Tesla

Refresh delta (since 2026-06-17)

What changed — the five load-bearing deltas

1 · Q2 volumes printed, and they are a step-change — not a continuation of the decline. 480,126 deliveries, +25.0% YoY vs 384,122 in Q2 2025, and +34.1% sequential vs 358,023 in Q1 2026 [primary: 8-KForm 8-KA filing for something that happened between reports and matters enough to tell shareholders now. Ex-99.1, 2026-07-02; Q1 comparatives primary: 8-K Ex-99.1, 2026-04-22]. Tesla delivered 28,368 more vehicles than it produced (480,126 − 451,758) — the quarter was partly an inventory drawdown, which matters because Q1 ended with days-of-supply at 27, the highest in the disclosed five-quarter series [primary: Q1 2026 update]. The prior dossier's spine — "a declining-volume automaker" — is now, at minimum, on hold pending tonight's price/margin data.

2 · The AI-compute buildout is disclosed, ramping, and ahead of its own guide. Tesla publishes an "Installed Annual Capacity — AI Training Compute" table in each quarterly update. The delta across two quarters [primary: 8-K Ex-99.1, 2026-01-28 and 2026-04-22]:

Q4 2025 guide: "In the first half of 2026, we plan to more than double the size of onsite compute in Texas (in terms of H100 equivalents)" [primary: 8-K Ex-99.1, 2026-01-28]. That doubling was met and exceeded in Q1 alone (>100k → >230k ≈ 2.3x). Q1 2026 narrative: "Cortex 2 is now online and has started running training workloads" [primary: 8-K Ex-99.1, 2026-04-22].

3 · A $2.0B AI-hardware acquisition, disclosed in one sentence and named nowhere. Note 14, Subsequent Events, Q1 2026 10-QForm 10-QThe quarterly version of the annual report. Lighter, and not audited.: "In April 2026, the Company entered into an agreement to acquire an AI hardware company for up to $2.00 billion in Tesla common stock and equity awards, of which approximately $1.8 billion is subject to certain service conditions and/or performance milestones dependent on the successful deployment of the company's technology." [primary: 10-QForm 10-QThe quarterly version of the annual report. Lighter, and not audited., filed 2026-04-23]. The target is not named in the filing, the consideration is stock (not cash), and 90% of it is contingent on deployment milestones — i.e. Tesla structured it as an acqui-hire / earn-out on unproven hardware. It appears in neither the shareholder deck nor the earnings call . This is the single most under-covered item in the delta window and it is squarely an AI-infrastructure event.

4 · Tesla became a related-party seller of AI-datacenter power hardware — and the counterparty went public-facing. Q1 2026 10-Q, Note 12: Tesla recognised $87M of revenue and $65M of cost of revenue from SpaceX for Megapack purchases in the quarter [primary: 10-Q, 2026-04-23]. Separately, in the delta window, SpaceX's amended S-1 (published 2026-06-03) disclosed xAI's Megapack purchases from Tesla: $191M (2024), $506M (2025), $303M (Jan 1–Apr 30 2026, of which $269M was April 2026 Megapacks) — ~$1B cumulative. Tesla also holds a $2.00B equity stake in SpaceX common (converted from a former xAI preferred position), acquired March 2026, <1% ownership, equity-method with the fair-value option, marked quarterly through Other expense, net [primary: 10-Q Notes 1 and 12]. Mechanical consequence: once SpaceX is publicly marked, Tesla's GAAP net income inherits SpaceX's price volatility. Q1 2026's "Other (expense) income, net" was already −$535M [primary: 8-K Ex-99.1, 2026-04-22].

5 · "Megapod" — a trademark filing, not a product. Tesla filed an intent-to-use USPTO application (serial 99893717) in June 2026 for "modular data center hardware systems for artificial intelligence computing" — servers, networking, power distribution and cooling as an integrated unit. Nothing beyond the trademark exists in the public record: no specs, no price, no customer, no mention in any Tesla filing. Treat it as a directional signal about intent, not a product. It is the only genuinely new AI-infrastructure product signal inside the 35-day delta window.

