OptimusM Abundance

Automotive · autonomy · energy · robotics · silicon

Tesla

Last updated 22 Aug 2026 · Europe/Zurich

Status

Q2 2026: rev $28.2B; deliveries 480,126; energy storage 13.5 GWh; FSD subs 1.48M; FCF −$1.1B; 2026 capex >$25B. H1 BEV: Tesla 838k vs BYD 867k. Cybercab listed Production at Giga Texas.

Next

Cybercab public launch, Austin — this month (day unconfirmed). Clark County commercial rides ~30 days after 20 Aug NTA. Semi Europe specs 15–20 Sep IAA Hannover. AI5 mid-2027.

Competition

BYD leads global BEV. Waymo ~500k rides/week vs Tesla’s small unsupervised fleet. Figure 03 on a BMW line; Optimus still internal. BESS: Tesla 4th at 7.1% H1 integrator share.

On this page

Tesla

Last updated: 2026-08-22 (Europe/Zurich)

Research note for a living Elon-company science/engineering and business dashboard. Scope is Tesla only. Political and social-anthropology commentary is excluded. Every non-obvious number, date, quote, or program claim below is sourced. Items that could not be confirmed are listed at the end and flagged in-line as unconfirmed.

Primary snapshot date: Tesla’s Q2 2026 shareholder update and Q&A webcast, 22 July 2026 (period ended 30 June 2026). Post-quarter developments through 22 August 2026 are noted separately.

How to read this

  • Company-stated = Tesla SEC filing, IR deck, earnings call, or Tesla-branded X account.
  • Independent = NHTSA, competitor IR, or a named reporter citing a named source.
  • Unconfirmed = widely repeated but not found in a primary document, or sources conflict.
  • Tesla does not publish robotaxi fleet size, weekly rides, Optimus unit output, or Dojo FLOPS in its quarterly update. Those figures, when given, come from third parties and are labeled as such.

Business snapshot (Q2 2026, company-stated)

Metric Q2 2026 Q1 2026 Q2 2025 YoY
Total revenue $28.236B $22.387B $22.496B +26%
Automotive revenue $20.516B $16.234B $16.661B +23%
Energy generation & storage revenue $3.139B $2.408B $2.789B +13%
Services & other revenue $4.581B $3.745B $3.046B +50%
Total GAAP gross margin 16.8% 21.1% 17.2% −41 bp
Automotive GM (GAAP) 16.9% 21.1% 17.2%
Automotive GM ex-credits (non-GAAP) 16.3% 19.2% 15.0%
Energy GM (CFO on call; also = (3.139−2.499)/3.139) 20.4% 39.5% (CFO)
Operating income (GAAP) $0.398B $0.941B $0.923B −57%
Operating margin 1.4% 4.2% 4.1% −269 bp
GAAP net income $1.114B $0.477B $1.172B −5%
Diluted EPS (GAAP / non-GAAP) $0.32 / $0.33 $0.13 / $0.41 $0.33 / $0.40
Adjusted EBITDA $3.273B $3.668B $3.401B −4%
Operating cash flow $4.697B $3.937B $2.540B +85%
Capex $5.789B $2.493B $2.394B +142%
Free cash flow −$1.092B +$1.444B +$0.146B
Cash + short-term investments $43.524B $44.743B $36.782B +18%
Vehicle production 451,758 408,386 410,244 +10%
Vehicle deliveries 480,126 358,023 384,122 +25%
Energy storage deployed 13.5 GWh 8.8 GWh 9.6 GWh +41%
Active FSD subscriptions 1.48M 1.28M 0.95M +56%
Regulatory credit revenue $146M $380M $439M −67%
R&D expense $2.371B $1.946B $1.589B +49%

H1 2026 (sum of Tesla’s two reported quarters, not a Tesla-published H1 line): production 860,144; deliveries 838,149; energy storage 22.3 GWh; revenue $50.623B. Cumulative vehicle deliveries reached 9.7 million. Trailing-twelve-month revenue crossed $100B for the first time (company statement). TTM free cash flow still positive at $5.762B despite the Q2 deficit.

Capital allocation (company-stated, 22 Jul 2026): 2026 capex still guided at more than $25B, with further sequential increase expected in H2. Capex expected to grow for the next two to three years (robotaxi fleet, Optimus capacity, semiconductor fab, solar manufacturing, AI compute, plus automotive). CFO Vaibhav Taneja said Tesla is securing debt facilities that will give it capacity to borrow up to $30B. Q2 GAAP net income includes a $1.005B unrealized mark-to-market gain on Tesla’s SpaceX equity (excluded from non-GAAP); Q1 cash flow statement shows a $2.002B SpaceX equity purchase.

Demand / mix (company-stated): Tesla exited Q2 with its “largest order backlog since 2023.” Sequential delivery growth: Americas +60%, APAC +27%, EMEA +12%. Model 3/Y were 467,762 of 480,126 deliveries. “Other models” (S/X wind-down, Cybertruck, Semi) were 12,364 delivered / 8,822 produced. Inventory fell to 15 days of supply from 27 in Q1. Average automotive revenue per delivery in Q2 was about $42,730 ($20.516B / 480,126); that is arithmetic from Tesla’s tables, not a Tesla-published ASP.

Services: record Services & Other gross profit $648M and 14.1% gross margin (CFO: up from 9.2% in Q1). Includes used cars, Supercharging, service, insurance, and robotaxi infrastructure spend.

Sources: Tesla Q2 2026 Update, Exhibit 99.1; Form 10-Q for quarter ended 30 Jun 2026; Q2 2026 production/deliveries 8-K summary; Q2 2026 earnings call transcript; Q1 2026 Update.

Vehicles / manufacturing

Status (late August 2026)

Tesla remains a high-volume EV manufacturer first, with the investment cycle now explicitly aimed at autonomy, robotics, batteries, solar, and chips. Q2 was a record second quarter for deliveries (480,126, +25% YoY) on 451,758 units produced. Deliveries beat production, drawing down inventory. Model 3/Y remain the entire volume story.

Installed annual capacity Tesla listed as “Production” in the Q2 deck:

Region Product Capacity Status
California (Fremont) Model 3 / Model Y >550,000 Production
Shanghai Model 3 / Model Y >950,000 Production
Berlin Model Y >375,000 Production
Texas Model Y >250,000 Production
Texas Cybertruck >125,000 Production
Texas Cybercab >125,000 Production
Nevada Tesla Semi TBD Commissioning
TBD Roadster — Design development

Tesla’s own footnote: installed capacity ≠ current production rate. Battery pack capacity is “the main limiting factor to near-term vehicle production volume increase.” 4680 cells in Texas listed at >40 GWh, Production; Berlin 4680 Construction; Nevada LFP 7 GWh, Early Ramp (energy products); Texas cathode 10 GWh, Early Ramp; Texas lithium refining 30 GWh, Early Ramp. Supercharger network: 8,704 stations / 82,357 connectors (+17% connectors YoY; >2,400 net new stalls in the quarter).

