GOVERNANCE WATCH
By InsidEntity Editorial Desk · Aug 6, 2026 · 11 min read
For several quarters this platform has tracked a recurring pattern in AI-era earnings: some of the largest reported profits in the market have rested substantially on fair value gains on private holdings rather than on operating performance. Tesla’s second quarter carried a $1.005 billion mark on its SpaceX stake. Alphabet’s carried roughly $99 billion in gains on equity securities, led by Anthropic. Amazon’s carried $53.4 billion, disclosed as arising primarily from Anthropic.
On 4 August 2026, one of those marks became a market. SpaceX reported its first quarterly results as a public company following its June listing. There is now a share price, an income statement, a segment table, and a Company Risk Rating where previously there was an estimate. That rating is 2.59, Caution, the lowest of any listed company this series has examined.
The numbers are more interesting than the headline suggests, and not in the direction the headline points.
The headline beat
SpaceX reported second quarter revenue of $7.814 billion, up 92% year on year against a consensus near $6.81 billion. All three segments beat. Net loss narrowed to $541 million from $1.008 billion. Adjusted EBITDA nearly tripled to $3.538 billion. Management holds close to $100 billion in cash, equivalents and marketable securities, and pointed analysts toward an internal ambition of $1 trillion in annual revenue by 2030.
The market was less enthusiastic. The stock fell more than 8% after hours before recovering part of the move.
The reason sits in three places: the segment table, the capital expenditure line, and the identity of the customers.
The segment arithmetic
The most revealing disclosure was not the consolidated income statement. It was the breakdown by division.
| Segment | Revenue | Growth | Operating result |
|---|---|---|---|
| Connectivity (Starlink) | $4.291bn | +66% | +$1.656bn |
| AI | $2.561bn | +247% | -$1.257bn |
| Space | $962m | +29% | -$542m |
| Group | $7.814bn | +92% | -$143m |
Take Starlink’s operating profit, subtract the two loss-making divisions, and you arrive at the group’s consolidated operating loss of $143 million.
That is the whole company in one line. SpaceX is not three businesses growing together. It is one profitable business funding two that are not, with the profitable one currently a little short of the task. Starlink alone accounts for roughly 55% of group revenue and is the only division turning an operating profit.
Cross-subsidising a capital-intensive build from a cash-generative division is a legitimate strategy and it is how a great deal of infrastructure has been financed. But it is a materially different company from the one implied by a 92% growth headline, and durability lives in the concentration rather than the aggregate.
Who is paying for the beat
The AI segment grew 247% on the strength of new cloud services agreements worth $14.1 billion. The customers behind that growth are Anthropic and Google.
Set that against what this series has already documented. Alphabet holds a stake in Anthropic and marked it upward, contributing that same $99 billion of gains on equity securities to a quarter in which its operating business earned $40.8 billion. Amazon holds a stake in Anthropic and marked it upward, contributing $53.4 billion of non-operating pre-tax income to a quarter in which its operating business earned $27.5 billion. Both supply Anthropic with compute: Amazon through AWS, Bedrock and Trainium, Alphabet through Google Cloud and TPUs. Tesla holds a stake in SpaceX and marked it upward by $1.005 billion in a quarter whose core result was otherwise close to breakeven.
Now add this filing. SpaceX’s fastest-growing segment is paid, in significant part, by Anthropic and by Google.
So the compute revenue that drove SpaceX’s beat comes from a company whose valuation produces the marks in Alphabet’s and Amazon’s earnings, and from Alphabet itself. Meanwhile Tesla’s earnings carry a mark on SpaceX. Four public companies, one private one, and a set of relationships in which the same dollars appear as revenue in one set of accounts, as capital expenditure in another, and as a fair value gain in a third.
None of this is improper and none of it is concealed. Every element was disclosed by the company reporting it. Compute contracts between technology companies are ordinary commerce, and an equity stake in a fast-growing private company is an ordinary investment. What is not ordinary is the density of the loop, or how little of it is visible from any single company’s earnings release. The circularity this series flagged individually at AMD, where a customer holds a warrant over roughly 10% of the supplier it buys from, and at Alphabet and Amazon, where suppliers mark the customer they supply, is not a set of isolated arrangements. It is a structure.