What held

• The vertical-integration moat is intact and now primary-sourced. In-house 4680 (40 GWh Texas, Production), LFP Nevada (7 GWh, Early Ramp), cathode materials (10 GWh, Early Ramp), lithium refining (30 GWh, Early Ramp) [primary: 8-K Ex-99.1, 2026-04-22]. Tesla's own statement that "battery pack capacity continues to be the limiting factor on ramping our vehicle production" [primary: same] is the named bottleneck — carried forward from Lens 2 of the prior dossier, now sourced rather than inferred.
• Energy storage is still the best-margin business in the company. FY2025 energy gross profit $3,802M on $12,771M revenue = 29.8% GM, vs automotive-segment $13,292M on $82,056M = 16.2% [primary: 10-KForm 10-KA company’s audited annual report to the US regulator. The most complete thing it publishes. FY2025]. Energy is 13.5% of revenue but 22.2% of gross profit.
• Regulatory credits are decaying, exactly as the prior dossier argued — but from a different base. FY2025 $1,993M, −28% YoY from $2,763M [primary: 10-K FY2025]; Q1 2026 $380M vs $595M PY [primary: 10-Q]. Tesla's own attribution: "In 2025, governmental and regulatory actions, such as OBBBA, have restricted certain regulatory credit programs tied to our products" [primary: 10-K].
• Optimus timelines keep slipping. Q4 2025 deck promised the Gen-3 unveil "In Q1 of this year" [primary: 8-K Ex-99.1, 2026-01-28]; it did not happen in Q1 and was pushed again. The credibility-tax argument in the prior dossier survives the refresh.

What is now falsified

See the dedicated section below — five figures, all of which the prior pass took from web summaries and none of which survive contact with the filings.

---

The live question — Cortex 2 and the "500 MW" claim

The KB claim under test: "Tesla Cortex 2 — 500MW GPU cluster at Giga Texas", with GPU counts and Megapack figures sourced only to web-search snippets.

Verified against primary sources

• Cortex 2 exists, is at Gigafactory Texas, and is online. FY2025 10-K: "We are currently building Cortex 2 at Gigafactory Texas to further increase our AI training compute capacity" [primary: 10-K FY2025, filed 2026-01-29]. Q1 2026 update: "Cortex 2 is now online and has started running training workloads" [primary: 8-K Ex-99.1, 2026-04-22].
• The GPU count is disclosed, and it is >130k H100e for Cortex 2 in "Early Ramp", alongside Cortex 1 at >100k H100e in "Production" — >230k H100e combined [primary: 8-K Ex-99.1, 2026-04-22, "Supporting Infrastructure — Installed Annual Capacity" table]. Both entries carry Tesla's caveat that "Early ramp capacity includes expected capacity" — i.e. the 130k is a ramp-target figure, not a census of racked silicon.
• Photographic corroboration in the same filing: the Q1 2026 deck contains two captioned exhibits, "CORTEX 2 – BUILDING CONSTRUCTION AT GIGAFACTORY TEXAS" and "CORTEX 2 – AI TRAINING CLUSTER" [primary: 8-K Ex-99.1, 2026-04-22].

NOT verified — and it cannot be, from Tesla

• The 500 MW figure has no primary source. Tesla discloses AI training capacity exclusively in H100-equivalent GPUs — in the 10-K, the 10-Q, and both quarterly update decks. The strings "megawatt", "MW", and any power-capacity figure for Cortex do not appear in the FY2025 10-K, the Q1 2026 10-Q, or either shareholder deck. The 500 MW number traces only to enthusiast and aggregator outlets (gearmusk.com, financespotnews.com, basenor.com, tesorb.com, teslahubs.com) — none of which is a primary source, several of which read as SEO/AI-generated, and none of which cite a Tesla document. Label it: unverifiable, web-only, not corroborated by the issuer.
• The "~150 Megapacks at Giga Texas powering Cortex 2" figure is likewise web-only. Tesla discloses its own Megapack manufacturing capacity, never its internal deployments.
• Arithmetic sanity check on 500 MW. At a customary all-in datacenter draw of ~1.2–1.4 kW per H100-class GPU (silicon plus networking, cooling and facility overhead), the disclosed Cortex 2 ramp target of >130k H100e implies roughly ~170 MW, and the combined >230k H100e implies ~300 MW. 500 MW is not reconcilable with the disclosed GPU counts as an IT load. It is only coherent as (a) a site power envelope / interconnect capacity for the whole north campus rather than a cluster IT load, or (b) a projection of a fully built-out Cortex 2 well beyond today's disclosed ramp. Either reading makes it a different quantity from the one the KB claim implies.

Also stale, and worth correcting in the KB

The widely-repeated "the Giga Texas compute cluster is ~100,000 Nvidia H100/H200 GPUs" is Cortex 1's number and is now a year out of date. The disclosed combined figure is >230k H100e [primary: 8-K Ex-99.1, 2026-04-22].

Recommended KB disposition

Replace the 500 MW/GPU-count claim with the primary-sourced table above, carry the 500 MW figure only as an explicitly-labelled unverified web assertion with its arithmetic inconsistency noted, and flag Tesla as a company that discloses AI compute in H100-equivalents and never in megawatts — a fact that is itself useful when normalising a datacenter KB across issuers.