Product changes confirmed:

  • Model S and Model X lines at Fremont decommissioned in Q2 to free the floor for Optimus. Final S/X builds were in early May 2026 (press, consistent with Tesla’s “decommissioned” language). “Other models” production fell 34% YoY to 8,822.
  • Model YL launched in the US in July 2026 (company). First builds at Giga Texas.
  • Cybercab listed as Production at Giga Texas with >125,000 nameplate. First production units, employee campus rides, and public-road engineering tests all in Q2/July (company). Unboxed process claimed in later investor conversations (JPMorgan note, not in the Q2 deck).
  • Tesla Semi: Musk on the 22 Jul call: “We have started production with the Tesla semi-truck.” The Q2 capacity table still says Nevada Semi is Commissioning with capacity TBD. Trade press reported the first unit off a new high-volume line in Sparks/Reno on 29 April 2026, with a 50,000/year factory target, two trims (325-mile Standard Range ~$260k; 500-mile Long Range ~$290k), 82,000 lb GCW, 800 kW tri-motor, 1.2 MW Megacharger. Treat the 50,000/year, prices, and April 29 date as trade-press, not Tesla IR. Tesla’s Q2 outlook wording is only: “Tesla Semi and Megapack 3 remain on schedule for production starting in 2026.” TechCrunch (22 Jul) noted Tesla removed 2026 “volume production” language for Cybercab, Semi, and Megapack 3 that had been in the Q1/January materials.
  • Cybertruck still in the capacity table at >125,000. Tesla does not break out Cybertruck deliveries; they sit inside “Other models” (12,364 Q2 deliveries including residual S/X and any Semi).
  • Roadster: still “Design development.” No 2026 production date in the Q2 deck.

Business constraint: Taneja said production growth is limited by the supply chain — batteries and electronic components — and that Tesla is “securing strategic deals with suppliers.” Automotive GM ex-credits fell sequentially 19.2% → 16.3%; Taneja attributed most of the drop to a non-repeat of a $230M Q1 warranty true-down plus tariff relief. Controlling for that, he said GM ex-credits would have been “approximately flat.” Regulatory credits collapsed to $146M (from $439M a year earlier) and are structurally shrinking.

Next milestone

Event Timing Confidence
Q3 2026 production/deliveries print Early October 2026, by historical pattern Tesla has not posted a date as of 22 Aug 2026. Unconfirmed date.
Q3 2026 earnings + deck Late October 2026, by historical pattern (Q3’25 was 22 Oct 2025; Q2’26 was 22 Jul) Unconfirmed date. Tesla IR has not announced it.
Cybercab public passenger service in Austin Tesla Robotaxi X contest (rides 17–23 Aug; winners 25 Aug) plus The Information reporting a public launch “as soon as this month.” Tesla has not published an event date. Launch event is company-confirmed as coming; calendar day unconfirmed.
Semi volume ramp / first large fleet deliveries Einride says first of 500 Semis start September 2026, phased over 24 months. WattEV: 50 of 370 in 2026, rest by end-2027 (ACT Expo, May). Tesla: factory commissioning, volume not guided. Einride is a primary company press release. Tesla 2026 Semi unit guide does not exist.
Semi unsupervised driving Musk on 22 Jul call: “probably around the end of this year or early next year,” with 3/Y/Cybercab taking priority for the next ~6 months. Company-stated, not a committed date.
Europe Semi Trade press: 2027 entry. Not in the Q2 deck. Unconfirmed.
Battery/pack unblocking so 3/Y can run closer to nameplate Ongoing; Tesla calls pack capacity the limiter Continuous, not a single date.
Terafab location announcement / Austin development fab Musk: location “soon”; equipment orders already placed for an Austin development fab combining mask, logic, memory, packaging, test. Company-stated, no date.

Competition (vehicles)

Global BEV volume: BYD passenger BEV wholesale 557,090 in Q2 2026 vs Tesla deliveries 480,126. H1 2026: BYD BEV 867,479 vs Tesla 838,149. BYD’s broader NEV (BEV+PHEV) H1 total was 1,808,511, with overseas ~792k (~44% of mix) and June overseas a record. Tesla is BEV-only. Sources: BYD monthly/H1 disclosures as compiled by NextBigFuture and ElectricCarsReport; Tesla IR. Treat BYD monthly wholesale vs Tesla deliveries as not perfectly like-for-like.

June 2026 model ranking (CleanTechnica, plugin vehicles): Tesla Model Y still #1 globally; BYD Song #2; Tesla Model 3 #3. OEM plugin share that month: BYD 19.2%, Geely 9.9%, Tesla 8.9%, VW Group 7.3%. BEV-only: Tesla 12.6%, BYD 13.9%. These are tracker estimates, not Tesla/BYD IR.

US: Tesla still roughly half of US EV sales in H1 2026 in secondary market-share roundups (~50.5% of ~463k). Unconfirmed at the tenth-of-a-percent level without an Automotive News / Cox primary table in hand.

China / other EV makers: Xiaomi EV delivered 31,267 in July 2026 and 216,322 in the first seven months (CnEVPost / CPCA), 39% of a 550,000 full-year target, with a new Sky Nomad EREV series unveiled 30 Jul 2026 priced below the China Model Y L (339,000 yuan). Tesla does not break out China deliveries in the quarterly update.

Rivian (US premium / R2 ramp): Q2 2026 production 12,613, deliveries 12,194 (company newsroom, 2 Jul 2026). H1 deliveries 22,559. FY2026 delivery guide raised to 65,000–70,000. R2 customer deliveries began 9 Jun 2026. Q2 revenue $1.658B, gross profit $179M (11% GM), cash $5.31B. Autonomy+: hands-free highway now; point-to-point “later this year” per CNBC (15 Aug 2026); $49.99/month or $2,500 lifetime vs Tesla FSD at $99/month. Rivian is ~2.5% of Tesla’s Q2 delivery volume.