The EBITDA question
The AI segment reported an operating loss of $1.257 billion and positive adjusted EBITDA. The gap between those figures is largely depreciation on compute infrastructure.
This is where earnings quality becomes the whole question. EBITDA excludes depreciation. For a business whose defining economic characteristic is an enormous and accelerating capital build, depreciation is not a distortion to be adjusted away. It is the cost.
Management effectively conceded the point, though probably not intentionally. The chief financial officer told analysts that AI compute capital is achieving a payback period of under a year, and is becoming something close to a cost of goods sold item because it moves so quickly.
Read that carefully. If AI capital expenditure genuinely behaves like cost of goods sold, then adjusted EBITDA for the AI segment is not a profit measure at all, because it adds back a cost the chief financial officer describes as functionally equivalent to COGS. The honest version of the argument is that the segment is loss-making today and management believes the assets will earn out quickly. That may well prove right. It is a forecast, not a result.
The capital expenditure line
Total capital expenditure was $18.369 billion for the quarter against an average estimate of $13.22 billion, with $15.828 billion of it in AI, up from $749 million a year earlier. That quarterly run rate annualises to about $73.5 billion against a full-year consensus near $48.7 billion. Compute capacity reached 1.4 gigawatts from 0.4 a year ago. Management indicated the next two quarters will run at similar levels.
The balance sheet absorbs this for now, largely because of the listing. What the IPO has bought is runway. What it has not answered is what the steady state looks like once the proceeds are deployed.
One operating detail deserves flagging. Starlink passed 12 million subscribers, roughly double a year earlier. Average revenue per user was $66 against $85 a year ago, a decline of 22%. Subscriber doubling with a 22% ARPU decline produces the 66% revenue growth reported. The growth is real, and the mix is shifting toward lower-value subscribers, which matters in a division being asked to fund two loss-making ones. Enterprise and government revenue, which more than doubled to $1.806 billion, is the offsetting line.
The governance layer the segment table does not show you
Inspection produces more than an income statement. It produces a governance structure, and SpaceX’s is the most concentrated this series has examined.
| Component | Score | Read |
|---|---|---|
| Director Independence | 5.0 | Benchmark, by the platform’s default pending completed questionnaires |
| Shareholder Influence | 3.0 | Good, and the first time this pillar has scored below Benchmark in this series |
| Director Capacity | 2.38 | Caution |
| Auditor Independence | 0.0 | Nil, tenure past the point the framework credits |
| Overall Rating | 2.59 | Caution |
(1 Risk, 2 Caution, 3 Good, 4 Excellent, 5 Benchmark. The overall is the average of the four pillars: (5.0 + 3.0 + 2.38 + 0.0) ÷ 4 = 2.595, displayed as 2.59. Ratings update regularly; check the live report for the current view.)
Two things about that table are worth more than the headline number.
The first is Shareholder Influence at 3.0. Across every company this series has profiled, from Nike to Alphabet to the six largest South African banks, this pillar has scored 5.0, because it tests economic ownership against a 20% materiality threshold and no holder has crossed it. Musk holds roughly 42% of SpaceX’s economics, more than double the line, and the pillar registers it.
Set that against Tesla. Tesla scores 5.0 on the same pillar, because Musk’s Tesla stake sits in the mid-teens, below the threshold, while his control there runs through a compensation structure and the influence of the founder rather than through a majority economic holding. This series noted at the time that the framework’s pillars did not capture Tesla’s related-party surface, and that the exposure lived in the record rather than the ratio.