---

Phase A — Understand the business

Company Overview

Tesla's own self-description opens the FY2025 10-K: "We are focused on bringing artificial intelligence ('AI') into the real world, through products and services like Full Self-Driving ('FSD') (Supervised) and Robotaxi, as well as working to develop and commercialize AI robots ('Bots') (including Optimus). We intend to leverage our current operations, in which we design, develop, manufacture, sell and lease high-performance fully electric vehicles and energy generation and storage systems…" [primary: 10-K FY2025]. The vehicles are framed by the issuer itself as the leverage, not the object.

Two reportable segments only — (i) automotive (which includes services & other and reg-credit sales) and (ii) energy generation & storage [primary: 10-K FY2025, Note on Segment Reporting]. The CODM is the CEO and evaluates segments on gross profit, not operating income; Tesla explicitly states the CODM "does not evaluate operating segments using asset or liability information" [primary: same]. That is why no segment operating-income line exists anywhere in this dossier — it is not disclosed, not withheld here.

Where Tesla actually touches AI infrastructure — four distinct positions, all primary-sourced:

1. First-party compute buyer/operator. Cortex 1 + Cortex 2 at Gigafactory Texas, >230k H100e disclosed [primary: 8-K Ex-99.1, 2026-04-22]. Tesla states an explicit purchasing discipline: "We aim to maximize capital efficiency by scaling training compute judiciously, including when the training backlog gets too long or in anticipation of greater demand from our engineers" [primary: 8-K Ex-99.1, 2026-01-28]. That is a buyer with a backlog-driven procurement rule — a materially different demand signature from a hyperscaler selling capacity.
2. Custom silicon designer, on two tracks. AI5 inference processor — "In April, we completed the final chip design of our next-generation AI5 inference processor" [primary: 8-K Ex-99.1, 2026-04-22]. And training silicon: "We are also continuing our custom silicon development with Dojo 3 in an effort to reduce the cost of training over time" [primary: same]. Dojo is not dead — it was consolidated. FY2025 Note 17: Tesla recognised $390M of expenses within the automotive segment "related to charges for supercomputer assets, contract terminations and employee terminations" from "convergence of AI chip design efforts" [primary: 10-K FY2025]. A supercomputer-asset write-down plus a surviving Dojo 3 line is the primary-record shape of that story.
3. Would-be fab owner. "Our partnership with SpaceX aims to build the largest chip fab ever: vertically integrating logic, memory and advanced packaging… This begins with the Tesla-owned Research Fab on our Gigafactory Texas campus" [primary: 8-K Ex-99.1, 2026-04-22]. The FY2025 10-K separately notes a Samsung collaboration "to manufacture advanced semiconductors for AI inference and training in the U.S." [primary: 10-K FY2025]. Press reporting attaches the name "Terafab", an Intel fabrication partnership (7 Apr 2026), and a $20–25B cost said to be outside the 2026 capexCapital expenditureMoney spent on long-lived things — buildings, machines, servers — rather than on running costs. plan. None of the Intel involvement, the "Terafab" name, or the $20–25B figure appears in any Tesla filing. Primary names SpaceX; web names Intel and xAI. Flagged as a conflict, not resolved.
4. Arms dealer to everyone else's datacenter. The 10-K makes the pitch in its own words: "As AI infrastructure drives rapid load growth, Megapack helps to, among other things, increase utilization of existing generation and transmission capacity" and "as AI infrastructure drives rapid load growth, we see opportunities for our energy storage products to stabilize the grid, shift energy when it is needed most and provide additional power capacity" [primary: 10-K FY2025]. This is Tesla explicitly repositioning the energy segment as datacenter infrastructure.

Supply Chain — CARRIED, with one addition

Unchanged in structure; see the previous dossier. Two primary-sourced amendments:

• The named bottleneck is now the issuer's own: "Battery pack capacity continues to be the limiting factor on ramping our vehicle production, and the team is actively working on initiatives to increase capacity" [primary: 8-K Ex-99.1, 2026-04-22].
• The AI-compute supply chain is a second, parallel chain the prior dossier did not map: Nvidia (H100/H200-class silicon for Cortex) → Samsung and/or the SpaceX/Research-Fab venture (AI5 + Dojo 3) → Tesla-owned clusters at Giga Texas → Tesla's own FSD/Optimus training workloads. It is entirely internal on the demand side — Tesla has no merchant compute business to sell into (a point the Megapod trademark may or may not be intended to change).

