Home
BTHE BARATELLI INSTITUTE · Mentoring at Scale
Case Studies  /  ELON DAILY
LIVING REFERENCE · PROVENANCE-TAGGED

Elon Daily

Every Musk-controlled company in one ledger — and a tag on every number telling you whether anyone had to sign it.

For twenty years, writing about Musk's companies meant writing about numbers nobody was legally responsible for. That changed in June 2026, when SpaceX went public and pulled xAI and X onto a consolidated income statement with it. Four of the six businesses below now sit inside a single SEC registrant; two remain private, and this page prints "not disclosed" rather than filling the gap with a press estimate. Every figure carries a tag showing what stands behind it.

6Companies tracked
4Inside one SEC filer
$36.0BCombined Q2 2026 filed revenue
−$24.7BCombined H1 2026 free cash flow
1.4 GWSpaceX nameplate AI compute
2Still fully private
Editorial close: 19 August 2026 · EDGAR scanned daily · Figures change only when a new primary document appears
How to read this page

The five tags

Most coverage of these companies mixes audited figures, company statements, leaked decks and secondary-market marks into one paragraph and lets the reader sort it out. An advisor planning around that mix cannot tell which numbers would survive a deposition. So every figure here is tagged. Section 8 sets out what each rung means in full and lists every document the page was built from.

FiledSubmitted to the SEC under signature and Section 18 liability — 10-Q, 10-K, S-1, a 424(b)(4) prospectus, a 13D/G, or the body of an 8-K reported under an item other than 2.02 or 7.01.
FurnishedIn an exhibit to a Form 8-K, which the cover page says is not to be deemed filed. Real and public, one rung down. Most of the operating detail lives here.
ReportedA named publication carried it, attributed but unsigned. Frequently correct, legally weightless.
EstimatedA private round, a secondary mark, an analyst model. Nobody is representing that this is what anything is worth.
Not disclosedNo primary source contains it. The page prints the words rather than importing a plausible number. An empty cell is information.

This table defines the vocabulary rather than using it, so it carries no rung of its own. Every table below carries one — a rung column, or a note like this one where the whole table shares a rung. Only Filed means signed under Section 18, and nothing is promoted to it to fill a table.

Master ledger

The six companies, and what is actually knowable about each

Company What it is Disclosure status Latest hard number Tag As of
SpaceX Launch, satellite broadband, and — since February 2026 — frontier AI. Nasdaq: SPCX. CIK 0001181412. SEC registrant. Files 10-Q, 10-K, 8-K. Q2 2026 revenue $7,814M; net loss $541M Filed 30 Jun 2026
Tesla Vehicles, energy generation and storage, services. Nasdaq: TSLA. CIK 0001318605. SEC registrant since 2010. Q2 2026 revenue $28,236M; net income to common $1,114M Filed 30 Jun 2026
xAI Grok and the AI compute platform. Wholly-owned SpaceX subsidiary since 2 Feb 2026. No separate filings. Consolidated into SpaceX's AI segment. AI segment Q2 2026 revenue $2,561M Filed 30 Jun 2026
X The real-time platform, formerly Twitter. Acquired by xAI 28 Mar 2025; now two levels inside SpaceX. No separate filings. Advertising line within SpaceX's AI segment. Advertising Q2 2026 revenue $367M (down from $426M) Filed 30 Jun 2026
Neuralink Implantable brain-computer interface. Private. No periodic reports. No filed financial statements exist Not disclosed
The Boring Company Tunnelling and underground transit. Private. No periodic reports. No filed financial statements exist Not disclosed

"Latest hard number" means the most recent figure the Institute has read in a primary source, not the most recent figure in circulation. Combined Q2 2026 filed revenue of $36.0B is the arithmetic sum of SpaceX's $7,814M and Tesla's $28,236M; the two companies are separate registrants and do not combine for any accounting purpose.

SpaceX · Nasdaq: SPCX

1. SpaceX — the rocket company is no longer mostly a rocket company

The single most useful thing in the first 10-Q is the segment table. SpaceX reports in three segments: Space (designing, building and flying reusable rockets), Connectivity (the Starlink broadband network), and AI (Grok, X, AI solutions, and computational infrastructure). In the quarter ended June 30, 2026, Space produced $962 million of the company's $7,814 million in revenue. That is 12.3%. Connectivity produced 54.9% and AI produced 32.8%.

Put a sharper point on it: for the six months ended June 30, 2026, Space revenue was $1,581 million against $1,611 million in the same period of 2025 — a decline — while total revenue grew 54%. The launch business did not shrink because anything went wrong with it; it shrank as a share of the whole because two other businesses grew around it. Anyone still valuing this company on launch cadence is valuing an eighth of it.

Consolidated results

$ millionsQ2 2026Q2 2025H1 2026H1 2025
Revenue7,8144,07112,5088,138
Cost of revenue3,4952,2825,8834,244
Gross profit (computed)4,3191,7896,6253,894
Gross margin (computed)55.3%43.9%53.0%47.8%
Research and development3,5481,9587,0623,515
Selling, general and administrative912
Loss on debt extinguishment(1,545)
Net loss(541)(1,008)(4,817)(1,536)
Net loss attributable to shareholders(5,488)
Loss per share, basic and diluted$(0.09)$(0.34)$(1.12)$(0.53)
Weighted average shares (millions)5,8642,9294,8792,902

All figures Filed — Form 10-Q for the quarterly period ended June 30, 2026. Em-dashes indicate a line the Institute did not read for that column, not a zero. Gross profit and gross margin are Institute arithmetic on filed inputs and are labelled as computed.

Read the H1 loss carefully. The six-month net loss of $4,817 million includes a $1,545 million loss on debt extinguishment — a refinancing charge, incurred once, in connection with restructuring the capital stack around the offering. Roughly a third of the half-year loss is that charge. The second-quarter loss on its own is $541 million, against $1,008 million a year earlier: the operating trend and the headline half-year number point in opposite directions, and only one of them is about the business.

Revenue by segment

$ millionsQ2 2026Q2 2025H1 2026H1 2025
Launch services6484909781,056
Launch & development314256603555
Space segment9627461,5811,611
Consumer2,4851,7214,6333,213
Enterprise & government1,8068672,9151,849
Connectivity segment (Starlink)4,2912,5887,5485,062
Advertising367426710870
AI solutions & infrastructure2,1943112,669595
AI segment2,5617373,3791,465
Total revenue7,8144,07112,5088,138

All figures Filed. Enterprise & government revenue includes Starlink Mobile service offerings, per the filing's own footnote. All products revenue is attributable to the Connectivity segment.

Three things in that table deserve to be said out loud. First, Starlink's enterprise and government line more than doubled year over year, from $867 million to $1,806 million in the quarter — growing twice as fast as the consumer line and now approaching half of Connectivity revenue. Second, AI solutions and infrastructure went from $311 million to $2,194 million in a single year, a sevenfold move that is the arrival of a compute-services business rather than the maturation of a model business. Third, and least discussed: X's advertising revenue is still falling. $367 million against $426 million in the quarter; $710 million against $870 million for the half. Inside a segment growing 130%, the advertising line is down 18%. Consolidation does not fix a business, it hides one.

Operating income by segment

Revenue by segment is the easy disclosure and most companies stop there. SpaceX goes further and publishes a full cost build for each segment — cost of revenue, research and development, and selling, general and administrative — down to income from operations. That is unusually generous, and it is the single most useful table this company produces, because it shows that the three businesses are in three completely different financial states.

$ millionsQ2 2026Q1 2026Q2 2025H1 2026H1 2025
Space — revenue9626197461,5811,611
  Cost of revenue329281330610627
  Research and development1,0769306932,0061,219
  Selling, general and administrative997087169175
  Impairment529
Space — loss from operations(542)(662)(369)(1,204)(439)
  Segment Adjusted EBITDA(205)(351)(93)(556)131
  Capital expenditure1,1741,0529462,2261,705
Connectivity — revenue4,2913,2572,5887,5485,062
  Cost of revenue2,0601,6511,4013,7112,615
  Research and development294205143499266
  Selling, general and administrative281213121494225
Connectivity — income from operations1,6561,1889232,8441,956
  Segment Adjusted EBITDA2,5972,0871,5834,6843,200
  Capital expenditure1,3671,3321,1302,6991,944
AI — revenue2,5618187373,3791,465
  Cost of revenue1,1064565511,5621,002
  Research and development2,1782,3791,1224,5572,030
  Selling, general and administrative532463398995699
  Restructuring charges (credits)2(11)190(9)194
AI — loss from operations(1,257)(2,469)(1,524)(3,726)(2,460)
  Segment Adjusted EBITDA1,146(609)(276)537(387)
  Capital expenditure15,8287,72374923,5513,316
Consolidated loss from operations(143)(1,943)(970)(2,086)(943)
Consolidated capital expenditure18,36910,1072,82528,4766,965

All figures Furnished — Q2 2026 earnings release, Exhibit 99.1 to Form 8-K. Segment Adjusted EBITDA is a non-GAAP measure the company defines as segment income or loss from operations before depreciation and amortisation, share-based compensation, restructuring and impairment; it is presented here because the company presents it, not because the Institute endorses it. Three checks were run before publishing: the three segment revenue lines sum to consolidated revenue in all five columns; the three segment operating results sum to the consolidated result in all five columns; and consolidated capital expenditure of $28,476 million for the half ties exactly to purchases of property, plant and equipment in the Filed statement of cash flows. A furnished table that reconciles to a filed one is worth more than a furnished table that does not.

Read the capex row before anything else. The AI segment spent $15,828 million of capital in a single quarter — more than eleven times the Space and Connectivity segments combined, and twenty-one times what the AI segment itself spent in the same quarter a year ago. For the half, AI capex of $23,551 million is 82.7% of every dollar of capital this company deployed. Whatever SpaceX is called, its capital budget belongs to a data-centre company that also happens to own the world's leading launch business and the world's largest satellite constellation.

The three segments are running three different companies. Connectivity is the only one making money: a 38.6% operating margin in the quarter, up from 35.7% a year earlier, throwing off $2.6 billion of Adjusted EBITDA against $1.4 billion of capex — self-funding with room to spare. Space is losing more money than it did a year ago and losing it deliberately: revenue grew 29% while research and development grew 55%, to $1,076 million, which is more than the segment's entire revenue. That is a company spending a dollar-and-a-tenth on Starship development for every dollar of rocket revenue it earns. AI made the quarter's most dramatic move — the operating loss halved sequentially, from $2,469 million to $1,257 million, and Segment Adjusted EBITDA crossed into positive territory at $1,146 million from negative $609 million three months earlier. Almost all of that came from one place: $1.6 billion of incremental infrastructure revenue recognised against $14.1 billion of newly signed Cloud Services Agreements. Whether that is an inflection or a pull-forward depends entirely on the shape of those contracts — and the shape is disclosed, in the registration statement rather than the earnings release. It is $1.25 billion per month through May 2029, terminable by either party on ninety days' notice. The terms and what they imply are set out in the data centre schedule in section 5.

