekly unsupervised miles are growing at double-digit rates week-over-week, a pace the VP of AI says has been sustained for several months and is expected to continue through year-end. The expansion to new cities has required less incremental effort per city than anticipated — management describes the time-to-launch-per-city trending toward zero as the software generalizes.
The strategic logic behind Tesla's approach is now being tested in the real world. Unlike Waymo's LiDAR-based system requiring pre-mapped geographies, Tesla's vision-only AI generalizes to any geography from the start. The implication is a faster and cheaper scaling path: once the technology works in a handful of cities, it works everywhere in the same jurisdiction. Management's stated goal is to cover half the U.S. population by year-end 2026.
FSD v15, which is partially running on the Robotaxi fleet already, is described by the AI team as incorporating seven major improvement tracks simultaneously. The early v15 builds in the fleet have 40% of those tracks merged. As v15 completes, Tesla expects to sustain the current compounding growth rate in fleet miles through end of year. The Cybercab, purpose-built to minimize cost per mile, has commenced production and will progressively replace the higher-cost Model Y as the primary Robotaxi vehicle, eventually driving cost per mile potentially below $0.30.
FSD has crossed a threshold from feature to primary purchase driver. In FY26Q2, 55% of North America deliveries included an FSD subscription at the time of delivery. Active paid FSD customers globally reached 1.48 million as of FY26Q2, up 51% YoY compared to FY26Q1. FSD churn is declining as the software improves, and existing subscribers are driving more miles — behavioral evidence of genuine product value.
Tesla's transition to subscription-only FSD pricing (eliminating the upfront purchase option) creates a near-term headwind to automotive margins but builds a recurring, high-margin revenue stream. Management explicitly states that the bulk of FSD monetization growth will come from subscriptions going forward. With 55% attach rates in North America and much lower rates in Europe and China where regulatory approvals are still pending, the approval pipeline represents a meaningful unrecognized growth vector.
Europe is the most important near-term catalyst: Netherlands approval was received in FY26Q1, and EU-wide review is underway. Management expects EU-wide approval in Q2 2026. In China, approval is being pursued for Q3 2026. Given that neither Europe nor China currently contributes materially to FSD revenue, approval in these markets would add incremental subscribers to a high-margin revenue base at minimal marginal cost.
Importantly, the FSD flywheel interacts with Robotaxi in a compounding way. The roughly 6 billion supervised FSD miles already accumulated from the customer fleet provide training data that makes the Robotaxi system safer. A safer Robotaxi expands faster and covers more geography. More geography and more supervised customer miles produce more training data. This loop is why Tesla's training compute has more than doubled in H1 FY26 — it is feeding an improving product.
FY26Q2 deliveries of 480,126 units were a record Q2 performance, up 25% YoY. Sequentially, volume grew 60% in the Americas, 27% in APAC, and 12% in EMEA. Tesla exited FY26Q2 with its largest order backlog since 2023 — a management observation that frames the constraint correctly: Tesla is not fighting for demand; it is fighting to supply it.
The production bottleneck is battery pack capacity, not cell supply or final assembly. Tesla is addressing this through multiple parallel efforts: ramping 4680 cells in Berlin packs, adding LFP module capacity in China, and retooling Reno for additional pack output. As these additions come online in the coming quarters, they convert a pipeline of orders directly into deliveries.
The recovery has occurred even without two of the largest potential demand catalysts: FSD is not yet approved in Europe or China, and the personal fleet Airbnb model (allowing owners to add their vehicles to the Robotaxi fleet) is not yet launched. When both arrive, they represent discrete step-ups in demand, particularly as the Airbnb model creates a financial incentive for ownership that reduces the effective cost of a Tesla.
Automotive gross margin, excluding regulatory credits, improved from 15.4% in FY25Q3 to 19.2% in FY26Q1 before normalizing to 16.3% in FY26Q2, with management noting the sequential decline was mostly explained by one-time FY26Q1 benefits from warranty true-downs and tariff reversals. Controlling for those items, underlying automotive margins were approximately flat — evidence that the demand recovery is happening without fresh price cuts.
Energy storage is a legitimate second growth engine, not a side business. Revenue grew 27% in FY25 to $12.8B, deployments grew 49% YoY to 46.7 GWh, and FY26Q2 deployments of 13.5 GWh were up 41% YoY.
The structural demand driver is compelling: U.S. grid average power consumption is approximately half the peak generation capacity. Batteries allow utilities to run power plants continuously at full output and store the excess, effectively doubling grid energy output per year with no new generation investment. AI data center demand is amplifying this — hyperscalers are using Megapacks not just for backup power but to smooth the extreme power fluctuations during AI training runs. This is a high-quality, defensible demand base.
Energy margins peaked at a one-time 39.5% in FY26Q1 due to tariff reversals and normalized to 20.4% in FY26Q2, including a roughly $240M warranty true-up on legacy cell issues. Management guided the business to a mid-to-low 20% normalized gross margin range, reflecting real competitive pressure and tariff exposure on Chinese LFP cells. The offset is U.S. LFP cell manufacturing coming online in Nevada, which will reduce tariff exposure as it scales. Megapack 3, beginning production at the Houston Megafactory in 2026, offers further cost reduction. Given the growth in deployments and the scale of the backlog, the energy business will continue generating growing absolute profit even with margin compression.
Tesla has built a supply chain structure that no other automaker can replicate quickly. U.S. vehicle content is approximately 85% USMCA-compliant on a weighted average basis, giving Tesla lower effective tariff exposure than any major competitor. While tariffs added over $400M per quarter across automotive and energy in FY25, Tesla's exposure is meaningfully lower than rivals sourcing more components from Asia.
The domestic infrastructure being commissioned in 2026 deepens this advantage:
The proprietary AI chip program is particularly important. Tesla's AI team argues the AI5 chip will be approximately 40x better than AI4 by key metrics, and the chip's significant simplicity — designed specifically for Tesla's software stack with no need to serve external customers — allows deletion of entire logic blocks that general-purpose chips require. This enables performance-per-watt and performance-per-dollar advantages that management believes are significant multiples of competing silicon.
This vertical integration is not incidental; it is a planned defense against the geopolitical supply chain risk that management views as the primary constraint on Tesla's growth in three to four years. Building a domestic semiconductor fab (TerraFab), partnering with Samsung and TSMC in U.S.-based facilities, and securing Micron memory allocations are all steps toward removing a potential ceiling on how fast Tesla can scale Optimus and Robotaxi.