Competitive Advantages

The auto/battery/charging moats are carried from the prior dossier. What the primary record now lets you say about the compute moat, which the prior pass could not:

• Scale, sourced. >230k H100e of owned training capacity [primary: 8-K Ex-99.1, 2026-04-22] puts Tesla in the small set of non-hyperscaler entities operating clusters at that order — and it is captive capacity with no external revenue obligation.
• Data, sourced. "Our global fleet can collect the equivalent of over 500 years of continuous driving data per day" [primary: 8-K Ex-99.1, 2026-01-28]. FSD (Supervised) v14.3 shipped April 2026 with a rewritten AI compiler and a runtime change Tesla says cuts "inference latency by up to 20%" [primary: 8-K Ex-99.1, 2026-04-22].
• The vertical stack, sourced. Tesla's stated method: "at the system level where we identify the limiting factor and develop bespoke and scalable solutions (batteries, power electronics, inverters, software, AI silicon, etc.)" [primary: 8-K Ex-99.1, 2026-01-28].
• What is NOT established: any evidence that owned compute at this scale has converted into a capability advantage. FSD remains (Supervised) in every customer vehicle, in Tesla's own labelling, with the standing footnote "Active driver supervision required; does not make the vehicle autonomous" [primary: 8-K Ex-99.1, 2026-04-22]. A compute moat around an unconverted capability is an input, not an output.

Segments

Annual, by segment [primary: 10-K FY2025]:

Inside automotive, FY2025 [primary: 10-K FY2025]: auto sales $65,821M (−9%), regulatory credits $1,993M (−28%), leasing $1,712M (−6%), services & other $12,530M (+19%). Reg credits are 2.1% of total revenue but 15.0% of automotive-segment gross profit.

Quarterly, by segment [primary: 10-Q Q1 2026]:

Geography — this is where the prior dossier's China narrative breaks. Revenue by sales location [primary: 10-K FY2025 and 10-Q Q1 2026]:

Read. FY2025's entire revenue decline came from "other international" — US and China were each flat to within 0.2%. And Q1 2026's re-acceleration came from the same bucket (+60% YoY), which matches the issuer's narrative of "continued growth in demand for our vehicles in markets in APAC and South America, while also seeing a rebound of demand in both EMEA and North America" [primary: 8-K Ex-99.1, 2026-04-22]. China has been flat, not collapsing — the prior dossier's framing of China share pressure as "the single biggest geographic risk" is not supported by the geographic revenue table.

---

Phase B — Measure performance

Latest print — PRE-PRINT SETUP (Q2 2026 reports tonight)

This lens carries no Q2 results. None exist. What follows is the last reported quarter, the two Q2 facts Tesla has published, and the consensus bar.

Last reported quarter — Q1 2026 (quarter ended 2026-03-31) [all primary: 8-K Ex-99.1 filed 2026-04-22, cross-checked to 10-Q filed 2026-04-23]:

The prior dossier's "free cash flow turned negative in Q1'26" is flatly wrong — Q1 2026 FCF was +$1.444B, and TTM FCF was $7.000B, the highest TTM reading in the disclosed 15-quarter series [primary: 8-K Ex-99.1, 2026-04-22, GAAP-to-non-GAAP reconciliation].

The GAAP/non-GAAP gap is the real earnings-quality issue, and it is widening. Q1 2026 GAAP EPS $0.13 vs non-GAAP $0.41 — a 3.2x gap, of which $0.23/share is stock-based comp and $0.05/share is digital-asset losses [primary: same reconciliation]. SBC per share nearly doubled YoY ($0.12 → $0.23), driven per the issuer by "AI and other R&D projects, 2025 CEO award SBC and SG&A" [primary: 8-K Ex-99.1, 2026-04-22]. TTM SBC: $3,282M vs $2,048M a year earlier [primary: same].

The two Q2 2026 facts that exist [primary: 8-K Ex-99.1, 2026-07-02]:

• Production 451,758 · Deliveries 480,126 (Model 3/Y 467,762; other models 12,364)
• Energy storage deployed 13.5 GWh

The consensus bar for tonight [all ``, no primary]:

• Revenue ~$26.2–26.4B · Estimize crowd $25.9B
• Non-GAAP EPS $0.50–0.53; range cited $0.27–$0.74; Deutsche Bank at $0.36
• GAAP EPS ~$0.36
• Adjusted EBITDA ~$4.00B
• Capex ~$6.7B and free cash flow ~−$3.25B

Earnings calls / management framing

Tone across the last two calls has moved from promising to pre-emptively lowering:

• Q4 2025 (2026-01-28): "2025 marked a critical year for Tesla as we further expanded our mission and continued our transition from a hardware-centric business to a physical AI company" [primary: 8-K Ex-99.1, 2026-01-28]. Gen-3 Optimus unveil promised for Q1 2026; Optimus SOP "before the end of 2026"; 1M robots/yr eventual capacity.
• Q1 2026 (2026-04-22): the deck's own language is hedged and infrastructure-first — "We continued to make meaningful progress on the build out of the infrastructure and AI software that underpins our Robotaxi and future robotics businesses" [primary: 8-K Ex-99.1, 2026-04-22]. On the call, Musk reportedly framed ramps with "Whenever you have an all-new product with an all-new supply chain… its growth curve is always a stretched S-curve" and said "I don't think unsupervised FSD or Robotaxi revenue would be super material this year, but I do think it will be material probably in a significant way next year". CFO Vaibhav Taneja: "our current expectation for 2026 is over $25 billion of CapEx".