What we could not check, and one thing the release got wrong. Segment operating income is furnished, not filed. The Institute attempted to read the equivalent segment note inside the Form 10-Q and could not: the retrieval available to us truncates that document partway through the notes, and rather than assert that the 10-Q "does not contain" a segment note we could not reach, we are telling you what we read and where. Separately, the AI segment highlights in the release state that compute capacity is "up from 1.0 GW in Q1 2026 and 0.4 GW in Q2 2026." The second reference is a typographical error for Q2 2025 — the table on the same page carries 0.4 GW in the Q2 2025 column. We are noting it rather than silently correcting it, because a reader comparing this page against the source should not have to wonder who made the mistake.

Balance sheet

$ millions30 Jun 202631 Dec 2025
Cash and cash equivalents93,52224,747
Marketable securities6,487
Total current assets108,04730,952
Property, plant and equipment, net65,73642,602
Goodwill11,64511,809
Total assets192,77092,079
Deferred revenue (current and non-current)14,28612,116
Debt and finance leases, current2,525928
Debt and finance leases, non-current36,83921,968
   of which related party13,3294,507
Total liabilities65,54650,754
Redeemable convertible preferred stock (temporary equity)38,752
Total shareholders' equity127,2242,573
Total liabilities, temporary equity and shareholders' equity192,77092,079

All figures Filed. The related-party total is the Institute's addition of the filed current ($2,039M) and non-current ($11,290M) related-party components at 30 June 2026, and of $455M and $4,052M at 31 December 2025. The footing line is shown because a balance sheet that does not visibly balance is a balance sheet the reader has to take on trust: total liabilities, temporary equity and shareholders' equity ties to total assets in both columns — $65,546M + nil + $127,224M = $192,770M at 30 June 2026, and $50,754M + $38,752M + $2,573M = $92,079M at 31 December 2025. The temporary-equity line is the reason the prior-year column does not foot without it: $38,752 million of redeemable convertible preferred stock sat outside permanent equity under ASC 480-10-S99 because redemption was outside the company's control. It converted at the IPO, which is most of why book equity moves from $2.6B to $127.2B in six months.

The number a practitioner should actually carry. Cash and marketable securities of $100.0 billion against debt and finance leases of $39.4 billion — roughly $60 billion of net cash, on a company that has never reported an annual profit and burned through a $4.8 billion half-year loss. This is the structural fact that governs everything else: for the next several years, SpaceX's spending is constrained by management's judgment, not by its lenders. Note also that a third of the debt is related-party, which means a third of the leverage sits on terms that were not set by an arm's-length lender.

Consolidated statement of cash flows

The income statement says SpaceX lost $4.8 billion in six months. The cash flow statement says something more useful, and something more alarming, and the two facts sit four lines apart. This is the full statement as filed, six months to 30 June, both years.

$ millions, six months ended 30 June20262025
Net loss(4,817)(1,536)
Depreciation and amortisation5,2902,970
Share-based compensation1,470694
Deferred income taxes(9)120
Unrealised loss (gain) on digital assets539(252)
Impairment and loss on disposal4054
Loss on extinguishment of debt1,545
Other non-cash(72)126
Accounts receivable(2,003)(470)
Inventories(827)(360)
Prepaid expenses and other assets102(2,125)
Accounts payable(88)309
Deferred revenue2,169680
Other operating liabilities127141
Net cash provided by operating activities3,466351
Purchases of property, plant and equipment(28,476)(6,965)
   of which paid to related parties329101
Capitalised interest(20)(22)
Proceeds from product rebates1,195
Purchases of marketable securities(13,630)(601)
Maturities of marketable securities7,248543
Sales of marketable securities1,173
Equity-method investment(86)
Purchases of intangible assets(856)
Other investing52(74)
Net cash used in investing activities(34,487)(6,032)
Net proceeds from initial public offering85,675
Proceeds from issuance of debt51,81210,943
Repayments of debt(39,396)(5,990)
Debt issuance costs(124)(61)
Premium paid on debt extinguishment(1,153)
Issuance of capital stock, net8,3195,047
Employee equity plans316155
Repurchases of capital stock(4,426)(520)
Taxes paid on net share settlement(559)(238)
Principal payments on finance leases(173)(137)
Net cash provided by financing activities100,2919,199
Effect of exchange rate changes(42)75
Net increase in cash and restricted cash69,2283,593
Cash and restricted cash, beginning of period25,12411,501
Cash and restricted cash, end of period94,35215,094

All figures Filed — Form 10-Q. Every one of the four sections was recomputed line by line and ties to the filed subtotal exactly, in both years. Two notes on presentation. First, the related-party capex line is a disclosure of how much of the $28,476 million went to related parties; it is shown positive because it is a component of the line above, not an additional outflow. Second, the beginning and ending balances are cash, cash equivalents and restricted cash, which is why they exceed the balance-sheet cash figures by $377 million and $830 million respectively — that difference is the restricted portion, and it is a reconciling item, not a discrepancy.

The one number to take from this statement. SpaceX generated $3,466 million of operating cash and spent $28,476 million on property, plant and equipment. Free cash flow for the half was negative $25,010 million — a quarter of a hundred billion dollars of cash consumed in six months, against negative $6,614 million a year earlier. It was funded almost entirely by the capital markets: $85.7 billion of IPO proceeds and $51.8 billion of gross debt issuance. This is the sentence the income statement will not say. The $4.8 billion net loss is a rounding error next to the capital intensity; the company's actual cash position is a function of the window staying open, and the window was very open in the first half of 2026.

Three further things worth reading twice. First, depreciation and amortisation of $5,290 million now exceeds the entire net loss, and grew 78% year over year — the depreciation charge from the 2025-26 capex wave has barely begun to land, and it will be the dominant force on the income statement for years. Second, the debt was largely refinanced, not raised: $51,812 million issued against $39,396 million repaid, at a cost of $1,545 million of extinguishment loss and $1,153 million of premium actually paid in cash. SpaceX paid roughly $2.7 billion to term out its balance sheet around the IPO. Third, deferred revenue provided $2,169 million of operating cash, more than half the total. That is customers paying in advance — a genuine strength, and also a reminder that this line reverses when growth slows.

The offering

ItemFiled detailTag
DateJune 2026; final prospectus filed under Rule 424(b)(4) on 12 June 2026Filed
Shares sold638.9 million shares of Class A common stock, including full exercise of the underwriters' over-allotment optionFiled
Price$135.00 per shareFiled
Net proceeds$85,675 million, after $575 million of underwriting commissions and offering costsFiled
Pre-offering splitFive-for-one forward stock split effected May 2026; all prior share and per-share data retroactively adjustedFiled
Preferred conversionAll redeemable convertible preferred stock converted automatically into Class A and Class B common at the IPOFiled
Shares outstanding afterClass A 7,607 million; Class B 5,569 million; Class C nil, at 30 June 2026Filed
Musk's economic and voting stakeGoverned by the prospectus and the dual-class structure. The Institute has not read the specific percentages in a primary source and does not print them.Not verified
Subsequent issuance — Cursor merger, 14 August 2026The merger of X67 Inc. into Anysphere, Inc. (“Cursor”) became effective 14 August 2026. Cursor's common and preferred stock converted into 389,289,254 shares of Class A common stock, plus 1,752,426 shares for vested Cursor restricted stock units — 391,041,680 Class A shares issued at closing, at an implied equity value for Cursor of $60.0 billion. Unvested Cursor awards were assumed and converted into approximately 29,128,326 restricted stock units and 44,365,047 options over Class A stock. Issued under the Section 4(a)(2) private-placement exemption, not registered.Filed

$575 million of costs on $86,250 million of gross proceeds is roughly 0.67% — an extraordinarily low all-in cost of issuance, and itself a data point about the negotiating position of an issuer of this size.

On the Cursor row, and why it is tagged Filed rather than Furnished. Every other 8-K figure on this page sits at the Furnished rung because it was read in an exhibit to a Form 8-K, and Tesla's own cover says an Item 2.02 exhibit is not deemed filed for Section 18 purposes. This one is different: the share counts and the $60.0 billion valuation appear in the body of the Form 8-K, reported under Item 2.01 and Item 3.02, signed by the Chief Financial Officer, with no furnishing legend anywhere on the cover. An 8-K body reported under an item other than 2.02 or 7.01 carries Section 18 liability. Tagging it Furnished because it arrived on a Form 8-K would be demoting a filed disclosure on the strength of the form number rather than the item number, and this page grades by liability, not by envelope.

The Institute's arithmetic on top of the filed figures, and labelled as ours: 391.0 million shares against the 7,607 million Class A shares outstanding at 30 June 2026 is an increase of roughly 5.1% in the Class A count; with the assumed unvested awards included the fully-diluted figure is roughly 6.1%. The $60.0 billion divided by the 389,289,254 conversion shares implies about $154 per share, which is the seven-day volume-weighted average close the merger agreement specifies and not a market quote on any single day. Estimated — and note that adding an August issuance to a 30 June balance-sheet count produces a pro-forma, not a current share count. SpaceX has not filed a share count as of a date after 30 June 2026, so this page does not print one.

Customer concentration

The filing discloses two customers above 10% of consolidated revenue and names neither of them. Customer A represented 18.3% of revenue in the quarter and 17.9% for the half (19.9% in the first half of 2025), with revenue in all three segments. Customer B represented 19.5% of the quarter and 12.2% of the half, entirely within the AI segment.

Disclosed customer concentration, Q2 2026Customer ACustomer BBoth
Share of consolidated revenue, per the filing18.3%19.5%37.8%
Implied revenue, $ millions1,4301,5242,954
Segments the filing places it inAll threeAI only
Share of consolidated revenue, first half 202617.9%12.2%30.1%
Share of consolidated revenue, first half 202519.9%below 10%

Percentages Filed — Form 10-Q concentration note. Implied dollars are the Institute's multiplication against filed consolidated revenue of $7,814 million and are therefore accurate only to the rounding of the filed percentage, roughly plus or minus $4 million. The movement in Customer B is the story. It was below the 10% disclosure threshold in every prior period shown and is now the company's largest single customer, which is what a cloud contract signed in May does to a set of books. Together the two named-only-by-letter customers are 37.8% of consolidated revenue in a single quarter. For the terms attaching to Customer B's likely contract, and why the enforceable portion is far smaller than the reported one, see the data centre schedule in section 5.