Class 8 electric: US+Canada Class 7/8 EV registrations were cited at 2,829 in the TT News Semi piece (context for Tesla’s 50,000/year ambition). Einride 500-unit Semi deployment (18 Aug 2026 company PR) is the largest public Tesla Semi commitment; WattEV 370. Freightliner eCascadia and Volvo VNR Electric remain the incumbent BE Class 8 names; Tesla has not published a Semi backlog.

Sources

FSD & Robotaxi

Status

FSD (Supervised) — consumer ADAS, not autonomy. Tesla is explicit in every footnote: “Active driver supervision required; does not make the vehicle autonomous.” Q2 metrics:

  • 1.48 million active FSD subscriptions (+56% YoY, +200k QoQ). Definition (CEO performance award): upfront payment or monthly subscription, excluding free trials. Mix on the call: 55% upfront / 45% subscription. Tesla has removed the purchase option in most markets; Taneja expects future growth to be mostly subscriptions.
  • >55% of new North America deliveries had FSD enabled at delivery (company).
  • FSD is a stated demand driver. Musk: in FSD-approved markets, some buyers are “buying Tesla full self-driving with a car attached.”
  • FSD v14 Lite began rolling to AI3 / Hardware 3 customers in the US, and South Korea in July. Distills AI4 v14 behavior onto AI3 camera/compute; adds destination parking options and speed profiles.
  • Summer Release (July): in-app Self-Driving Stats sharing, expanded Grok (calls, music, climate), Automatic Navigation beyond home/work.
  • Europe: following the Netherlands, Tesla received FSD approvals in Lithuania, Estonia, Denmark and Belgium. Customers in those countries had driven >50 million km (~31 million miles) on FSD as of July (company).
  • Robotaxi fleet is already running early FSD v15. Elluswamy: v15 has ~seven major improvement tracks in parallel; ~40% of those tracks are merged and running on the robotaxi fleet. Same v15 models are intended to run on Cybercab.

Robotaxi — Tesla-operated, vision-only, no lidar. Q2 deck coverage table:

State Metro Status (as of 22 Jul 2026)
California SF Bay Area Safety driver (CA TCP permit #TCP0046782-A)
Texas Austin, Dallas, Houston Ramping unsupervised
Florida Miami, Orlando, Tampa Ramping unsupervised (Florida launches in July)
Arizona Phoenix Preparations underway
Nevada Las Vegas Preparations underway

That’s seven live metros (Bay Area + three Texas + three Florida). Elluswamy (22 Jul): >380,000 unsupervised miles across six cities in two states, “zero notable incidents”; any reports were “other actors impacting us when we were stationary.” Unsupervised miles growing at “double-digit” week-over-week rates since the start of 2026. Tesla tracks miles, not cars, because vehicles run near-continuously.

Post-quarter (independent, 20 Aug 2026): The Verge, citing Ethan McKanna’s crowdsourced Robotaxi Tracker, reported that all 170 Austin rides over two weeks, across 54 cars, were unsupervised. Dallas/Houston also saw a jump (~30 unsupervised vehicles in the past week, per McKanna). Tesla does not publish fleet size and did not comment. Waymo comparison in the same article: >300 driverless vehicles in Austin alone, ~4,000 across 10 other cities.

Cybercab path: production started at Giga Texas; employee campus rides in July; public-road engineering tests of production units. Purpose-built, no steering wheel or pedals. Musk: Cybercab-specific miles must be accumulated (including units retrofitted with controls) before large numbers go on the road. Starlink to be integrated on Cybercab (and eventually other vehicles) for coverage holes cellular cannot fill. Grok (SpaceXAI, formerly xAI) is in Tesla cars; SpaceXAI is also building a “manager” model for Optimus (earnings call / CNBC).

Safety / regulation:

  • NHTSA EA26002, opened 18 Mar 2026, upgraded from PE24031. Subject: FSD collisions in reduced roadway visibility; estimated ~3.20 million US vehicles. Nine crashes in the failure-report summary (including one fatality), plus six more SGO reports under review. Focus is whether FSD’s degradation-detection system detects glare/fog/dust and warns the driver in time. ODI resume: https://static.nhtsa.gov/odi/inv/2026/INOA-EA26002-10023.pdf (this research pass got HTTP 503 fetching the PDF; facts above are from the public ODI resume text as quoted by Electrek and NHTSA-mirrors on 19 Mar 2026). Status of EA26002 as of 22 Aug 2026 (open vs closed vs recall) is not independently re-verified this week.
  • A separate NHTSA PE on FSD traffic-violation behavior (PE25012) and a crash-reporting inquiry were described as concurrent in March 2026 coverage. Current status unconfirmed this week.
  • Electrek reported 207 Autopilot/FSD crashes in May 2026 from Standing General Order data. That is a journalist tally, not an NHTSA conclusion of defect rate.
  • Lars Moravy on the call: federal FMVSS work is moving toward acceptance of purpose-built AVs; he credited NHTSA. State-level sensor-mandate talk (he flagged New Jersey as “disheartening”) is the legal risk to a camera-only stack. Tesla asked Nevada for a large robotaxi permit; secondary sources say the state approved a permit on 27 Jul 2026 but capped the fleet at 10. Tesla IR does not state the Nevada cap; treat as unconfirmed until the Nevada Transportation Authority order is read.

Business of FSD: at $99/month, 1.48M paid subs would be ~$1.76B annualized if all were monthly and none churned — a back-of-envelope used in Tesla coverage, not a Tesla disclosure. Actual recognized FSD revenue is inside automotive ancillary / services and is not broken out.

Next milestone

Event Timing Confidence
Cybercab launch event / first public Cybercab rides in Austin Late Aug / early Sep 2026 per contest + The Information Event is coming (Tesla Robotaxi X). Date unconfirmed. Contest window ends 23 Aug 2026.
Remaining v15 tracks merged; unsupervised FSD on customer HW4/AI4 cars Tesla told JPMorgan (Fremont visit, reported 20 Aug) that HW4/AI4 can run v15 unsupervised; AI4.5 (~+10% compute, ~2× memory) rolling out; AI5 delayed to mid-2027. JPMorgan note relaying Tesla. Not a Tesla 8-K. Unsupervised customer FSD has been promised on successive hardware generations since 2016.
Robotaxi in Phoenix and Las Vegas “Preparations underway” as of 22 Jul Company. No public-launch date.
California unsupervised (remove safety driver) Requires CPUC/DMV path beyond current TCP supervised permit Unconfirmed timing. Tesla has not given a CA unsupervised date.
FSD approvals in more EU countries Ongoing after NL/BE/DK/EE/LT Next countries not named in the Q2 deck.
Q3 robotaxi miles / city count 22 Jul call: continue double-digit weekly mile growth; “very, very rapidly” Directional; Tesla still will not publish fleet size.