What they stopped saying: any volume-growth target for autos. The Outlook section's Volume paragraph is now purely capacity-utilisation language with no number [primary: both decks]. What they started saying: capex, compute, fabs, and "AI, software and fleet-based profits."

Transcripts remain absent from the shelf — transcripts/ is empty and the call quotes above are ``. Flagged as the top sourcing gap for the next refresh.

Comps — CARRIED with a price update, not re-derived

Peer multiples were n/a in the prior pass and remain so; nothing on this shelf sources them. What is updated:

The $125 to $600 price-target spread — a 4.8x range across the sell side — is the single most informative line in this lens. It is a dispersion, not a consensus.

Price catalysts

Nothing in the 2026-06-17 → 2026-07-22 window was a disclosed material event with a Tesla filing behind it except the 2026-07-02 Q2 production/deliveries release, which beat a widely-cited Bloomberg consensus of ~397,466 by ~21%. Stock context: $378.93 at the 2026-07-21 close, +2.53% on the day, down ~6% on the month, and up ~41.7% over twelve months. The 12-month gain against a −46% YoY collapse in GAAP net income [primary: 10-K FY2025] is the whole valuation argument in one comparison.

Other delta-window items, none of them filings: Austin unsupervised robotaxi service area extended metro-wide (~245 sq mi) on 2026-06-03 with a reported ~20-vehicle active fleet; the Megapod trademark (2026-06-21); SpaceX's amended S-1 disclosing the xAI Megapack purchases (2026-06-03).

---

Phase C — Judge people & books

Management

The prior dossier's central governance claim is a year out of date. The 2025 CEO Performance Award was approved by shareholders on 2026 — no: on November 6, 2025, at the 2025 Annual Meeting: 1,892,235,822 for / 564,940,908 against / 12,227,846 abstained / 302,456,274 broker non-votes [primary: 8-K Item 5.07, filed 2025-11-07]. It is granted, live, and being expensed — the Q1 2026 deck footnotes its operational metrics "In accordance with our 2025 CEO Performance Award" [primary: 8-K Ex-99.1, 2026-04-22], and the same deck attributes part of the opex increase to "2025 CEO award SBC" [primary: same]. There is no upcoming pay-package vote.

What actually happened since, and it is a real delta [all primary: 10-Q Q1 2026, Item 5 Other Information]:

• 2025-12-19 — the Delaware Supreme Court reinstated the 2018 CEO Performance Award, reversing the Chancery decision that had rescinded it.
• 2026-03-18 — Chancery entered the final order implementing that reversal.
• 2026-04-21 — the disinterested directors determined this constituted a "Tornetta Decision Event", causing Musk's forfeiture of the 2025 CEO Interim Award, per the "no double dip" principle.
• 2026-04-21 — the Board approved an Implementation Agreement governing Musk's exercise of the 2018 award, expressly designed "to mitigate any negative impact of significant share sales on the Company" — the Special Committee having considered "the negative impact on the Company resulting from the likely sale of a significant number of shares of the Company's stock to cover the exercise price and tax obligations."

That last item is a disclosed, board-acknowledged supply overhang, and it is a mechanic the prior dossier did not have. Tesla's own risk factors separately flag the pledged-share scenario: Musk has historically pledged Tesla stock to secure personal loans, and "If Mr. Musk were to choose to partially secure any loans with pledges of Tesla common stock… and the price of our common stock were to decline substantially, Mr. Musk may be forced by one or more of the banking institutions" to sell [primary: 10-K FY2025, Risk Factors].

Capital allocation, primary-sourced. No dividend, ever, and none anticipated [primary: 10-K FY2025]. Capex history: $6.51B (2021) → $7.16B (2022) → $8.90B (2023) → $11.34B (2024) → $8.53B (2025, −25%) [primary: 8-K Ex-99.1, 2026-01-28] → >$20B guided for 2026 [primary: 10-K FY2025]. In Q1 2026 Tesla also raised $4.331B of debt (vs $625M in Q1 2025) and deployed $2.002B into SpaceX equity [primary: 10-Q]. Net position remains ~$35.5B net cash .

A risk factor Tesla wrote about its own pay package, worth quoting because it is unusual for an issuer to say: "There is a risk that the technologies and initiatives associated with the 2025 CEO Performance Award, including the product goals, are misaligned with current or future consumer demand… the product goals may not be indicative of the types of products or services that would, in the long run, generate financial returns necessary to justify the significant market capitalization goals of the 2025 CEO Performance Award" [primary: 10-K FY2025, Risk Factors].

Forensic red flags

Every item below is now filing-sourced rather than press-sourced.