Where a widely-repeated figure gets downgraded. The "roughly one-fifth of revenue from the U.S. government" framing that appears in much of the coverage is directionally consistent with Customer A — but the 10-Q says "Customer A," not "the United States government." Identifying Customer A requires the risk-factor and business sections of the prospectus, which is a different document. Until that identification is made from a primary source, this page reports the concentration percentage as filed and the identity as Not verified. The concentration risk is real either way; the attribution is what we decline to assert.

Backlog and other filed detail

ItemFiled detail
Backlog$47,461 million at 30 June 2026, of which $14,286 million sits in deferred revenue. Approximately 56% expected within one year, 34% in one to three years, 10% thereafter.
Property, plant and equipment, gross$83,071 million, of which servers and networking equipment $34,771 million, satellites $13,788 million, machinery and equipment $9,453 million, construction in progress $12,554 million.
Depreciation$2,735 million in the quarter; $5,064 million for the half. Against $1,310 million and $2,547 million a year earlier.
Goodwill by segmentAI segment $11,130 million; Connectivity $515 million. There is no goodwill in the Space segment.
SpectrumEchoStar AWS-4 and H-Block licences covering 50 MHz, plus up to 15 MHz of unpaired AWS-3. FCC approval 12 May 2026; transfer to the holding trust closed 22 May 2026. $856 million paid to the trust and carried as a prepaid asset until acquisition closing.
Starlink subscribers and ARPUNot in the Form 10-Q. Disclosed one rung down, in the Q2 2026 earnings release: 12.0 million subscribers at 30 June 2026, ARPU $66 per month. Furnished

All figures Filed — Form 10-Q for the quarterly period ended June 30, 2026, and the notes thereto — except the final row, which is Furnished. The split matters: the balance-sheet and backlog figures above sit inside an SEC periodic report; the subscriber count sits in an 8-K exhibit. Both are real. Only one carries Section 18 liability.

On servers versus satellites. The gross PP&E table is the clearest single statement of what this company has become. SpaceX now carries $34.8 billion of servers and networking equipment against $13.8 billion of satellites — two and a half times as much computing hardware as orbital hardware, with another $12.6 billion in construction in progress that the filing says is primarily AI infrastructure and facilities. Depreciation more than doubled year over year. The capital intensity of this business is migrating from space to data centres, and it is doing so faster than the revenue mix is.
The Starlink numbers, and the one that should worry you. Subscribers doubled year over year, from 6.0 million to 12.0 million, and rose 1.7 million in the quarter alone. ARPU went the other way: $85 per month a year ago, $66 today — a 22% decline. Both facts are furnished in the same table, and read together they say something the subscriber headline alone does not. Connectivity revenue grew 66% while the subscriber base grew 100%, because the marginal Starlink customer is worth roughly three-quarters of the average customer of a year ago. That is what international expansion into lower-income geographies and a shift toward cheaper consumer tiers looks like in a single line. It is not necessarily bad — Connectivity operating income still grew 79%, faster than revenue — but anyone modelling this business off subscriber growth alone is modelling the wrong variable.
Tesla · Nasdaq: TSLA

2. Tesla — revenue up a quarter, operating income down more than half

The headline and the engine room disagree. Tesla's second quarter of 2026 produced revenue of $28,236 million, up 25.5% from $22,496 million a year earlier. Over the same span, income from operations fell from $923 million to $398 million — a 57% decline. Operating margin went from 4.1% to 1.4% while the top line grew a quarter.

Net income to common stockholders was $1,114 million, down 4.9%. The gap between a collapsing operating line and a roughly flat bottom line is closed almost entirely below the operating line: interest income of $422 million and other income of $590 million, against $81 million of interest expense, contributed $931 million to a pre-tax result of $1,329 million. Seventy percent of Tesla's pre-tax income this quarter came from things other than making and selling anything.

$ millionsQ2 2026Q2 2025H1 2026H1 2025
Automotive revenue20,51616,66136,75030,628
Energy generation and storage3,1392,7895,5475,519
Services and other4,5813,0468,3265,684
Total revenue28,23622,49650,62341,831
Total cost of revenue23,48518,61841,15234,800
Gross profit4,7513,8789,4717,031
Research and development2,3711,5894,3172,998
Selling, general and administrative1,9821,3663,8152,617
Total operating expenses4,3532,9558,1325,709
Income from operations3989231,3391,322
Interest income422392856792
Interest expense(81)(86)(173)(177)
Other income, net59032055201
Income before income taxes1,3291,5492,0772,138
Provision for income taxes201359458528
Net income to common stockholders1,1141,1721,5911,581
Earnings per share, basic / diluted$0.34 / $0.32$0.36 / $0.33$0.49 / $0.45$0.49 / $0.45

All figures Filed — Form 10-Q for the quarterly period ended June 30, 2026. One note on the operating expense block: research and development plus selling, general and administrative equal total operating expenses exactly in three of the four columns, but fall $94 million short in the first half of 2025. The filed subtotal is the one printed. The Institute did not read the additional line that accounts for the difference and does not name it here.

Gross profit by line of business

Tesla does not publish a segment cost build the way SpaceX does. It reports two segments — automotive, and energy generation and storage — and it stops at gross profit. But the quarterly update deck furnished as an exhibit to Form 8-K breaks revenue and cost of revenue into matched pairs, and a matched pair is a gross profit whether or not the company labels it one. The table below is that arithmetic, done here and checked three ways.

Q2 2026, $ millionsRevenueCost of revenueGross profitMargin
Automotive sales20,00616,8663,14015.7%
Automotive regulatory credits146146100%
Automotive leasing36418717748.6%
Total automotive20,51617,0533,46316.9%
Energy generation and storage3,1392,49964020.4%
Services and other4,5813,93364814.1%
Total28,23623,4854,75116.8%

Revenue and cost of revenue Furnished — Q2 2026 update, Exhibit 99.1 to Form 8-K. Gross profit and margin columns are the Institute's subtraction and division, not printed figures. Three checks were run: the three total-line gross profits ($3,463 + $640 + $648) sum to $4,751 million, which is the Filed consolidated gross profit in the Form 10-Q above; the derived automotive margin of 16.9% matches the GAAP automotive gross margin the company states in its own highlights; and the derived services gross profit of $648 million matches the record figure the company names in the same document. Regulatory credits are a separate line on the face of the income statement, not a component of automotive sales: $20,006 + $146 + $364 = $20,516 million. They carry no cost of revenue, which is why the margin column reads 100%.

Where Tesla actually makes its money, and where it does not. Energy is the highest-margin line in the company at 20.4%, four points above automotive — and it is the smallest of the three. Automotive, at 16.9%, includes $146 million of regulatory credits that cost nothing to produce; strip them out and the margin on physically manufacturing and selling cars falls to 16.3%. Services and other is now Tesla's second-largest revenue line at $4,581 million, larger than energy, growing 50% year over year, and earning the thinnest margin of the three at 14.1%. The mix is shifting toward the lowest-margin line, which is most of why gross margin sat flat at 16.8% while revenue grew a quarter.
What we could not publish, and why the table stops at one quarter. The 8-K exhibit presents this matched-pair detail for the quarter; the Institute did not obtain a comparable prior-year breakdown from a primary document, and will not construct one from secondary sources. Separately, Tesla's Form 10-Q segment note reports segment revenue and segment gross profit for the two reportable segments — a different cut than the one above, since Tesla's automotive segment combines the automotive and services-and-other revenue lines. The one segment figure we could tie is that combined automotive segment revenue: $25,097 million for the quarter and $45,076 million for the half, which reconciles exactly to the furnished lines ($20,516 + $4,581 = $25,097; $19,979 + $25,097 = $45,076). We could not read the full segment note itself, for the same truncation reason described on the SpaceX panel, and so we are not printing segment operating income for Tesla. It may well be there. We did not see it.
Balance sheet, $ millions30 Jun 202631 Dec 2025
Cash and cash equivalents15,21916,513
Total assets148,524137,806
Total liabilities61,00554,941
Redeemable noncontrolling interests in subsidiaries (temporary equity)5458
Total stockholders' equity86,85882,137
Noncontrolling interests in subsidiaries607670
Total liabilities and equity148,524137,806
Common shares outstanding (millions)3,9493,751

All figures Filed. Both columns foot to total assets, and the footing is the point: $61,005 + $54 + $86,858 + $607 = $148,524 million at 30 June, and $54,941 + $58 + $82,137 + $670 = $137,806 million at 31 December. Adding liabilities to total stockholders' equity alone leaves $661 million unaccounted for, because two separate minority-interest lines sit outside that subtotal — the redeemable noncontrolling interests are carried in temporary equity above the equity section under ASC 480-10-S99 since holders can require redemption, and the ordinary noncontrolling interests sit below the parent's equity subtotal. Weighted average basic shares for the quarter were 3,237 million and diluted 3,540 million; the 3,949 million figure is the period-end count from the balance sheet, which is the correct denominator for an ownership-percentage calculation and the wrong one for earnings per share.

Consolidated statement of cash flows

$ millions, six months ended 30 June20262025
Net income1,6191,610
Depreciation, amortisation and impairment3,2092,880
Share-based compensation2,1811,208
Inventory and purchase commitment write-downs187248
Foreign currency transaction loss, unrealised59954
Deferred income taxes(301)9
Unrealised gain on SpaceX equity investment(1,005)
Digital assets, net loss (gain)334(159)
Non-cash interest and other3373
Accounts receivable377601
Inventory(1,663)(2,407)
Operating lease vehicles26065
Prepaid expenses and other assets(1,028)(1,137)
Accounts payable and accrued liabilities3,3411,333
Deferred revenue491318
Net cash provided by operating activities8,6344,696
Purchases of property, plant and equipment(8,282)(3,886)
Purchase of SpaceX equity investment(2,002)
Purchases of short-term investments(16,281)(13,500)
Maturities of short-term investments15,62112,791
Purchases of intangible assets(7)
Net cash used in investing activities(10,951)(4,595)
Proceeds from issuances of debt4,6793,050
Repayments of debt(3,922)(4,129)
Debt issuance costs(4)(1)
Exercises of stock options and other issuances468528
Principal payments on finance leases(37)(67)
Proceeds from directors in shareholder settlement277
Recovery (payment) of settlement legal fees116(176)
Distributions to noncontrolling interests(91)(36)
Net cash provided by (used in) financing activities1,209(554)
Effect of exchange rate changes(83)151
Net decrease in cash and restricted cash(1,191)(302)
Cash and restricted cash, beginning of period17,61617,037
Cash and restricted cash, end of period16,42516,735

All figures Filed — Form 10-Q. All four sections were recomputed line by line and tie to the filed subtotals exactly in both years. The net income line is $1,619 million, not the $1,591 million attributable to common stockholders shown in the income statement above; the $28 million difference is income attributable to noncontrolling and redeemable noncontrolling interests, which is cash inside the consolidated group and therefore belongs at the top of this statement. Beginning and ending balances include restricted cash of $1,103 million and $1,206 million respectively, which is why they exceed the balance-sheet cash figures. Non-cash items disclosed separately: $2,633 million of property and equipment still sitting in liabilities and accrued expenses at period end, against $1,639 million a year earlier.