Competition (autonomy)

Waymo (Alphabet) is the US scale leader by a wide margin. Independent roundups as of August 2026, triangulating Waymo’s own infrequent disclosures: roughly 3,500–4,000 vehicles, ~500,000 paid rides/week, ~11 public US metros, 20M+ lifetime paid trips, target 1M weekly trips and 20+ cities. Freeway rides paused late May 2026, began returning 29 Jul starting in Phoenix. New markets announced 8 Jul 2026 (Las Vegas, Denver, San Diego, Tampa) — employee-first, public by end-2026. London later in 2026. Alphabet Other Bets Q2 2026: $382M revenue, $1.8B operating loss (segment, not Waymo-only). Waymo raised $16B at a $126B valuation in Feb 2026 (widely reported). Sensor stack: lidar + radar + cameras + HD maps, currently Jaguar I-PACE; cheaper next-gen vehicle rolling out. Tesla’s 380k unsupervised miles vs Waymo’s cumulative autonomous miles in the hundreds of millions is the honest technical comparison, not weekly city count.

Baidu Apollo Go is the global volume leader. Baidu Q2 2026 results (18 Aug 2026, primary): footprint 28 cities; >350 million autonomous km, of which >240 million fully driverless km. 2026 expansion: London open-road testing with Uber and Lyft; Dubai fully driverless commercial (own app + Uber); Hong Kong first fully driverless test permits (Airport Island, RHD/LHT); Switzerland testing with PostBus. Earlier Baidu print: Q4 2025 3.4 million fully driverless trips that quarter, weekly peak >300k, cumulative rides >20 million by Feb 2026. No US public service.

GM Cruise: public robotaxi shut down December 2024. Not restarted as of August 2026. GM is rebuilding personal-vehicle autonomy (ex-Tesla Autopilot lead Sterling Anderson as CPO; Super Cruise 1 billion hands-free miles as of April 2026; eyes-off / Level 3 target on 2028 Cadillac Escalade IQ). Robotaxi optionality is “eventually,” not a 2026 product.

Others: Zoox public (free) rides in Las Vegas and select SF. Avride via Uber in Dallas (safety specialist onboard at launch). WeRide / Pony.ai expanding internationally via Uber (Middle East, some Europe); no US public robotaxi. Rivian Autonomy+ is ADAS, not robotaxi, with a 2027–28 Uber-related personal-AV story in secondary coverage.

Architectural split that actually matters: Waymo/Apollo Go = geofenced, lidar-heavy, HD-mapped, currently far more unsupervised miles. Tesla = camera-only, generalized stack, tiny unsupervised fleet, huge supervised FSD data engine (1.48M subs). The 2026–27 question is whether v15 + Cybercab closes that gap on safety evidence and cost per mile, not whether Tesla can add cities.

Sources

Optimus

Status

Tesla is converting Fremont’s former Model S/X space into the first-generation Optimus production line. Q2 deck (22 Jul): “We have decommissioned the manufacturing lines for Models S & X at our Fremont Factory and are installing the first-generation lines for Optimus, where we expect to start production soon. The initial Optimus builds will be used in our Optimus Academy for training data collection and further functionality development.” Giga Texas Optimus building: “construction now in full swing.” Capacity table: Optimus California Construction, Texas Construction. No unit output, no ASP, no external shipments.

Musk on the call (calibrating, not hyping the ramp):

  • “Hardest product to scale manufacturing that we’ve ever made at Tesla, because everything on the robot is new.”
  • “There is no existing supply chain.” Tesla is building it or in-sourcing.
  • Initial S-curve “will be quite flat and long.”
  • Claims no currently shipping humanoid does generalized tasks (vs preprogrammed/teleop demos). Optimus target: pixels-in/controls-out, same end-to-end stack as FSD; human-level (or better) hand dexterity.
  • Gen 3 at Fremont; Optimus 4 at Austin, “much more vertically integrated,” aspirational ~10 million/year vs ~1 million/year for Optimus 3 — Musk’s words, with the manufacturing-caveat attached in the same answer.
  • Samsung and TSMC “putting in tens of billions” for AI compute; Panasonic on cells; Micron on memory allocation. Karn Budhiraj (on the call; Tesla manufacturing/supply) added MIM parts, FPCs, and in-source when no partner exists.
  • SpaceXAI model as “manager” of (Digital) Optimus; Digital Optimus = computer-use agent (photons in, controls out) that physical Optimus also needs for touchscreens.

JPMorgan Fremont visit (reported 20 Aug, Tesla claims relayed by the bank, not an 8-K): Gen 3 design “finalized”; supply chain “essentially locked in”; external commercial sales “as early as” H2 2027; long-term capacity ambitions ~1 million/year Fremont, ~10 million/year Texas. Treat H2 2027 external sales as Tesla-to-investor guidance via a bank note, not a booked launch.

Tesla has not held an AI Day / Optimus Gen 3 reveal in 2026 as of 22 August. No Tesla-published count of robots doing useful factory work.

Next milestone

Event Timing Confidence
First Fremont line units into Optimus Academy (internal) Q2 deck: “start production soon” / “production in 2026.” Musk in April had pointed at late July/August. As of 22 Aug, Tesla has not issued a “first unit off the line” IR note. Production intent is company-stated. First-unit date unconfirmed.
Gen 3 public reveal Repeatedly slipped; still outstanding Unconfirmed
External sales “As early as H2 2027” (JPMorgan note) Tesla-via-bank; not a contract
Giga Texas Optimus volume Construction; secondary reports say summer 2027 target Texas building is company-stated. Volume date unconfirmed in the Q2 deck.
Useful internal tasks (beyond data collection) Academy is explicitly for training data, not customer deployment No Tesla date for “first useful factory job.”

Competition (humanoids)

Figure AI + BMW (primary BMW, 2026): After an 11-month Figure 02 pilot at Spartanburg that BMW says supported production of >30,000 X3s, BMW is deploying Figure 03 on logistics sequencing (unsorted parts → trolleys). Figure CEO Brett Adcock, quoted in BMW’s release: the 02 deployment “proved that humanoids are no longer lab experiments.” This is the most concrete OEM-documented humanoid production use case in 2026. Tesla has not published an equivalent internal-task metric.

Boston Dynamics Atlas (primary, CES 5 Jan 2026): fully electric product version unveiled; production started at Boston HQ; all 2026 deployments committed to Hyundai (RMAC) and Google DeepMind; additional customers “early 2027.” Hyundai Mobis supplies actuators. Google DeepMind foundation-model partnership announced the same day. Not for general sale in 2026.