SpaceX shows the other half of that observation. The framework is not blind to concentration. It measures economic concentration, and when a founder’s holding is large enough, it says so. Tesla’s structure routes control around that measure. SpaceX’s does not.
| Tesla | SpaceX | |
|---|---|---|
| Director Independence | 5.0 | 5.0 |
| Shareholder Influence | 5.0 | 3.0 |
| Director Capacity | 2.81 | 2.38 |
| Auditor Independence | 0.0 | 0.0 |
| Overall | 3.20 | 2.59 |
Same chairman. Same nil auditor pillar. A 0.61 point gap, driven by a shareholder pillar that reads one structure and not the other, and a board with less capacity to spare.
The second is what the 5.0 on Director Independence is standing on.
Elon Musk held 85.1% of shareholder voting power at the time of the listing filings, settling at 82.4% after the June IPO, delivered through Class B shares carrying ten votes each on that same 42% of the economics. The filings also disclose that he can be removed as chairman, chief executive or chief technology officer only by a vote of the Class B shareholders, a class he dominates, which in practice makes his removal a decision only he can take.
As a controlled company under exchange rules, SpaceX is exempt from requirements for independent board oversight that apply to most listed companies. That exemption sits directly underneath a Director Independence pillar scoring Benchmark by default, which is the starkest version of that default this series has encountered: a company legally excused from independent board oversight, scored 5.0 on independence because the questionnaires establishing it have not been completed. Shareholder proposals require a 3% holding, a threshold worth billions that only a handful of institutions clear. The offering documents include jury trial waivers and mandatory arbitration provisions. Three of America’s largest public pension funds, together managing more than a trillion dollars, wrote to the company objecting to the structure and calling for one share one vote, a majority-independent board, and independent approval of related-party transactions.
That last request is the one this quarter makes concrete. Tesla, chaired by the same person who controls SpaceX, holds SpaceX equity and marked it upward this quarter. Musk has publicly floated a possible Tesla and SpaceX combination. Any such transaction would run through boards negotiating with their own chief executive’s controlling interest on the other side of the table. The pension funds asked for independent approval of related-party transactions before any of that was on the record. It is now on the record.
The full map of these interlocks is exactly what a director-level record exists to hold. Elon Musk’s InsidEntity profile connects the SpaceX control position, the Tesla seat and the rest of the portfolio in one place.
Why this matters beyond SpaceX
When a private company is held at a mark, the holder’s earnings depend on a valuation judgment. When it lists, the judgment is replaced by a price, and the price is subject to disclosure, quarterly scrutiny and lockup expiries.
SpaceX’s first filing shows what inspection produces: a strong revenue beat, a consolidated operating loss, one division carrying two, an EBITDA presentation that adds back the central cost, a capital programme running well ahead of forecast, a customer base that overlaps with the companies marking the same asset class upward, and a control structure under which the chief executive cannot be removed by anyone but himself.
None of that was visible from the outside. All of it was there.
That is not an argument that SpaceX is a poor business. Starlink’s subscriber trajectory and operating profit are genuinely impressive and the payback thesis on compute may prove correct. It is an argument that a reported profit resting on the revaluation of an unlisted holding is a claim about a company nobody can inspect, and that such earnings deserve a discount until the underlying asset can be examined directly.
What to watch
Four items. Whether capital expenditure converts, since management’s case rests on a sub-year payback and the next two quarters at similar spending will begin to test it. Whether Starlink’s ARPU stabilises, because the funding division’s unit economics matter more than its subscriber count. Whether the AI segment’s customer concentration is disclosed in more detail, since the identity and share of those counterparties is now material to the group’s growth rate. And the share register after lockup: a very large block of pre-IPO shares became eligible for sale on the release of these results, and register composition following that expiry is a governance signal, not merely a supply one.
SpaceX’s Company Risk Rating of 2.59, with the pillar breakdown and board record behind it, is on InsidEntity. For a company that spent two decades as an estimate on someone else’s balance sheet, the first filing is a useful correction. The numbers are neither as good as the headline nor as bad as the after-hours reaction. They are simply, for the first time, checkable.
Know your entity. This is independent research, not financial advice.
Better entity intelligence, more risk mitigation | #KnowYourEntity