Tesla's free cash flow has almost disappeared. Operating cash of $8,634 million less capital expenditure of $8,282 million leaves $352 million of free cash flow for the half — down from $810 million a year ago, on revenue that grew 21%. Capex more than doubled while operating cash grew 84%. In the second quarter alone the company's own furnished figures show operating cash of $4,697 million against capex of $5,789 million, which is negative $1,092 million of free cash flow. Tesla is now, for the first time in this cycle, spending more on plant than the business produces. It has $43.5 billion of cash and short-term investments to do it with, so this is a choice rather than a constraint — but it is the same choice SpaceX is making, at one-thirtieth the scale.

Two lines here are new and neither is small. Tesla bought $2,002 million of SpaceX equity during the half and marked it up by $1,005 million — a 50% unrealised gain on a position it acquired in the same six months, recognised in earnings and then backed out of operating cash as a non-cash item. That $1,005 million is a fifth of Tesla's entire pre-tax income for the half. It is a mark on a related-party position in a company controlled by the same person, and it is the single item on Tesla's statements a governance-minded reader should look at hardest. Separately, share-based compensation of $2,181 million grew 81% and is now more than the company's entire net income — a real economic cost that never touches the cash flow statement as an outflow.

Where the R&D went. Research and development rose 49% year over year in the quarter, from $1,589 million to $2,371 million, and now consumes 8.4% of revenue against 7.1% a year ago. Gross margin was roughly flat at 16.8%. So the operating-income collapse is not a manufacturing story — gross profit actually grew $873 million — it is a spending story: operating expenses grew $1,398 million against $873 million of incremental gross profit. Tesla is funding something out of current earnings, and the filing's line items say most of it is engineering.
Musk's Tesla stake — what we can and cannot say. A Schedule 13G/A naming Elon Musk as the filing person and Tesla, Inc. as the subject was filed on 17 June 2026 under accession number 0001104659-26-075203. The Institute has confirmed the filing's existence, date, filer and subject directly on EDGAR. It has not read the share count or percentage in the primary document, and therefore prints neither. Two widely-circulated figures for Musk's Tesla position were tested against the filed share count of 3,949 million shares outstanding at 30 June 2026 and could not be reconciled to the percentages attributed to them. If you need this number, read the 13G/A itself. Not verified
xAI & X

3. xAI and X — why there is no separate set of books, and what that hides

The chain of ownership, in order. Twitter, Inc. was acquired by Elon Musk in October 2022. X.AI Corp. began operations in March 2023. On 28 March 2025, xAI — then X.AI Holdings Corp. — acquired X Holdings Corp. and X.AI Corp., making both wholly-owned subsidiaries. On 2 February 2026, SpaceX completed its acquisition of X.AI Holdings Corp. Each step was effected through a share exchange.

Both transactions were between entities under common control, and that phrase does specific accounting work. In a common-control transaction there is no acquirer and no acquiree in the purchase-accounting sense: no purchase price allocation, no fair-value step-up of the acquired assets, no new goodwill from the transaction itself. The receiving entity carries the transferred businesses at their existing book values, and — this is the part that trips people up — restates prior periods as though the combination had always existed.

The practical consequence for anyone reading these statements. SpaceX's "Q2 2025" comparative of $4,071 million in revenue is not what SpaceX reported or would have reported in 2025. It is the retrospectively combined figure for SpaceX, xAI and X together. Year-over-year growth of 92% is therefore a like-for-like organic comparison, not an acquisition effect — which is genuinely useful. But it also means no filed document anywhere states what SpaceX alone earned in 2025, and no filed document states a standalone valuation for xAI or for X. Any figure purporting to be one is an estimate, whatever confidence it is delivered with.
What people want to knowWhat is actually availableTag
xAI standalone revenueDoes not exist as a filed figure. The nearest primary datum is SpaceX's AI segment: $2,561M in Q2 2026, $3,379M for the half.Filed (segment)
X advertising revenue$367M in Q2 2026 against $426M in Q2 2025; $710M for the half against $870M. Disclosed as the Advertising line within the AI segment.Filed
AI solutions & infrastructure$2,194M in Q2 2026 against $311M in Q2 2025.Filed
Goodwill attributable to AI$11,130M at 30 June 2026, down from $11,296M at 31 December 2025 — the movement is currency translation, not impairment.Filed
Preferred converted in the xAI merger1,987 million preferred shares carrying $37,476M converted into 1,424 million common shares. $2,413M of common was repurchased from current and former xAI employees.Filed
A combined or standalone valuation for xAI or XNo filing contains one, and by construction none can: common-control accounting produces no transaction value.Does not exist
X user counts, engagement, ARPUNot disclosed. The 10-Q carries a "user base" intangible of $754M net, which is an amortising acquired asset, not a user metric.Not disclosed

The editorial point. The most interesting fact about X inside SpaceX is not the size of the advertising line but its direction. It is the only revenue line in the entire consolidated filing that is smaller than it was a year ago in both the quarter and the half. In a group where Starlink's government business doubled and AI infrastructure grew sevenfold, an eighteen-percent decline is easy to miss — and consolidation is precisely the structure that makes it easy to miss. That is not an accusation of anything; it is what segment reporting is for, and it is why a practitioner reads the segment note before the press release.

Neuralink & The Boring Company

4. The two that are still dark

Neuralink and The Boring Company remain private, file no periodic reports, and have no filed financial statements of any kind. There is no revenue figure, no loss figure, no balance sheet, and no audited valuation for either company. Everything published about their finances derives from private financing rounds, secondary-market marks, or people describing documents they will not produce.

This page will not carry those figures under a tag that implies more than they deserve. When either company registers securities, files a Form D that contains something substantive, or is consolidated into a registrant, it will appear here with filed numbers. Until then the honest entry is the blank one.

Why the blank matters more than a number would. Two of the six companies in this portfolio are structurally unobservable. An advisor building a client's exposure picture around this group is working with roughly two-thirds coverage, and the uncovered third is where the most speculative valuations sit. That ratio — how much of a concentrated position you can actually see into — is a better first question than any individual valuation. The family-office coordination framework treats it as the opening question for exactly this reason.
Rejected during preparation. Several specific claims about Neuralink — a patient-implant count and a mid-2026 financing round in a named jurisdiction — appeared only on content-farm aggregators with no primary source and no named reporter behind them. They are not on this page. When aggregator content and primary sources disagree, or when aggregator content simply has no primary source behind it, the aggregator loses.
Build rate

5. The physical plant — what is actually being built, and how fast

Financial statements measure a company in dollars, which is the right unit for a lender and the wrong unit for understanding what these two companies are. The comparison that gets made is to Henry J. Kaiser — the industrialist who took shipbuilding from a craft that produced a vessel in eight months to a process that produced one in a matter of days, not by inventing a better ship but by rebuilding the method of building. The claim implicit in that comparison is about rate: not what a company owns, but how quickly it can bring new productive capacity into existence.

That claim is testable. Both registrants disclose installed capacity, and both disclose the throughput they actually achieved against it. This panel puts the two numbers next to each other, because a nameplate figure with no utilisation figure beside it is marketing, and a utilisation figure with no nameplate beside it is noise.

Tesla — installed annual manufacturing capacity

SiteProductInstalled annual capacityStatus
CaliforniaModel 3 / Model Y>550,000 vehiclesProduction
ShanghaiModel 3 / Model Y>950,000 vehiclesProduction
BerlinModel Y>375,000 vehiclesProduction
TexasModel Y>250,000 vehiclesProduction
TexasCybertruck>125,000 vehiclesProduction
TexasCybercab>125,000 vehiclesProduction
NevadaTesla SemiNot statedCommissioning
RoadsterNot statedDesign development
Vehicles, stated capacity>2,375,000
CaliforniaMegapack40 GWhProduction
ShanghaiMegapack20 GWhProduction
NevadaPowerwall>6 GWhProduction
TexasMegapackNot statedCommissioning
Energy storage, stated capacity>66 GWh
CaliforniaOptimusNot statedConstruction
TexasOptimusNot statedConstruction

All figures Furnished — Q2 2026 update, Exhibit 99.1 to Form 8-K, "Installed Annual Manufacturing Capacity". The two subtotals are the Institute's addition of the stated lines and inherit the company's "greater than" qualifier; they exclude every site marked Commissioning, Construction or Design development, because those have no stated capacity. Status labels are the company's own vocabulary and they are load-bearing: Production, Commissioning, Construction and Design development describe four different distances from a saleable unit, and only the first is generating revenue.

The column that belongs in this table and cannot yet be built: floor area. Installed capacity measures what a plant can make. It does not measure how much plant exists, and for a thesis about industrial build rate, square footage is arguably the better number — it is the physical scale of what has actually been erected, independent of what the tooling inside it is rated at. We tried to add it as a fifth column here and could not. The reason is worth stating precisely, because it is a retrieval failure and not a disclosure failure. Tesla's Q2 2026 shareholder update was read in full, all thirty-two pages: it contains no square footage figure for any site. Site-level floor area for a US registrant normally appears in Item 2, Properties, of the Form 10-K — an annual disclosure that does not repeat in interim reports, and one that sits beyond the character limit this page has been unable to clear on long EDGAR documents. The same is true of SpaceX: the full 150-page filed 10-Q was searched and returns zero occurrences of “square feet.” So the honest position is that the figure is not disclosed in any interim document and has not been retrieved from the annual one. Widely-circulated building sizes for individual sites — including the 810,000 square foot Southaven warehouse carried in the reported table further down — come from press coverage and county records, not filings. They will not be promoted into this table. When Item 2 is read at the source, the column goes in.

Tesla — supporting infrastructure

SiteFacilityInstalled capacityStatus
TexasCortex 1 — AI training compute>90 MWProduction
TexasCortex 2 — AI training compute>115 MWProduction
Texas4680 battery cell>40 GWhProduction
TexasLithium refining30 GWh equivalentEarly ramp
TexasCathode materials10 GWh equivalentEarly ramp
NevadaLFP battery cell7 GWhEarly ramp
Berlin4680 battery cellNot statedConstruction

All figures Furnished — same exhibit, "Supporting Infrastructure". The company states in the same document that it "more than doubled" its onsite compute in Texas, measured in megawatts, during the first half of 2026.