Unitree: ships low-cost humanoids (G1 often quoted ~$14–15k) at far higher volume than Western peers. Secondary (Omdia via industry blogs): Unitree ~5,500 humanoids in 2025 vs Tesla/Figure/Agility ~150 each. Unitree STAR-market IPO process reported in 2026. Volume leader, not a US auto-OEM deployment leader.

Others: BMW is separately piloting Hexagon AEON at Leipzig. Agility Digit has had Amazon-related warehouse work in prior years; 2026 unit status not re-verified here. Apptronik, Fourier, AgiBot appear in China-centric shipment tables; treat shipment league tables as research-house estimates.

What a technically literate reader should actually compare: (1) documented hours on a real production task (Figure/BMW has this; Tesla does not publish it); (2) actuator/hand dexterity and reliability (Atlas and Optimus Gen 3 both claim this; neither has public MTBF); (3) who can actually make 10k+ units (Unitree on cheap hardware today; Tesla/Figure/BD on expensive hardware tomorrow); (4) end-to-end neural control vs teleop/demo. Musk’s claim that no humanoid currently does generalized un-programmed tasks is a definition fight — BMW is using Figure for a specified factory task, which is the commercially relevant 2026 bar.

Sources

Energy

Status

Energy is Tesla’s second product engine and the one most directly tied to AI-datacenter demand. It is lumpy, ASP-competitive, and currently margin-compressed.

Q2 2026 (company): 13.5 GWh deployed (+41% YoY, +53% QoQ), second-best quarter, record TTM deployments, record EMEA deployments, Shanghai Megafactory at record. Energy revenue $3.139B (+13% YoY). Energy cost of revenue $2.499B → ~20.4% GM, matching Taneja on the call. That GM dropped from 39.5% in Q1. Three stated reasons: (1) ~$240M warranty true-up on vendor cell issues in legacy deployments; (2) Q1 tariff benefit >$200M did not repeat; (3) industrial-storage ASPs “coming down amidst growing competition.” Long-term energy GM guide: mid- to low-20%. Backlog “robust”; demand from data centers + electrification. Musk: SpaceX is a large Megapack buyer for training-run power smoothing (70% power drop in ~100 ms). He also restated the “batteries can theoretically double US energy output” argument (nameplate ~1.2–1.3 TW vs ~0.5 TW average load).

Capacity table (Q2): Lathrop, CA Megapack 40 GWh Production; Shanghai Megapack 20 GWh Production; Texas Megapack Commissioning (as of 22 Jul); Nevada Powerwall >6 GWh Production.

Megapack 3 / Megablock / Brookshire — post-quarter, company-branded: On 6 Aug 2026, @Tesla_Megapack posted that Megapack 3 is “officially rolling off the line” at the new Megafactory in Brookshire, Texas: groundbreaking-to-operation 16 months, designed capacity 50 GWh/year dedicated to Megapack 3. Specs Tesla has used since the Sep 2025 reveal (repeated in Electrek’s 6 Aug piece, which embeds the X post): ~5 MWh/unit vs 3.9 MWh Megapack 2 XL (~+28%) via 2.8-liter cells; 78% fewer thermal-system connection points. Megablock = up to four MP3s (~20 MWh) with transformer/switchgear factory-integrated; Tesla claims 248 MWh/acre, 1 GWh in 20 business days, ~23% faster install. These density/install claims are Tesla marketing specs, not third-party measured.

Q2 outlook still said “on schedule for production starting in 2026”; the 6 Aug X post is the production-start confirmation. “Volume” in 2026 was the language TechCrunch says was dropped. Brookshire 50 GWh is designed capacity, not 2026 output.

Powerwall: Powerwall 3P (three-phase, native three-phase inverter) available in Germany; Tesla wants other three-phase markets “as quickly as possible.” Residential remains a smaller slice than Megapack; Tesla does not break out Powerwall vs Megapack GWh.

Solar: Q2 deck: site work progressed for solar manufacturing (cells and panels, silicon-to-module in Musk’s telling). Taneja: US solar manufacturing “we are just going to multiply it with an order of magnitude.” On 20–21 Aug 2026, Electrek (two installer-network sources) plus Tesla’s own site behavior: Solar Roof tiles discontinued as not economically viable; tesla.com/solarroof redirects to tesla.com/solarpanels; Energy nav is Solar Panels / Powerwall / Megapack only. Tesla has not issued an IR press release on the discontinuation. Conventional panels continue.

Deal (primary-adjacent): 24 Jun 2026, Energy-Storage.News: NatPower + Tesla multi-year supply and execution agreement covering >25 GWh Megapack across Italy and the UK (five initial projects; broader program targeting >100 GWh). Tesla supplies hardware, EPC, and Autobidder with long-term revenue warranties; NatPower owns/operates. Aggregate construction value on the full scope US$4–5B; NatPower cites >$15B projected revenues over 20 years (NatPower figure, not Tesla). Tesla speaker on the deal: Mike Snyder, VP of energy and charging. This is the largest publicly described European Megapack award of 2026.

Cells: Tesla still buys stationary LFP cells (CATL / others) while ramping its own Nevada LFP (7 GWh, early ramp) and has discussed LGES supply for later Megapack 3 (secondary: LGES LFP deal from ~2025, supply starting Aug 2027 — not re-read as a 2026 primary this week; treat timing as unconfirmed). Q2 energy GM was hit by tariffs on LFP cells. Own cell production “isn’t coming close to 50 GWh” (Electrek commentary, consistent with Tesla’s 7 GWh LFP line).

Next milestone

Event Timing Confidence
Megapack 3 / Megablock first customer deployments from Brookshire Production started 6 Aug 2026; first grid COD not published Start of production confirmed. First commercial COD unconfirmed.
Energy GM “normalize” to mid/low-20s after the $240M warranty one-off H2 2026 / 2027 Company long-term guide, not a quarter target
Powerwall 3P beyond Germany “As quickly as possible” No country list
US solar cell/panel factory Site prep in Q2; Musk: silicon-to-module No production date in the Q2 deck
NatPower first Italy/UK COD Multi-year; five initial projects No COD dates in the June coverage
H2 2026 storage deployments > 2025 full year Taneja (per Benchmark): 2026 deployments still expected to exceed 2025 Directional company comment via analyst

Competition (BESS)

This market moved against Tesla on share, not on absolute GWh growth.