1. The GAAP↔non-GAAP wedge, widening. FY2025 GAAP diluted EPS $1.08 vs non-GAAP $1.66 [primary: 8-K Ex-99.1, 2026-01-28]. Q1 2026: $0.13 vs $0.41. TTM SBC $3,282M and rising fast on the CEO award [primary: 8-K Ex-99.1, 2026-04-22]. Non-GAAP EPS is 3.2x GAAP; the bridge is compensation.
2. Reg credits — decaying, but from a base the prior pass had wrong. $1,993M FY2025 (−28%), $380M in Q1 2026 [primary: 10-K, 10-Q]. Tesla does not project the line to zero anywhere in its filings; that projection is a third-party forecast. What Tesla says is that OBBBA "restricted certain regulatory credit programs tied to our products" [primary: 10-K FY2025]. At 15.0% of automotive-segment gross profit, the line still matters; the honest statement is a decaying prop, not a legislated cliff to $0.
3. Non-recurring items in Q1 2026 margins — confirmed by the issuer's own bridge. Tesla lists among the drivers of the YoY operating-income increase: "increase in automotive one-time benefits related to warranty and tariffs" and "increase in energy one-time benefits related to tariffs" [primary: 8-K Ex-99.1, 2026-04-22]. The $230M/$250M magnitudes remain `` — Tesla names the items but not the amounts. Ex a $250M energy benefit, Q1 energy GM would be ~29.2% rather than 39.5% — i.e. flat to the FY2025 run-rate, not a step-change.
4. A tariff-refund contingency that can cut both ways. "In February 2026, the U.S. Supreme Court issued a ruling invalidating certain tariffs previously imposed under [IEEPA]… we may be eligible for a refund… we have not recognized a receivable… We continue to monitor these developments and their potential impact on our results of operations, including reduction of revenue for any potential refunds to certain energy storage customers for which a contractual obligation exists" [primary: 10-Q Q1 2026, Note 10]. Tesla is telling you an unrecognised asset exists and a contra-revenue obligation exists against it.
5. Inventory building faster than revenue. Total inventory $12,392M (Q4'25) → $14,434M (Q1'26), +16.5% in one quarter against −10.1% sequential revenue. Energy inventory rose 31% ($2,714M → $3,554M) [primary: 10-Q]. Days of supply 15 → 27 [primary: Q1 2026 deck]. Q2's 28,368-unit delivery-over-production gap partially unwinds the auto half of this.
6. Fair-value marks now run through net income. The $2.00B SpaceX stake is carried at fair value with gains/losses in Other expense, net [primary: 10-Q, Notes 1 and 12]. Q1 2026 Other expense, net was −$535M; Q4 2025 was −$592M [primary: 8-K Ex-99.1, 2026-04-22]. Digital assets ($786M at Q1'26, down from $1,008M) add a second mark-to-market line. GAAP net income is now partly a function of asset prices Tesla does not operate.
7. A $2.0B acquisition disclosed only in a subsequent-events footnote, with an unnamed target and ~90% contingent stock consideration [primary: 10-Q Note 14]. Not a red flag in itself; the disclosure venue is the flag.
8. The Dojo write-down, named. $390M of charges "for supercomputer assets, contract terminations and employee terminations" on "convergence of AI chip design efforts" [primary: 10-K FY2025, Note 17]. A capitalised-compute program that got written off inside the same year the company guided >$20B of AI capex is the most useful base rate on this shelf for how the next compute program might go.
9. Related-party revenue is concentrated in the segment being repositioned as AI infrastructure. $87M of Q1 2026 Megapack revenue came from SpaceX [primary: 10-Q Note 12]; the xAI figures in SpaceX's S-1 imply materially more across 2024–26. When the growth narrative for the energy segment is "AI datacenters buy Megapacks," the fact that the highest-profile buyers are the CEO's other companies is a disclosure-quality issue.

Regulatory findings. SEC: 0 Litigation Releases and 0 AAERs naming Tesla in the 2021-06-17→2026-06-17 window (AAER coverage partial — the EFTS search returned HTTP 500) . Not re-run this pass (<7 days old at the prior dossier; unchanged filings). 10-K Item 3 points to Note 13 rather than enumerating; Tesla's own summary sentence is "To our knowledge, no government agency in any such ongoing investigation has concluded that any wrongdoing occurred" while conceding "we cannot predict the outcome or impact of any such ongoing matters, and there exists the possibility that we could be subject to liability, penalties and other restrictive sanctions" [primary: 10-K FY2025]. The NHTSA probes, the Florida Autopilot verdict and the robotaxi crash reports carried from the prior dossier remain `` and were not re-verified this pass — treat those figures as of 2026-06-17.