What the plant actually produced

Tesla, Q2 2026AchievedAgainst nameplate
Vehicles produced451,75876%
Vehicles delivered480,12681%
Energy storage deployed13.5 GWh82%
Supercharger stations8,704
Supercharger connectors82,357
Active FSD subscriptions1,480,000

Achieved figures Furnished. The "against nameplate" column is the Institute's arithmetic: the quarterly figure multiplied by four and divided by the stated annual capacity above. It is a rough measure and should be read as one — a single quarter annualised is not a run rate, deliveries can exceed production out of inventory, and storage deployed is not the same thing as storage manufactured. It is printed because a capacity table without it invites the reader to treat nameplate as output, which is the most common error made about this company in both directions.

SpaceX — compute, orbit and where the capital actually went

Physical throughputQ2 2026Q1 2026Q2 2025
Nameplate AI compute capacity1.4 GW1.0 GW0.4 GW
Customer launches1079
Internal launches283337
Total launches384046
Mass to orbit, metric tons485556652
Average mass per launch, metric tons12.813.914.2
Starlink subscribers, millions12.010.36.0

Compute, launch and subscriber figures Furnished — Q2 2026 earnings release. Average mass per launch is the Institute's division. See the warning on the SpaceX panel regarding a typographical error in the release's own description of the 0.4 GW figure.

SpaceX gross property, plant and equipment at 30 June 2026$ millionsShare
Servers and networking equipment34,77141.9%
Satellites13,78816.6%
Construction in progress12,55415.1%
Machinery and equipment9,45311.4%
Data centre infrastructure3,9914.8%
Launch sites3,1183.8%
Land and buildings2,9583.6%
Flight vehicle hardware1,5571.9%
Leasehold improvements8811.1%
Gross property, plant and equipment83,071100%

Dollar figures Filed — Form 10-Q. Share column is the Institute's division; components sum to the filed gross total. Construction in progress is described in the filing as primarily AI infrastructure, which means the compute share of this balance sheet is understated by the table, not overstated. Share percentages are rounded independently and sum to 100.2%; the dollar components sum to the filed total exactly.

The data centre schedule — every site the filings actually name

Until this revision the page reported the AI build as a dollar figure and a capacity figure and never once said where. That was a real gap, and it turns out to be closeable. The registration statement SpaceX filed for its June 2026 listing names its data centres, places them by road and by state, and gives the build times — which means facility-level detail on this company is Filed, not merely reported, and belongs on this page at the top rung. What follows is every facility named in a document SpaceX has filed with the Commission, and then, kept deliberately separate, the sites that appear only in press coverage and county records. The line between the two tables is the entire purpose of this page.

Facility, as the filing names itLocation, as the filing places itWhat the filing saysRung
COLOSSUSPaul R. Lowry Road, Memphis, TennesseeThe company's “flagship data center.” Its first cluster was brought online in 122 days, in the repurposed shell of an existing factory.Filed
COLOSSUS IIMemphis, Tennessee and Southaven, MississippiDefined in the plural — “our data centers” — and described as a single coherent gigawatt-scale training cluster spanning two states and two utility jurisdictions. First cluster online in 91 days. Grok-5 was being trained here at the time of the filing.Filed
Both, togetherTennessee and Mississippi“Approximately 1.0 gigawatt of compute power, with additional power capacity available for data center operations.”Filed

All three rows Filed — final prospectus, Form 424(b)(4), added to the sources table in section 8. Two of these facts are worth holding onto. First, the 1.0 GW in the prospectus and the 1.0 GW in the Q1 2026 column of the earnings release are the same number arrived at independently, one on the Filed rung and one on the Furnished rung, for the same period. That is the strongest form of corroboration available without an audit. Second, the filing supplies its own benchmark for the build times: it states that an industry benchmark to bring a 100 megawatt greenfield data centre online is approximately two years. COLOSSUS II's first cluster took 91 days. The comparison is the company's own and is presented as such.

Reported, and not found in any filing read hereLocationWhat is reportedRung
The COLOSSUS host buildingA former appliance plant, south Memphis, TennesseeGrid supply of 150 MW approved in November 2024 and energised May 2025, with a further allocation of up to 150 MW approved in February 2026. Installed GPU counts of 200,000 and of more than 220,000 both circulate; they do not agree with each other, and neither is filed.Reported
The Tulane Road buildingWhitehaven, Memphis, TennesseeA warehouse and adjacent parcels acquired by an affiliate in early 2025 for roughly $80 million.Reported
The Stateline Road buildingSouthaven, DeSoto County, MississippiA former logistics warehouse of roughly 810,000 square feet on about 48 acres, acquired December 2025.Reported
A fourth Memphis buildingTulane Road corridor, Memphis, TennesseeAnnounced publicly in July 2026, with a reported configuration of 220,000 next-generation processors. No document SpaceX has filed and this page has read names it.Reported
Southaven generationDeSoto County, MississippiA 1.2 GW gas plant of 41 turbines permitted under the Clean Air Act in March 2026. Separately, 69 trailer-mounted turbines at the same site are under an agreed decommissioning order with the state running into 2027.Reported

All five rows Reported — press coverage and state permitting records, not filings. They are shown because a reader deciding what to believe is better served by seeing the reported layer than by having it silently withheld, and worse served by having it presented at the same weight as the prospectus. None of these figures is used anywhere else on this page, and none feeds any calculation.

One naming trap, because it has already caught people. Press coverage refers to two of the Memphis buildings as “Macrohard” and “Macroharder,” after rooftop signage. In SpaceX's own filed definitions, Macrohard is not a building. The prospectus defines it as a software platform under development with Tesla, designed to emulate digital workflows and operate computers autonomously. A reader who takes the rooftop as the corporate nomenclature will end up counting a software project as a data centre, and may double-count capacity. The filed vocabulary has exactly two data-centre names in it: COLOSSUS and COLOSSUS II.

Who is paying whom, and who owns the buildings

Naming the buildings answers half the question. The other half is commercial: somebody is writing a very large cheque every month for the compute inside them, somebody holds title to the shells, and somebody owns the accelerators. Those are three different parties, and the filings separate them — sometimes explicitly, sometimes only by implication, and in one case not at all. What follows is every counterparty relationship the documents name, with the rung attached to each.

CounterpartyDirectionWhat the document saysRung
Anthropic PBCPays SpaceXCloud Services Agreements signed May 2026 covering approximately 325,000 NVIDIA GPUs across COLOSSUS and COLOSSUS II, at $1.25 billion per month through May 2029, ramping in May and June 2026 at a reduced fee. After an initial three-month period, either party may terminate on 90 days' notice. This is the only cloud customer named by name in any document read here, and it is named in a filed one.Filed
GooglePays SpaceXAn agreement described as approximately $920 million per month for 32 months beginning October 2026. Discussed on the Q2 2026 earnings call. It appears in no filed document read here, and no dollar figure attached to it has been filed.Reported
“Customer B”Pays SpaceX19.5% of consolidated revenue in the quarter, disclosed as sitting entirely within the AI segment and absent from the comparable prior-year period — approximately $1,524 million. The 10-Q does not say who this is.Filed
“Customer A”Pays SpaceX18.3% of consolidated revenue in the quarter. Not attributed to a segment in the concentration note, and not named.Filed
CTC Property, LLCSpaceX subsidiaryNamed in the 10-Q as a subsidiary of the company, a joint-and-several guarantor of the SpaceX Bridge Loan alongside X Corp. and X.AI LLC, and the entity that holds the AI infrastructure assets subject to the sale-leaseback transactions below. The filings read here do not set out site-by-site title, acreage or purchase price.Filed
Valor Equity PartnersSpaceX paysAn equipment lease relationship with CTC, entered April 2026, disclosed as a related party: Valor's founder and chief executive, Antonio Gracias, is a director of SpaceX. The transactions were deemed failed sale-leasebacks and are therefore carried as borrowings rather than leases — $2,039 million current and $11,290 million non-current, $13,329 million in total, against $455 million and $4,052 million at 31 December 2025. Related interest expense was $327 million in the quarter and $513 million for the half.Filed
Tesla, Inc.SpaceX pays$295 million of Megapack products purchased from Tesla in the quarter and $329 million for the half, recorded in property, plant and equipment. As at 31 December 2025 the company had purchased $506 million of Megapacks and $131 million of Cybertrucks at manufacturer's suggested retail price. All other transactions with Tesla in the period are described as immaterial. Tesla is a vendor to the data centres, not a co-owner of them: no filing read here describes any joint ownership, joint venture or shared-title arrangement between SpaceX and Tesla over any facility.Filed
Unnamed cloud providersSpaceX paysSpaceX's own purchase commitments total $27,955 million, of which $22,244 million falls due in 2027, described as third-party cloud capacity. The counterparties are not named.Filed

Rungs as marked — every row Filed but Google. Two cautions, both of which have caught commentary already. First, do not equate Customer B with Anthropic. The concentration note discloses a measured percentage without attribution; the prospectus discloses a named contract without a revenue figure. The two are consistent with each other and the arithmetic is suggestive, but the filing does not join them, and neither does this page. Second, Google is on the Reported rung and stays there until it appears in a filed document as a customer. An earnings call is a named source and a real one; it is not a signed filing, and the monthly figure attributed to it has never been filed. To be exact about it, since this page is built on exactness: the word “Google” does appear once in SpaceX's filed 10-Q, and it has nothing to do with cloud services — it is in the legal proceedings note, naming Google among the owners of tracking technologies at issue in a putative class action filed against X.AI LLC in May 2026. No filed document read here names Google as a customer of anything.

The ownership question, answered as far as the filings answer it — and no further. SpaceX states at the company level that it owns and operates its data centres, and the property sits on its own balance sheet: $3,991 million of data centre infrastructure and $34,771 million of servers and networking equipment, all of it consolidated, none of it equity-method or off-balance-sheet. The holding vehicle named in the filings is CTC Property, LLC. What is not disclosed in any document read here is site-level title — which legal entity holds the deed to which building, on what terms, and whether any parcel is leased rather than owned. And the answer to the natural follow-up is a clean negative: no filing read here describes Tesla as owning any part of any data centre. Tesla appears once in this context, as a supplier of $295 million of batteries. The two companies are related parties, they transact, and they are separately capitalised. Combining their data centre estates in a single mental balance sheet is an error that the filings do not support.
Two numbers that should be read together, because they run in opposite directions. SpaceX's remaining performance obligations — revenue it has contracted to earn but not yet recognised — are $47,461 million. Its own purchase commitments are $27,955 million, with $22,244 million of that concentrated in a single year, 2027. So the company has roughly $1.70 of contracted future revenue for every dollar of contracted future spending, which sounds comfortable until the timing is laid alongside it: the spending has a hard date and the revenue, on the terms disclosed above, is substantially cancellable on ninety days. That is the same asymmetry the contract callout describes, expressed in the commitments note instead of the risk factors. A practitioner modelling 2027 for this company should start here.