  • Benchmark Mineral Intelligence, H1 2026 system-integrator shipments: BYD Energy Storage #1 at 9.1%; Sungrow 8.8%; CATL 8.0%; Tesla 4th at 7.1% (Tesla fell out of the global top 3 for the first time) despite +11.5% YoY Tesla shipments. Tesla’s 2025 full-year integrator position had still been top-tier; BYD shipped >60 GWh of systems in 2025 vs Tesla 46.7 GWh (Benchmark via Electrek, 13 May 2026).
  • Wood Mackenzie 2025 (report published 2026): Chinese integrators 76% of global BESS; 8 of top 10 HQ’d in China. Tesla + Sungrow still top two for 2025 in that ranking; BYD jumped to third. Tesla retained North America leadership (Megapack + Autobidder + US factories). NextEra entered NA top 3 as a self-integrator. US 48E ITC FEOC content rules start 2026 at 55% of project cost, rising to 75% by 2030 — a structural tailwind for Tesla’s US factories vs Chinese hardware.
  • InfoLink 1Q26: BYD, Sungrow, Huawei, Tesla, CRRC among the top five; Tesla 5th in that quarter. 2026 global ESS shipment forecast ~600 GWh.
  • Product spec war: BYD HaoHan (Sep 2025) 14.5 MWh standard config, ~3× a single Megapack; 20-ft variant 10 MWh. Tesla’s answer is Megapack 3 density + Megablock factory-integrated MV, not a 14 MWh single box. Fluence remains a US/EU integrator name in the 4–5 GWh “next 10” band (Benchmark). CATL is the cell king (22% H1 2026 cell share per Benchmark) and also integrates.

How to use this: Tesla’s energy bull case is US/EU content + Autobidder software + install speed (Megablock) + data-center power electronics, not winning a global GWh share war against BYD/Sungrow/CATL. Q2’s 20.4% GM (after one-offs, mid-20s) is the number to watch against Chinese ASP pressure. Brookshire 50 GWh designed + Lathrop 40 + Shanghai 20 = 110 GWh nameplate if all three run — vs a 600 GWh world.

Sources

Dojo / AI training

Status

Tesla’s training story in 2026 is Nvidia-based Cortex clusters plus a custom-silicon roadmap (AI4 → AI4.5 → AI5 → AI6 / Dojo 3). The original Dojo/D1 program was reported disbanded in August 2025 (Bloomberg) and restarted in January 2026 around AI5/AI6 rather than a distinct D2 training chip.

What Tesla printed in the Q2 2026 deck (the only IR-hard numbers):

Asset Location Capacity Status
Cortex 1 Texas >90 MW Production
Cortex 2 Texas >115 MW Production

“We have more than doubled our onsite compute in Texas (in terms of MW of compute) during the first half of 2026. Cortex 2 supports the development of both vehicle and humanoid robot autonomy software and will ramp further over the rest of the year.” A chart of “Tesla AI Training Capacity Ramp (MW of Compute)” is in the deck (axis to 400 MW) but Tesla does not print a single company-wide MW or H100-equivalent total in the tables. Dojo is not named in the Q2 deck.

Chips (company-stated on 22 Jul call + prior Musk X posts):

  • AI4 (HW4): current inference computer. Tesla told JPMorgan it can run v15 and unsupervised FSD (bank note, 20 Aug).
  • AI4.5: “upgraded AI4” / moderate improvement; JPMorgan note: ~+10% compute, ~2× memory. Musk: production “around the middle of next year” (i.e. mid-2027).
  • AI5: Musk: “hopefully in volume production around the middle of next year,” initially into Optimus. JPMorgan note says AI5 was delayed to mid-2027. Tape-out was discussed in 2026 coverage as having happened in April — not in the Q2 deck; unconfirmed here as a Tesla IR fact. Foundry partners named on the call: TSMC (Arizona) and Samsung (Texas).
  • AI6: Musk X (19 Mar 2026, per Teslarati): possible December 2026 tape-out “with some luck”; one AI6 ≈ dual-SoC AI5 in the same half-reticle / node. Samsung 2 nm supply has been reported at ~$16.5B (Reuters, earlier). Yield/timing of 2 nm unconfirmed.
  • Terafab: Tesla’s own advanced-packaging / logic / memory ambition. Musk: location announcement “soon”; equipment ordered for an Austin development fab with lithography mask + logic + memory + packaging + test under one roof. Framed as required to scale Optimus (otherwise “we simply won’t have enough AI chips”). Overlap with SpaceX; GC Brandon Ehrhart: 2026 investment + framework agreement with SpaceX.
  • Dojo 3: Musk X in January 2026 (hiring note to AI_Chips@Tesla.com) said Tesla would restart Dojo 3 now that AI5 design is “in good shape,” with AI7/Dojo 3 later discussed in the context of space-based compute. Not in the Q2 deck. No Tesla-published Dojo 3 MW, FLOPS, or in-service date.
  • Megapods: Musk described boxes of AI4 + x86 (Digital Optimus in a box), Megapack-like packaging, deployable at Superchargers (~7 GW of Supercharger power, his figure) as distributed inference. Concept-stage on an earnings call, not a shipped product line.

HW3 / AI3 customer path: v14 Lite is the software consolation; Musk said it will “make sense to upgrade all cars that have less than HW4” that have cameras, to a next-gen AI board, at some financially sensible point. No price, no start date, no eligibility list in the Q2 materials. This is a live customer-commitment and possible recall/upgrade-cost issue.

Business: R&D $2.37B in Q2 (+49% YoY), opex $4.35B (+47%), opex “will grow in 2026 and beyond.” Capex includes AI compute; Texas onsite MW more than doubled in H1. Tesla is a large Nvidia customer for Cortex; custom silicon is the cost-down and supply-insurance bet, not a 2026 training-revenue business. There is no Tesla AI-cloud product selling Dojo cycles to third parties.