---

Phase D — Project & stress-test

Forward view — RE-CAST AS PRE-PRINT SCENARIOS, NOT A FORECAST

No EPS projection is offered and no Brier forecast is logged. Q2 prints in hours; a projection authored now would be superseded before it could be read, and the Socratic rule for this refresh forbids a directional call. What follows is the arithmetic that constrains any projection.

The capex step-up is the whole 2026 story, and it is Tesla's own framing. From the 10-K, twice: "We currently expect our capital expenditures to be in excess of $20 billion in 2026, driven by our AI initiatives, including investments in compute infrastructure and data centers, the expansion and ramp of our manufacturing and R&D production lines and facilities, and growth in our fleet of company-operated AI-enabled assets and our retail, service and charging footprint" [primary: 10-K FY2025, MD&A and Risk Factors].

So: Q1 ran at a ~$10B annualised pace against a ≥$20B guide. Either Q2 shows a 2.4–3.0x sequential capex jump, or the guide is going to be walked. That single number — Q2 capex — is the most informative line in tonight's release for an AI-infrastructure reader, more than EPS.

The cash-flow mechanic that follows. TTM FCF at Q1 2026 was +$7,000M [primary]. If Q2 lands on the consensus −$3,254M, TTM FCF becomes ~$3,600M, a ~49% compression in one quarter — not because operations deteriorated but because the compute and robot factories are being paid for. Against $44.7B of cash and ~$35.5B net cash, that is fundable for several years without equity; Tesla has already begun pre-funding with debt ($4.331B raised in Q1 2026 vs $625M PY) [primary: 10-Q].

Energy-storage trajectory — the other half of the AI-infrastructure exposure [primary except where noted]:

H1 2026: 22.3 GWh vs H1 2025's 20.0 GWh = +11.5%. The lumpy Q1 was recovered; the growth rate is single-to-low-double digit, not the +49% of FY2025.

Installed Megapack capacity was quietly revised DOWN between decks — a primary-source delta worth flagging:

[primary: 8-K Ex-99.1, 2026-01-28 and 2026-04-22]. Installed Megapack capacity as stated fell from 80 GWh to 60 GWh across one quarter with no accompanying explanation. FY2025 deployed 46.7 GWh against 60 GWh — ~78% utilisation — meaning the Houston Megafactory (Megapack 3 / Megablock, SOP "later this year" per the issuer [primary: 8-K Ex-99.1, 2026-04-22]) is now the gating asset for energy-segment growth beyond ~2026.

Lenses 12 + 13 · Bull / bear and the short-seller case — REFRAMED AS FALSIFIABLE CONDITIONS

Per this refresh's Socratic constraint, no case is endorsed. Both are stated as what each side must be right about, so that tonight's print and the next two quarters can settle them.

The AI-infrastructure-optimist reading requires all of:

1. Q2 capex prints ≥ ~$5.8B — i.e. the >$20B guide is real and back-half loaded, not aspirational.
2. The Q2 compute table shows Cortex 2 moving from "Early Ramp" toward "Production" and total H100e above 230k. A flat compute table would be the loudest possible signal.
3. Auto GM ex-credits holds ≥ ~19% on 480k deliveries — i.e. the volume surge was demand, not discounting. Tonight settles this directly.
4. Energy GM holds near the ~29–30% structural level on 13.5 GWh, without a one-time tariff item doing the work.
5. Optimus SOP actually starts in 2026 as the Q4 2025 deck committed [primary: 8-K Ex-99.1, 2026-01-28], having already slipped the Gen-3 unveil.

The skeptical reading requires all of:

1. The 480k delivery quarter was bought — i.e. ASP or auto GM ex-credits falls materially, and the 28,368-unit inventory drawdown means Q3 starts thin.
2. The capex is being spent into an unproven capability: FSD is still labelled (Supervised) in every customer vehicle by Tesla itself; robotaxi is a ~20-vehicle Austin fleet plus Dallas and Houston; and Musk's own guidance is that robotaxi revenue is "not super material this year".
3. The $390M Dojo/supercomputer write-off [primary: 10-K FY2025 Note 17] is the base rate, not the exception, for Tesla's compute programs.
4. GAAP earnings continue to be a fraction of the headline: $1.09 TTM GAAP EPS supporting a $1.42T market cap is a 368x trailing multiple, with the gap to non-GAAP being compensation expense that is real dilutionDilutionIssuing new shares, so each existing share owns a smaller slice of the same company..
5. The disclosed 2018-award exercise overhang and the board's own Implementation Agreement to manage share sales [primary: 10-Q Item 5] act on supply.

The pre-mortem question tonight can begin to answer: did the capex arrive? If Q2 capex prints near $2.5B again, the AI-infrastructure thesis for Tesla is a press release, not a cash flow. If it prints near $6.7B with FCF at −$3B, Tesla has begun spending like a compute buyer — and every subsequent quarter is about whether the compute converts.