What the money actually bought, six months of it

SpaceX gross property, plant and equipment, $ millions30 Jun 202631 Dec 2025ChangeGrowth
Servers and networking equipment34,77122,69412,07753.2%
Construction in progress12,5544,6047,950172.7%
Data centre infrastructure3,9912,9601,03134.8%
The three compute-related lines51,31630,25821,05869.6%
Satellites13,78811,9491,83915.4%
Machinery and equipment9,4536,3433,11049.0%
Launch sites3,1182,40471429.7%
Land and buildings2,9581,8761,08257.7%
Flight vehicle hardware1,5571,689(132)(7.8%)
Leasehold improvements8817849712.4%
Everything else31,75525,0456,71026.8%
Gross property, plant and equipment83,07155,30327,76850.2%

Both dollar columns Filed — Form 10-Q, property, plant and equipment note. Change and growth columns are the Institute's arithmetic. Four checks were run: the nine components sum to the filed gross total in both columns; the two subtotals sum to the filed gross total in both columns; each change equals the difference of the two filed balances; and the total change of $27,768 million equals $83,071 less $55,303. Flight vehicle hardware is the one line that fell, which in a gross-cost schedule means retirement or disposal rather than depreciation.

The shells are the cheap part, and that is the finding. When a company says it spent $23.6 billion on data centres, the instinct is to picture buildings. The filing says otherwise. Data centre infrastructure — the structures, the power distribution, the cooling — is $3,991 million, 4.8% of everything SpaceX owns. The servers and networking equipment inside those structures are $34,771 million: 8.7 times the buildings that house them. That ratio is the whole economics of this business, because concrete depreciates over thirty years and accelerators depreciate over three to five. It is already showing up: SpaceX's depreciation for the half was $5,064 million against $2,547 million a year earlier, very nearly a doubling, and it will keep climbing as the $12,554 million sitting in construction in progress is placed in service. The cost of the build arrives on the income statement before the revenue from it does. Any model of this company that gets the depreciation schedule wrong gets the next three years wrong.
The data centres are not yet a profit driver, and the word matters. It is natural to describe this build as SpaceX's profit engine, and measured by revenue and by capital it plainly is the engine: AI took $23,551 million of the $28,476 million of capital deployed in the half, 82.7% of everything, and grew segment revenue from $737 million to $2,561 million year over year. But on the income statement the AI segment lost $1,257 million in the quarter and $3,726 million in the half. The profit driver is Connectivity, which earned $2,844 million of operating income on $7,548 million of revenue — a 37.7% margin for the half, 38.6% in the quarter — and which is currently funding the other two segments. AI's Segment Adjusted EBITDA did turn positive for the first time, at $1,146 million against negative $276 million a year earlier, and that is a genuine inflection. But Segment Adjusted EBITDA is defined by the company to exclude depreciation, and depreciation on a three-year server is not an accounting artefact — it is the principal cost of the product. A compute business that is EBITDA-positive and operating-loss-negative is telling you precisely one thing: the assets are not yet earning back their own depreciation. The correct sentence today is that AI is the revenue and capital driver and Connectivity is the profit driver. The correct sentence in four quarters may be different, and the line to watch is segment operating income, not Segment Adjusted EBITDA.
The largest compute contract is cancellable on ninety days' notice. The prospectus discloses the terms this page previously described as undisclosed, and they are worth reading closely. The Cloud Services Agreements signed in May 2026 cover approximately 325,000 NVIDIA GPUs across COLOSSUS and COLOSSUS II at $1.25 billion per month through May 2029 — a $15.0 billion annual run rate — with capacity ramping in May and June 2026 at a reduced fee. After the initial three-month period, either party may terminate on 90 days' notice. Two consequences follow. First, the furnished and filed documents corroborate each other: two months at the full rate would be $2.5 billion, the earnings release reports $1.6 billion of incremental revenue, and the 64% shortfall is exactly what the filed “reduced fee” ramp language predicts. Second, and less comfortably, the earnings release attributes $1.6 billion of a $1,743 million sequential increase in AI revenue — 92% of the segment's entire growth — to these agreements. The 10-Q narrows it further. An unnamed Customer B, disclosed as sitting entirely within the AI segment and absent from the comparable period a year earlier, was 19.5% of consolidated revenue in the quarter. Against $7,814 million of revenue that is approximately $1,524 million from one counterparty — 59.5% of everything the AI segment earned, and 87% of its sequential growth, from a customer that did not exist twelve months ago. The capital committed against it is measured in decades of concrete and years of silicon. That asymmetry between contract duration and asset duration is the single most important thing on this page for anyone underwriting the AI segment.
The most important sentence in the earnings release is in the definitions, and almost nobody reads it. The release announces “several Cloud Services Agreements totaling $14.1 billion in contracted sales.” That number looks small next to the coverage, and it is. The filed Anthropic terms — $1.25 billion a month over roughly 37 months to May 2029 — imply about $46 billion of nominal value on their own Estimated. Add the separately reported agreement with Google, at $920 million a month for 32 months from October 2026, and the headline is roughly $75 billion Reported. SpaceX reports $14.1 billion.

The gap is not a discrepancy. It is a definition, and the company prints it in Note 2: contracted sales means the value of contracts signed for the non-cancellable, enforceable period, and expressly excludes estimated revenue for future periods cancellable by either the company or the customer. So $14.1 billion is the part that is actually enforceable. The remainder — on these numbers about four-fifths of the headline — is capacity either side may walk away from.

That is SpaceX grading its own contracts, and grading them honestly. A reader working from the press figures carries roughly $75 billion of assumed revenue against $83 billion of gross plant. A reader working from the company's own definition carries $14.1 billion, or about nineteen cents of enforceable revenue for every dollar of headline. Both numbers are true. Only one of them is a receivable, and the company tells you which. Note that the Google agreement is Reported and is not named in any document read here — SpaceX names no cloud customer in the earnings release, and the 10-Q identifies its largest only as Customer B.
What a megawatt appears to cost, with the caveat attached. Nameplate compute rose from 1.0 GW to 1.4 GW during the second quarter while the AI segment spent $15,828 million of capital, which divides to roughly $39.6 million per megawatt energised Estimated. That figure almost certainly overstates the true cost of a megawatt, and knowingly so: a large share of the quarter's spending bought capacity that is not yet in service, which is why construction in progress rose $7,950 million over the half. The number is offered as a ceiling, not a measurement. A true cost per megawatt would require the company to disclose capital expenditure by facility and by in-service date, and it does not.
The build-rate finding, stated plainly. SpaceX took its AI compute from 0.4 GW to 1.4 GW in four quarters — three and a half times, adding a gigawatt of nameplate capacity in a year. Tesla's entire disclosed AI training estate, Cortex 1 and Cortex 2 together, is 205 MW: SpaceX is running roughly seven times Tesla's compute. On the balance sheet the same fact appears differently — servers, networking and data centre infrastructure are $38,762 million of SpaceX's $83,071 million of gross plant, 46.7% of everything the company owns, against 22.2% for satellites, launch sites and flight vehicle hardware combined. Measured by the assets rather than the name on the building, SpaceX is already more a compute company than a space company, by a factor of two.
And the finding that cuts the other way. The launch business — the actual Kaiser comparison, the thing that builds and flies hardware — went backwards in physical terms. Total launches fell from 46 to 38 year over year and from 40 sequentially. Mass to orbit fell from 652 to 485 metric tons, down 26%, and for the half from 1,102 to 1,041. Average mass per launch declined in each of the three periods shown. Space segment revenue still grew 29%, which means the revenue per unit of physical throughput rose sharply — better pricing, better mix, or both. But whatever else is accelerating at this company, tonnage to orbit is not. A thesis built on build rate has to survive its own metric, and this quarter that metric declined.

What a practitioner should take from this panel. The build-at-scale thesis is real but it is not uniform, and the two halves of it are moving in opposite directions. Capital is compounding into compute at a rate that has no obvious precedent in either company's history — $23.6 billion of AI capex at SpaceX in six months, a doubling of Tesla's Texas compute in the same six months. Meanwhile the manufacturing estate at Tesla is running at roughly three-quarters of its own stated nameplate, with four separate product lines sitting in Commissioning, Construction or Design development and therefore contributing nothing, and SpaceX's launch cadence is declining. Both of those are ordinary facts about industrial companies mid-buildout. Neither is a scandal. But an investor holding this thesis should know that what is being built at unprecedented speed today is data centres, and that the vehicle plants and the launch pads — the parts the Kaiser comparison actually invokes — are the parts currently running below their own capacity.

The Texas fab — the gap between what was announced and what has been filed

In August 2026 the Governor of Texas announced what would be, if built as described, the largest single industrial project either company has attempted: a semiconductor fabrication complex in Grimes County, north-west of Houston, referred to publicly as a “Terafab.” The announcement is specific and it is large. What follows is that announcement set directly against what the registrants have actually filed, because the distance between the two is the single most instructive thing about it.

The TerafabWhat the source saysRung
Capital, phase one$16.8 billion, per the Governor's office, 6 August 2026.Reported
Employment3,000 jobs, per the same announcement.Reported
Floor area100 million square feet at full build, per the same announcement.Reported
State incentiveA $30 million Texas Enterprise Fund grant.Reported
In SpaceX's Form 10-QZero mentions. A full-text search of the filed 10-Q returns no instance of “Terafab,” “semiconductor,” “fab” or “foundry.”Filed
In SpaceX's prospectusA framework only. The filed 424(b)(4) contemplates chip fabrication in general terms and states expressly that “any specific projects … have not yet been determined.”Filed
In Tesla's Q2 updateTesla names a fab, but it is its own, in Austin — not Grimes County, and not described as shared.Furnished
Capital committed on any balance sheetNot disclosed. No commitment, purchase obligation, construction contract or contingency in either registrant's filings read here is identified with this project.

The rung column here is doing unusual work and is worth reading literally. The Filed rows are not filed confirmations of the project — they are filed silences, and a filed silence is evidence. A registrant that had committed $16.8 billion of capital would ordinarily disclose it: in commitments, in construction in progress, in liquidity, or in risk factors. As of the 30 June 2026 balance sheet date, none of that appears.

Both things are probably true, and the sequencing is the point. There is no reason to doubt that the announcement happened or that the state grant is real; a governor's office and a state incentive fund are named, accountable sources. But an announcement on 6 August 2026 sits after a 30 June balance sheet date, so its absence from the 10-Q is not yet a contradiction — it is a timing gap that the next filing will close one way or the other. What can be said today, precisely: a project described publicly at $16.8 billion has no filed capital commitment behind it, and the registrant's own prospectus says the specific projects have not been determined. The practitioner discipline is to carry it at zero until a filing carries it at something, and to watch three places in the Q3 10-Q — the commitments note, construction in progress, and the subsequent-events note. If it is real at the scale announced, it will appear in at least one of them.