Next milestone

Event Timing Confidence
Cortex 2 further MW ramp “Rest of the year” (Q2 deck) Company, no MW target printed
Terafab location / development-fab details “Soon” (22 Jul) No date. Still unannounced as of 22 Aug.
AI5 volume production Mid-2027 (Musk + JPMorgan note) Slipped once already; treat as target
AI6 tape-out Dec 2026 “with luck” (Musk, March) Aspirational
Dojo 3 hardware in a Tesla data center Restarted on paper Jan 2026 No in-service date
HW3 → next-gen board upgrades “At some point” Unconfirmed start

Competition (training compute)

Tesla is not competing with OpenAI/Google on foundation-model leadership; it is competing with Nvidia’s pricing and availability and with every AV/humanoid lab’s training budget. Relevant 2026 markers, not a FLOPS league table Tesla does not publish:

  • Nvidia remains the default training vendor. Tesla’s Cortex 1/2 are MW-denominated Nvidia (and mixed) clusters. Dojo 1 did not displace them.
  • SpaceXAI / Colossus (Memphis) is a sister-ecosystem cluster for Grok, not a Tesla asset. Do not add Colossus MW into Tesla Cortex.
  • Waymo / Alphabet train on Google TPUs + Nvidia; Waymo in Aug 2026 published more on its onboard “brain” (The Verge, 20 Aug) — different stack (lidar fusion), far more geographically specialized models.
  • Custom inference silicon at the edge is Tesla’s actual differentiator vs Waymo’s vehicle computer and vs Nvidia Drive. AI5/AI6 success is a unit-cost and Optimus-volume story. Failure means Tesla stays on Nvidia + AI4 and Optimus is chip-constrained (Musk’s own Terafab rationale).

Sources

Named executives (science / engineering / business statements that matter)

How to follow them: Tesla Q&A webcasts (ir.tesla.com, replay ~1 year), Tesla, Tesla AI, Tesla Robotaxi, Tesla Megapack, and Elon Musk accounts on X, and SEC Exhibit 99.1 each quarter. Tesla does not run a 2026 AI Day. The 2025 Annual Shareholder Meeting was 6 Nov 2025 in Austin; the 2026 ASM date is not posted on Tesla IR as of this writing (10-K said the 2026 proxy would be filed within 120 days of 31 Dec 2025).

Person Role Why they matter Where their 2026 remarks are
Elon Musk CEO, Technoking Sets chip, Optimus, robotaxi, Terafab, Semi autonomy, and capex pace. Calibrated Optimus difficulty on 22 Jul; Cybercab Starlink; “March of 9s.” Q2 2026 call; X (@elonmusk); occasional factory posts
Vaibhav Taneja CFO The numbers: GM bridges, $25B+ capex, $30B debt capacity, energy mid/low-20s GM, FSD mix 55/45, 55% NA attach, seven robotaxi markets, opex growth Q2 2026 call opening remarks; every earnings call
Ashok Elluswamy VP, AI Software (Autopilot / FSD / Optimus software) Only Tesla AI executive who regularly speaks on earnings. 380k unsupervised miles; v15 seven tracks / 40% merged; camera-only argument; Optimus Academy data flywheel; same stack on Cybercab Q2 2026 call (extended prepared remarks). LinkedIn: VP AI Software since Oct 2024. Rare conference talks.
Lars Moravy VP, Vehicle Engineering Hardware, FMVSS/NHTSA relationship, purpose-built AV rules, city-by-city TNC regulation, Cybercab chassis Q2 2026 call (regulation + Cybercab scaling)
Franz von Holzhausen Chief Designer Cybercab / Cybertruck / Semi form factor. Not an earnings-call regular in Q2 2026. Engineering-relevant only when discussing packaging, aero, or the no-wheel Cybercab cabin. Follow Tesla design posts, not quarterly numbers. Design reviews; Tesla product unveils. No Q2’26 call appearance found.
Mike (Michael) Snyder VP, Energy and Charging Named on the NatPower 25 GWh announcement (Jun 2026). Hardware + Autobidder + EPC packaging. Energy-Storage.News quote, 24 Jun 2026. Not on the Q2 call.
Karn Budhiraj Spoke on Q2 call on Optimus supply chain (MIM, FPCs, in-source vs Samsung/TSMC) Useful for Optimus manufacturability, not a public-facing IR name Q2 2026 call
Dan Priestley Semi program director Quoted by Transport Topics at the April 2026 high-volume-line start (“many thousands” of Semis in 2026) TT News, ~30 Apr 2026. Not on the Q2 call.
Travis Axelrod Head of IR Runs the call, say.com questions, and what Tesla will/won’t quantify (fleet size, Dojo FLOPS) Every webcast
Brandon Ehrhart General Counsel SpaceX investment / framework agreement; what can be said about any Tesla–SpaceX combination Q2 2026 call
Tom Zhu Senior manufacturing leader, China (Giga Shanghai) Shanghai 3/Y + Megapack 20 GWh. Rarely on US calls. Factory / China media. Confirm title against latest proxy; 2026 proxy not reviewed in this pass.

Departed, still relevant to Dojo archaeology: Pete Bannon (Dojo silicon) left around the 2025 Dojo wind-down; some Dojo engineers reported to have moved under Elluswamy (Yahoo org-chart piece). Not a 2026 spokesperson.

Not currently a Tesla technical spokesperson: Andrej Karpathy (left 2022, returned briefly, left again — 2026 employment status not re-verified; do not treat his posts as Tesla).

Unconfirmed / could not confirm

Facts this pass could not pin to a primary document, or where sources conflict. Do not put these in a dashboard as “known.”