Management questions — CARRIED with five replacements

Q1–Q10 of the prior dossier's list stand (see the previous dossier), except questions 11 and 15, which are voided by the primary record (the pay vote already happened; the ~190x figure has moved). Five replacements, ordered by information value for an AI-infrastructure reader:

1. Q2 capex was $X against a Q1 of $2.493B and a filed guide of ">$20B" that the CFO verbally raised to ">$25B". Which number is the company standing behind, and what is the quarterly phasing for the rest of 2026?
2. The compute table says Cortex 2 is ">130k H100e, Early Ramp." What is the racked-and-running figure today versus expected capacity, what is the site's power envelope in MW, and when does Cortex 2 move to "Production"?
3. Note 14 discloses an agreement to acquire an AI hardware company for up to $2.00B in stock, ~$1.8B milestone-contingent. Who is it, what technology, what are the deployment milestones, and where does it sit relative to AI5 and Dojo 3?
4. Note 17 records $390M of charges for "supercomputer assets" on "convergence of AI chip design efforts." What was written off, and what specifically prevents the same outcome for Dojo 3 and the Research Fab?
5. Installed Shanghai Megapack capacity was restated from 40 GWh to 20 GWh between the Q4'25 and Q1'26 decks. What changed? And with Houston not yet in production, what is the installed- capacity path that supports energy-segment growth in 2027?

---

What the prior dossier got wrong — falsified by primary sources

Two further corrections of degree rather than fact: "record inventory" — days of supply was 27 in Q1 2026, elevated but not a record in the disclosed series (Q1 2025 was 22) [primary: Q1 2026 deck]; and "FY2026 capex >$20B (vs ~$9B FY2025)" — directionally right and now primary- confirmed at $8.527B FY2025, but the guide has since been verbally raised to >$25B and had not, as of Q1, begun to show up in the actuals.

Standing correction on method: all five errors came from taking a web summary's number instead of the filing's. Four of them were available in a document Tesla published before the prior dossier was written.

---

The numbers to watch tonight (2026-07-22, after close)

Ordered by information value on this desk's AI-infrastructure lens, not by headline prominence.

---

Exercised vs reasoned

Exercised (fetched and read in full this pass):

• SEC EDGAR submissions API for CIK 0001318605 (curl with declared User-Agent — returned 200; plain WebFetch is blocked, the declared-UA curl route works)
• FY2025 10-K (filed 2026-01-29) — read in full: MD&A, capex guidance, segment note, geographic revenue, Note 17 restructuring, risk factors, Item 3
• Q1 2026 10-Q (filed 2026-04-23) — read: statements, Note 10 contingencies, Note 12 related party, Note 13 segments, Note 14 subsequent events, Item 5 Other Information
• Q1 2026 shareholder update, 8-K Ex-99.1 (2026-04-22) — read in full incl. the Supporting Infrastructure and Manufacturing capacity tables and the GAAP↔non-GAAP reconciliations
• Q4 2025 shareholder update, 8-K Ex-99.1 (2026-01-28) — read for the compute-table baseline, the H1-2026 compute-doubling guide, and the 5-year financial summary
• Q2 2026 production/deliveries, 8-K Ex-99.1 (2026-07-02) — read in full
• 8-K Item 5.07 (filed 2025-11-07) — annual-meeting vote tabulations, read directly
• 10-K/A Amendment No. 1 (2026-04-30) — retrieved; cover/Part-III amendment, no financial restatement
• Research layer repopulated: 5 filings deposited to filings/; our figures, our figures, our figures, our figures, our figures written from primary

Reasoned / web-sourced only (labelled inline, not verified against a filing):

• Q2 2026 consensus (revenue, EPS, capex, FCF) — Bloomberg/Estimize via Electrek and Yahoo Finance, 2026-07-21
• The CFO's verbal ">$25B" 2026 capex raise and Musk's Q1-call remarks — no transcript on the shelf
• Price, market cap, multiples, analyst PT range — stockanalysis.com, 2026-07-21 close
• "Terafab", the Intel fabrication partnership, and the $20–25B fab cost — conflicts with the primary record, which names SpaceX and Samsung and gives no cost
• The 500 MW Cortex 2 figure and the ~150-Megapack deployment — no primary source exists
• xAI Megapack purchase figures — SpaceX's amended S-1 as read by Electrek; the S-1 itself was not fetched
• Robotaxi fleet size, city status, NHTSA probe status, the Florida verdict — carried from the 2026-06-17 dossier, not re-verified this pass

Not done: no transcripts ingested (transcripts/ remains empty — the top sourcing gap); our model not re-run (output <7 days old at the prior dossier and no new SEC filings in the enforcement window); no our model entry logged; no market call, no publish, no git operations, no writes outside companies/tesla/.

---