Solar — one sentence, no figures

Tesla's Q2 2026 shareholder update contains a single sentence indicating an intention to manufacture solar panels. That is the whole of it. There is no capacity figure, no site, no capital number, and no solar line anywhere in the installed annual manufacturing capacity table above — which is notable, because that table is where Tesla itself puts every product it can attach a number to, including products that are only in Construction. A product absent from a company's own capacity table is a product the company is not yet prepared to quantify.

Separately, press coverage describes a project referred to as “Crystal Sun,” a roughly $10.1 billion facility in Fort Bend County, Texas, targeted for the first quarter of 2029. That figure is Reported and appears in no filing read here. It is recorded on this page so that a reader who encounters it elsewhere knows where it sits, and it feeds no calculation.

What would move solar up a rung. A stated annual capacity in gigawatts, a named site, or a capital commitment in the commitments note. Any one of the three would put a solar row in the capacity table. Until then the honest statement is that Tesla has said it intends to build solar panels and has disclosed nothing measurable about doing so — which is a different thing from Tesla having no plan, and a different thing again from the $10.1 billion number being wrong. It is simply not yet a filed fact.
What is not on this page, and will not be until it is filed. Energy generated — as opposed to storage capacity manufactured and deployed — is not disclosed by either registrant in the documents read here. Neither is data centre power consumption, water usage, land area, headcount by site, or square footage. Megawatts of compute is a capacity figure, not a consumption figure, and the two are not interchangeable. On facility detail this callout has been narrowed: SpaceX's registration statement does name and locate its data centres, and those rows now sit at the top of the schedule above on the Filed rung. What remains undisclosed is everything quantitative below the site name — no filing read here gives capital expenditure by facility, in-service date by facility, square footage, acreage, GPU count, power draw, water consumption or headcount by site. The widely-circulated figures for those things trace to press coverage and county permitting records, are carried in the separate Reported table, and feed no calculation on this page. If they enter a filing, they move up a rung the next morning.
Autonomy

6. Robotaxi and Full Self-Driving — what is deployed, what is subscribed, and what is carefully not said

Autonomy is the part of the Tesla story where the distance between the narrative and the disclosure is widest, and it is also the part where the company's own documents are most useful — not because they say a great deal, but because what they choose to say and not say is unusually legible. Two products sit here. Cybercab is a vehicle and a service; Full Self-Driving is a software subscription attached to the existing fleet. They are financially unrelated today and are treated separately below.

Cybercab and the robotaxi service — the deployment table

Robotaxi and CybercabWhat the document saysRung
Cybercab production status“Started production” at the Texas plant.Furnished
Cybercab installed capacity>125,000 vehicles per year, Texas, listed under Production in the company's own capacity table.Furnished
Service footprintDescribed as operating across seven major metropolitan areas.Furnished
Bay Area operating basisThe company's own footnote states the Bay Area service runs with a safety driver, under a limousine-class Transportation Charter Party permit, TCP0046782-A.Furnished
Cybercabs actually builtNot disclosed. No unit figure for Cybercab appears in the production and delivery table, which reports vehicles only in aggregate.
Vehicles in the robotaxi fleetNot disclosed.
Rides completed, or miles driven in serviceNot disclosed.
Robotaxi revenueNot disclosed as a line. Any robotaxi revenue is inside the services and other line, which is $4,581 million in total and is not broken down.
Which seven metros, by nameNot disclosed as a complete list in the documents read here.

Positive figures Furnished — Q2 2026 update, Exhibit 99.1 to Form 8-K. Every “not disclosed” row above means exactly that: the Institute read the document and the figure is not in it. It does not mean the figure could not be found elsewhere, and it does not mean the company does not have it.

The one place Tesla's own document argues with itself, and it is worth sitting with. The same exhibit counts the Bay Area among a service footprint of “seven major metropolitan areas” and, in its own footnote, discloses that the Bay Area operation runs with a safety driver behind the wheel under a limousine-class charter permit. A chauffeured car dispatched by an app is a legitimate business and a sensible regulatory on-ramp. It is not a robotaxi. Tesla does not claim it is — the footnote is Tesla's own, and disclosing it was a choice the company did not have to make in that form. But a reader who takes “seven metros” at face value and does not read the footnote will materially overstate how much driverless service exists. Without per-metro disclosure of which markets are driverless and which are supervised, the honest count of genuinely autonomous metros is somewhere between one and seven, and this page cannot narrow it further. That is a disclosure gap, not an accusation.
What the capacity figure does and does not tell you. Cybercab sits in Tesla's capacity table at more than 125,000 vehicles a year, in Production, in Texas — the same line and the same status as Cybertruck. That is a real and checkable statement about installed tooling. It is not a statement about output: the production and delivery table reports 451,758 vehicles produced in the quarter without a model split, so the number of Cybercabs that physically exist is not derivable from any document read here. A nameplate of 125,000 and an output of zero would both be consistent with everything Tesla has disclosed. When the model split appears, this row becomes the most informative single figure on the autonomy story.

Full Self-Driving — the subscription business, and the accounting trap inside it

Full Self-DrivingWhat the document saysRung
Active subscriptions1,480,000, up 56% year over year.Furnished
Attach rateGreater than 55% of the eligible fleet.Furnished
Cumulative distanceMore than 50 million kilometres driven on the system.Furnished
Software version and marketsv14 “lite” released in the United States and South Korea; four new European approvals during the period.Furnished
ChinaAbsent from the document. No FSD status for China appears in anything read here.Not disclosed
Deferred revenue, total balance$4,050 million at 30 June 2026, against $3,870 million at 31 December 2025.Filed
Deferred revenue recognised, six months$468 million.Filed
Expected to be recognised, next twelve months$962 million.Filed
FSD subscription priceNot disclosed in either document read here.
FSD revenue, isolatedNot disclosed. See the caution immediately below.

Operating metrics Furnished — Q2 2026 update. Deferred revenue figures Filed — Form 10-Q revenue note.

The deferred revenue balance is not an FSD number, and reading it as one is the most common error made about this line. Tesla's filed revenue note describes the $4,050 million balance as covering Full Self-Driving, connectivity, Supercharging access and over-the-air software updates — four products in one bucket. The company does not split it. So the widely repeated move of dividing the deferred balance by the subscriber count to derive an implied FSD price per user is arithmetic performed on the wrong numerator, and it will overstate FSD every time by an unknown amount. Likewise the $962 million expected to be recognised over the next twelve months is a four-product figure, not an FSD run rate.

What can be said cleanly is narrower and still useful: the combined bucket grew $180 million over the half while recognising $468 million, which means gross additions to deferred revenue of roughly $648 million Estimated — the four products together are being sold faster than they are being recognised. That is the shape of a growing subscription business. Attributing it to FSD specifically requires a disclosure Tesla has not made.
The one autonomy figure that is unambiguous, and it is the attach rate. More than 55% of the eligible fleet subscribes, and subscriptions grew 56% year over year to 1.48 million. Neither figure depends on an allocation, a price assumption or a segment definition — they are counts, furnished by the company, and they are the cleanest evidence on this page that FSD is a real business rather than a feature. What they cannot tell you is what it earns. With no disclosed price and no isolated revenue line, a 1.48 million-subscriber business of entirely undisclosed unit economics is the accurate description. The gap between those two sentences is the whole difficulty of underwriting Tesla's autonomy story from public documents.
What would change this section. Four disclosures, in descending order of how much they would matter: a Cybercab unit production figure in the delivery table; a per-metro breakdown of which robotaxi markets are driverless and which are supervised; deferred revenue disaggregated by product; and a stated FSD subscription price. The first two are operating disclosures Tesla could make at any time. The third is an accounting policy choice. Any of them appearing in a filing moves several rows above from Not disclosed to Filed the next morning, and this page will say so when they do.
Cross-company view

7. Five observations that only appear when you put the filings side by side

Both companies burned free cash flow this quarter, and only one of them can afford to.

For the six months, SpaceX generated $3,466 million of operating cash against $28,476 million of capital expenditure — free cash flow of negative $25,010 million. Tesla generated $8,634 million against $8,282 million — positive $352 million, down from $810 million a year ago, and negative $1,092 million in the second quarter alone. The direction is identical and the magnitude is not: SpaceX is consuming cash at roughly seventy times Tesla's rate, and it is doing so with $100.0 billion of liquidity and a freshly opened equity market behind it. Tesla is consuming cash with $43.5 billion. Neither is in difficulty. But both companies have now crossed from funding their own expansion out of operations to funding it out of the balance sheet, in the same six months, under the same person. For a client with concentrated exposure to both, that is one bet, not two.

SpaceX now carries more book equity than Tesla.

Shareholders' equity of $127,224 million against Tesla's $86,858 million, and cash of $93,522 million against Tesla's $15,219 million. Six months ago SpaceX's book equity was $2,573 million. The entire move is the IPO and the preferred conversion, not earnings — the company lost $4.8 billion over the same period. Book equity measures capital raised and retained, not value created, and there is no cleaner illustration of the distinction in current markets.

One of these companies is capital-constrained and it is not the loss-maker.

Tesla generates profits and holds $15.2 billion of cash; SpaceX generates losses and holds $100.0 billion of cash and securities. If the two were being run by the same treasury function — they are not, and cannot be — the money would flow the other way from where the income statements suggest. For an advisor, the lesson is that liquidity and profitability are separate questions and get separate answers.

Both companies are spending into the same bet.

Tesla's R&D rose 49% year over year; SpaceX's rose 81% ($1,958M to $3,548M in the quarter) and now exceeds its cost of revenue. SpaceX's construction in progress is $12,554 million, primarily AI infrastructure. Two separately-governed public companies, controlled by the same person, are simultaneously converting current earnings and raised capital into compute and engineering. Whether that is conviction or correlation is a governance question, not an accounting one — but it is a question a concentrated holder should be asking.

The related-party line is the one to watch.

SpaceX carries $13,329 million of related-party debt and finance lease obligations, a third of its total. The filing directs readers to its related-party note for detail on property, plant and equipment arrangements. Related-party financing at this scale, inside a controlled company with a dual-class structure, is the single item on these statements most worth reading in full each quarter.

Net operating losses — the widely-assumed tax shield, and what the filings actually contain

A recurring claim about both companies is that each carries a very large net operating loss carryforward capable of sheltering most of its taxable income for years. The claim is plausible on its face — SpaceX has lost $4.8 billion in six months and Tesla lost money for most of its first fifteen years — and the tax mechanic behind it is real. It is also, in the documents read here, unsupported by any filed figure, and this page will not print a number it has not read.