  1. Q3 2026 earnings date. Tesla IR has not posted it. Secondary calendars (e.g. TrendMatrix) float 21 Oct 2026. Historical pattern is late October. Wait for ir.tesla.com.
  2. 2026 Annual Shareholder Meeting date. 2025 ASM was 6 Nov 2025. 2026 date not on tesla.com/ir as of 22 Aug 2026.
  3. Cybercab public-launch calendar day. Tesla confirmed an event via a Robotaxi-app contest (rides through 23 Aug, winners 25 Aug). The Information, via The Verge/Carscoops, says Austin public rides “as soon as this month.” No Tesla date.
  4. Robotaxi fleet size. Tesla will not publish it. Crowdsourced tracker numbers have moved around (dozens vs 54 unsupervised plates in Austin vs “186 Model Ys registered in Austin” in one Carscoops line). Do not treat any of those as Tesla-official. Elluswamy’s 380,000 unsupervised miles (22 Jul) is the only company figure.
  5. Nevada robotaxi permit cap of 10. Repeated in secondary trackers citing a 27 Jul 2026 Nevada decision. Did not read the Nevada Transportation Authority order in this pass.
  6. First Optimus unit off the Fremont line. Tesla said “soon” on 22 Jul and showed line photos. No “SOP” IR note by 22 Aug. Musk’s earlier late-July/August target may have been met internally; it is not externally confirmed.
  7. Optimus robots currently doing useful (non-demo, non-data-collection) factory work. Musk has both claimed and walked this back in prior cycles. Q2 language is Academy / training data only.
  8. Dojo 3 in-service date, MW, or FLOPS. Restarted on Musk’s X in Jan 2026. Absent from the Q2 deck. Cortex 1+2 MW (>90 + >115) are the only printed training numbers, and they are facility nameplate MW of compute, not a FLOPS rating and not a statement they are Nvidia-free.
  9. AI5 April 2026 tape-out. Repeated in chip blogs. Not in the Q2 deck.
  10. HW3 paid upgrade price, start date, and whether it is AI4.5 or AI5. Musk said upgrades “will make sense.” Nothing commercial.
  11. Semi 2026 unit volume. Priestley “many thousands”; factory 50,000/year nameplate; Tesla IR: commissioning / production this year / volume language dropped. No Tesla unit guide.
  12. Semi MSRP $260k / $290k. Trade press at April line-start. Not in the Q2 deck.
  13. EA26002 (NHTSA FSD visibility) current disposition. Opened 18 Mar 2026 as an Engineering Analysis on ~3.2M vehicles. This pass did not get a live NHTSA status page (PDF fetch 503). Recall vs still-open is unverified this week.
  14. PE25012 and the crash-reporting inquiry — still open? Not re-checked against NHTSA this week.
  15. US EV market-share to the tenth of a percent (the 50.5% / 463k H1 figure). Secondary compilation; no Cox/AN table retrieved.
  16. LG Energy Solution Megapack 3 cell supply starting August 2027. Appears in May 2026 BESS coverage. Contract not re-read.
  17. Solar Roof official IR confirmation. Website redirect + installer-network sources (Electrek, 20 Aug) are strong, but Tesla has not issued an 8-K.
  18. Franz von Holzhausen 2026 engineering comments. None found on the Q2 call or a 2026 AI/product day.
  19. Tom Zhu’s exact current title and 2026 proxy officer list. 10-K filed 28–29 Jan 2026; proxy not reviewed in this pass.
  20. Whether Tesla will hold an AI Day / Cybercab “day” distinct from the Austin launch event.
  21. Cumulative FSD miles as a single printed number. Q2 deck charts “Cumulative Miles Driven with FSD (Supervised)” on a billions-scale axis; this pass did not extract a precise total from the chart image. Do not invent “12 billion.”
  22. Cumulative paid robotaxi miles. Deck has a chart scaled to 2.5M; exact plotted value not extracted.
  23. Cortex 2 “500 MW by mid-2026” figures circulating on Tesla blogs from January Musk comments. The Q2 deck lists Cortex 2 at >115 MW Production — use that, not the 500 MW blog number, unless Tesla reprints it.
  24. Cruise 2026 headcount / remaining IP. Shutdown of the robotaxi service in Dec 2024 is well reported; residual GM AV program is the Escalade IQ 2028 story, not Cruise.
  25. Any Tesla–SpaceX merger. Musk and GC shut this down as not discussable on an earnings call. Collaboration (Grok, Starlink, Terafab, Digital Optimus, $2B equity) is real; combination is not a 2026 fact.

Source index (primary first)

  1. Tesla Q2 2026 Update (Exhibit 99.1, 22 Jul 2026): https://www.sec.gov/Archives/edgar/data/1318605/000162828026049213/exhibit991.htm
  2. Tesla Form 10-Q, quarter ended 30 Jun 2026: https://www.sec.gov/Archives/edgar/data/1318605/000162828026049270/tsla-20260630.htm
  3. Tesla Form 10-K, year ended 31 Dec 2025 (filed 28 Jan 2026): https://www.sec.gov/Archives/edgar/data/1318605/000162828026003952/tsla-20251231.htm
  4. Tesla Q2 2026 earnings call transcript, 22 Jul 2026: https://www.roic.ai/quote/TSLA:US/transcripts/2026-year/2-quarter
  5. Tesla Q1 2026 Update: https://www.sec.gov/Archives/edgar/data/1318605/000162828026026551/exhibit991.htm
  6. Tesla IR: https://ir.tesla.com
  7. NHTSA ODI EA26002 resume: https://static.nhtsa.gov/odi/inv/2026/INOA-EA26002-10023.pdf
  8. Baidu Q2 2026 results, 18 Aug 2026 (Apollo Go): https://ir.baidu.com/news-releases/news-release-details/baidu-announces-second-quarter-2026-results
  9. Rivian Q2 2026 production/deliveries, 2 Jul 2026: https://rivian.com/newsroom/article/rivian-releases-q2-2026-production-and-delivery-figures
  10. Einride–Tesla Semi, 18 Aug 2026: https://www.einride.tech/press/einride-partners-with-tesla-for-the-world%E2%80%99s-largest-tesla-semi-deployment
  11. Boston Dynamics Atlas CES, 5 Jan 2026: https://bostondynamics.com/blog/boston-dynamics-unveils-new-atlas-robot-to-revolutionize-industry/
  12. BMW Figure 03 Spartanburg: https://www.therobotreport.com/bmw-group-deploys-figure-03-humanoid-after-tests-previous-version/
  13. NatPower–Tesla 25 GWh, 24 Jun 2026: https://www.energy-storage.news/natpower-tesla-sign-25gwh-megapack-supply-and-execution-agreement-for-italy-and-uk-bess-projects/
  14. CNBC Tesla Q2 2026: https://www.cnbc.com/2026/07/22/tesla-tsla-q2-2026-earnings-report.html
  15. The Verge Q2 earnings: https://www.theverge.com/transportation/969311/tesla-q2-2026-earnings-revenue-profit-sales
  16. The Verge Austin unsupervised, 20 Aug 2026: https://www.theverge.com/transportation/982759/tesla-robotaxi-unsupervised-austin-robotaxi-tracker
  17. TechCrunch volume-production language, 22 Jul 2026: https://techcrunch.com/2026/07/22/tesla-spending-skyrockets-as-cybercab-semi-megapack-production-timeline-slips/
  18. Electrek Megapack 3 SOP, 6 Aug 2026: https://electrek.co/2026/08/06/tesla-megapack-3-production-starts-texas/
  19. Electrek JPMorgan Fremont note, 20 Aug 2026: https://electrek.co/2026/08/20/tesla-jpmorgan-fremont-fsd-v15-hw4-optimus-2027/
  20. Electrek Solar Roof, 20 Aug 2026: https://electrek.co/2026/08/20/tesla-discontinues-solar-roof-panels-only/
  21. Benchmark H1 2026 BESS: https://source.benchmarkminerals.com/article/who-shipped-the-most-bess-cells-and-systems-in-h1-2026-
  22. Wood Mackenzie 2025 BESS share: https://www.woodmac.com/press-releases/2026-bess-market-share-chinese-integrators-76-percent-global-dominance-2025/