Take the statutory mechanic first, because it is law rather than disclosure and can be stated with certainty. For losses arising in tax years beginning after 31 December 2017, the Tax Cuts and Jobs Act amended Internal Revenue Code section 172 so that federal net operating losses carry forward indefinitely but may offset only 80% of taxable income in the year of use. So the “80%” in the common formulation is correct as a description of the ceiling on post-2017 federal NOLs. It is a rule, not a fact about either company. Separately, section 382 limits the annual use of pre-change losses after an ownership change — a live consideration for a company that has just completed a public offering and converted its preferred stock.

Tax position, as disclosedSpaceXTeslaRung
Federal NOL carryforwardNot disclosedNot retrieved
State NOL carryforwardNot disclosedNot retrieved
Section 382 limitationNot disclosedNot retrieved
Deferred tax assets, 30 Jun 2026$354M$7,235MFiled
Deferred tax assets, 31 Dec 2025$141M$6,925MFiled
Income before tax, Q2 2026$(518)M$1,329MFiled
Income before tax, six months$(4,788)M$2,077MFiled
Income tax provision, Q2 2026$23M$201MFiled
Income tax provision, six months$29M$458MFiled
Effective tax rate, as statedNot statedNot stated

Dollar figures Filed — the two Forms 10-Q. Neither interim report states an effective tax rate; the ratios a reader can compute from the rows above are the Institute's division and are quarterly, which is not the same thing as an annual effective rate. The two “not” labels in the top three rows are deliberately different words and the difference is the whole discipline of this page. For SpaceX, Not disclosed is a finding: the complete 150-page filed 10-Q was searched end to end and returns zero occurrences of “net operating loss,” “carryforward” or “Section 382.” The figure is not in the document. For Tesla, Not retrieved is an admission: the carryforward detail lives in the annual tax footnote of the Form 10-K, and that footnote sits beyond the retrieval limit this page has been unable to clear. It may well be there. We did not see it, and so we are not printing it.

Two published NOL figures for SpaceX are circulating and should not be relied on. In the course of this work, search summaries returned specific federal and state carryforward figures for SpaceX attached to a link list containing a CIK that does not belong to SpaceX. Those figures do not appear in SpaceX's filed 10-Q, which contains no NOL disclosure at all. This page is not reprinting them, even to rebut them with numbers, because reprinting a wrong figure is how a wrong figure survives. If you encounter a precise SpaceX NOL number, ask which filed document it came from and check the CIK.
What can honestly be said about the tax shield, and it is less than the claim. Tesla carries $7,235 million of deferred tax assets, up $310 million over the half. A deferred tax asset balance of that size is consistent with substantial loss carryforwards and credits, and it is the closest thing to filed evidence for the shield thesis — but a DTA balance is a net-of-valuation-allowance accounting figure, not a carryforward, and the two are not interchangeable. On the other side, Tesla paid a $458 million provision on $2,077 million of pre-tax income for the half, which is not the profile of a company sheltering the overwhelming majority of its income. SpaceX's DTA balance more than doubled to $354 million while the company lost $4.8 billion, and it paid $29 million of tax on those losses — almost certainly foreign and state, though the filing does not break it out. The defensible statement today is that Tesla holds a large deferred tax asset and pays real cash tax, and that SpaceX discloses no NOL carryforward at all in its interim report. Anything more specific requires the annual tax footnote, and this page will carry those figures the moment it can read them at the source.
Sources & refresh

8. What this page is built on, and when it changes

Every figure tagged Filed or Furnished above was read directly from one of the following documents. No figure on this page is sourced to an aggregator, a summary service, or a secondary compilation. The table is ordered by rung and then by weight: the two periodic reports, the registration statement and the filed 8-K body first, then the two 8-K exhibits, then the 8-K cover page read for one legal sentence only, and last the one document whose existence is cited but whose contents were not read.

DocumentCoversFiler / CIKRung
SpaceX Form 10-Q Quarter and six months ended 30 June 2026; income statement; balance sheet; full statement of cash flows; property, plant and equipment detail; IPO terms; revenue by segment; common-control mergers; concentration; backlog; spectrum; related-party debt and the Valor equipment financing; remaining performance obligations and purchase commitments; deferred tax assets. Read end to end — all 150 pages, 440,810 characters, from the complete document rather than through the truncating retrieval path described below. This is what permits the page to say “not disclosed” rather than “not retrieved” about SpaceX net operating losses, square footage and semiconductor fabrication. Space Exploration Technologies Corp. / 0001181412 Filed
Tesla Form 10-Q Quarter and six months ended 30 June 2026; income statement; balance sheet; full statement of cash flows; share count Tesla, Inc. / 0001318605 Filed
SpaceX final prospectus
Form 424(b)(4)
Read for facility-level and contract-level detail that appears in no other document: the names, roads and states of COLOSSUS and COLOSSUS II; combined compute of approximately 1.0 gigawatt; build times of 122 and 91 days; the definition of Macrohard as a software platform rather than a building; and the full terms of the Cloud Services Agreements, including the monthly fee, the term, the GPU count and the ninety-day termination right Space Exploration Technologies Corp. / 0001181412 Filed
SpaceX Form 8-K
Items 2.01 and 3.02, 14 August 2026
Completion of the Cursor (Anysphere, Inc.) merger on 14 August 2026: the share consideration of 389,289,254 Class A shares plus 1,752,426 shares for vested restricted stock units; the implied equity value of $60.0 billion and the seven-day volume-weighted average pricing mechanic; the assumed unvested awards; and the Section 4(a)(2) exemption relied on. Read in the body of the report, not an exhibit — there is no furnishing legend on it. Space Exploration Technologies Corp. / 0001181412 Filed
SpaceX Q2 2026 earnings release
Exhibit 99.1 to Form 8-K
Operating income and full cost build by segment; Segment Adjusted EBITDA; capital expenditure by segment; Starlink subscribers and ARPU; launch counts and mass to orbit; nameplate AI compute; Cloud Services Agreements and, in Note 2, the definition of contracted sales that limits them to the non-cancellable period; the Cursor acquisition; the bond issuance Space Exploration Technologies Corp. / 0001181412 Furnished
Tesla Q2 2026 update
Exhibit 99.1 to Form 8-K
Revenue and cost of revenue by line; installed annual manufacturing capacity by site; supporting infrastructure including Cortex compute; production, deliveries and storage deployed; Supercharger network; free cash flow Tesla, Inc. / 0001318605 Furnished
Tesla Form 8-K cover Read for one purpose only: the Item 2.02 language stating that the information in the exhibit shall not be deemed “filed” for Section 18 purposes. This is the document that defines the Furnished rung used throughout this page. Tesla, Inc. / 0001318605 Filed
SpaceX Q2 2026 earnings call transcript
4 August 2026
Read for one thing only, and it is the sole basis for one row of the data centre contracts table: the reference to an agreement with Google. A call is a named, attributable source and a real one, but it is not a document filed under signature, which is why every figure sourced here carries the Reported rung and none of it feeds a calculation. No filed document read here names Google as a customer; its single appearance in the 10-Q is in an unrelated litigation disclosure. Space Exploration Technologies Corp. / 0001181412 Reported
Schedule 13G/A Existence, date (17 June 2026), filer and subject only. Contents not read; no figure taken from it. Musk Elon / 0001494730 Not read

These are the primary documents themselves, not a summary of them — every Filed and Furnished figure on this page was read in the document linked above and nowhere else. The Schedule 13G/A row is the exception that proves the rule: its existence is filed, its contents were not read, and so no figure anywhere on this page derives from it.

One limitation of method, disclosed — and narrowed this revision. The retrieval tooling available to the Institute truncates very long documents at roughly 110,000 characters. A Form 10-Q exceeds that comfortably, and for most of this page's life that meant both 10-Qs were read through the financial statements and partway into the notes, but not to the end. The consequence was a rule we held to strictly: this page never asserted that a 10-Q omitted something, because a disclosure we could not locate might simply sit past where we stopped reading. Where that happened, the page said the Institute did not reach it — a claim about our reading, not about the filing.

That limitation now applies to one of the two registrants rather than both. The SpaceX Form 10-Q has been read end to end from the complete document, all 150 pages and 440,810 characters, outside the truncating path. Statements on this page that a figure is Not disclosed by SpaceX are therefore findings about the filing: the document was searched in its entirety and the term does not appear. Tesla's position is unchanged — its interim report was read partway, and its annual tax and properties disclosures have not been reached at all, which is why the tax table above says “not retrieved” in the Tesla column and “not disclosed” in the SpaceX one. Those are two different statements and this page will keep them apart. Conflating them cost us a published error once already, and the method note is here so it does not cost us a second one.

Refresh cadence

The page is called Elon Daily, so EDGAR is scanned daily. An automated sweep runs every morning against the two registrants' filing histories and against the filing calendar. It is worth being precise about what that means, because a daily scan is not a daily rewrite: most mornings the correct outcome is that nothing on this page changes, and the sweep is built so that a run which changes nothing is a successful run. What runs daily is the check. What changes is the page, and only when a new primary document exists to change it.

This page is a living reference and updates on the filing calendar, not on the news cycle. Both SpaceX and Tesla are calendar-year filers, which means quarterly reports land in the weeks after 31 March, 30 June and 30 September, and an annual report after 31 December. Between those dates the page changes only when a Form 8-K, a Schedule 13D/G amendment, or a registration statement introduces a new filed figure. A story about one of these companies that is not accompanied by a filing does not move a number on this page; it may, if it is well-sourced, add a line tagged Reported.

What would change this page most. Three events, in order of impact: a SpaceX Form 10-K, which would bring an audit opinion and full-year segment data to figures that are currently unaudited interim; a prospectus-level read of the risk factors, which would let the Customer A concentration be attributed rather than merely measured; and any registration event at Neuralink or Boring, which would take this portfolio from two-thirds observable to fully observable for the first time.

Related Institute references

Editorial positioning — explicit reframe

This is a provenance-tagged reference page, not a valuation call. It records what six Musk-controlled businesses have disclosed, and what rung of disclosure each figure sits on. It is not investment advice, is not a recommendation to buy or sell any security, is not tax advice, and is not a forecast. Where a figure is absent, the page leaves the cell empty rather than filling it from an estimate. Where the Institute did not read far enough into a document to locate a disclosure, the page says so in those words rather than asserting the disclosure is absent. “Not in the part of the document the Institute read” and “not disclosed” are different claims, and only the first is ours to make.

Educational references and tools — not legal, tax, accounting, or investment advice, and not a recommendation to buy or sell any security. Consult a qualified professional about your specific situation. © 2026 The Baratelli Institute.