The AI promise, meet the ledger
There is a moment in every AI earnings season when the keynote slides come down and the income statement goes up. This week was that moment for the AI trade — and the two documents are not always on speaking terms.
Consider the cast. One company reported that its artificial-intelligence business now brings in more revenue than its rockets and its satellite network put together. Another reported flat revenue, falling car deliveries and a capital-expenditure bill that has more than doubled — and asked investors to keep believing in robotaxis, humanoid robots and a merger that does not exist yet. A third reported $64.6 billion in revenue, a $9.89 billion legal charge, and a brand-new class of shares invented to keep the AI spending spree funded. Earnings season did not declare the AI promise a fraud or a triumph. It did something subtler: it started auditing it, line by line.
And the audit has an uncomfortable question at its centre — one Meta’s and Tesla’s figures do not break out.
SpaceX: the receipt exists
Start with the company that has a receipt. SpaceX, in only its second quarterly report since its June IPO, posted total revenue of $14.6 billion, up 161% year on year. Inside that number sat something striking: its Super Intelligence segment brought in $7.6 billion — up 830% year on year, and now larger by revenue than the company’s Connectivity ($5.2 billion) and Space ($1.8 billion) segments combined. The SI growth was driven by new cloud-services agreements, the closed $60 billion acquisition of Cursor and enterprise and consumer adoption of Grok.
One caveat: revenue is not profit. SI income from operations was $431 million, well behind Connectivity’s $2.4 billion, and the Space segment ran a $7 million operating loss. A company famous for launching rockets now collects more revenue from selling intelligence than from launches and bandwidth combined — in a quarter where its Starlink business counted 13.6 million consumer subscribers.
SpaceX plans its first-ever Starship + Super Heavy booster catch attempt next month — “Success is not guaranteed on these first flights,” Musk said on the earnings call — and Gwynne Shotwell described plans for a terrestrial mobile network that would compete with T-Mobile, AT&T and Verizon. And both SpaceX and Tesla have formed special committees to discuss a merger, with Musk excluded as a conflicted party, no terms agreed and no guarantee of a deal. Analysts mostly expect SpaceX would absorb Tesla, if anything happens at all. If.
Tesla: the receipt is a promise
Now the other Musk company. Tesla reported net income of $1.69 billion on $28.0 billion of revenue, against $1.77 billion on $28.1 billion a year earlier — flat revenue, a profit drop of roughly 4.5%, and a miss against the $28.3 billion consensus. Deliveries fell 2.1% year on year. Automotive revenue was $20.5 billion, down 3%. And while the top line sat still, the spending did not: capital expenditure jumped 110% year on year to $4.7 billion, operating expenses rose 39%, and free cash flow came in at $1.5 billion after dipping negative the prior quarter.
On the earnings call, the AI story was told almost entirely in the future tense. Vehicle owners may be able to add their own cars to the robotaxi fleet from the second quarter of 2027, said VP Ashok Elluswamy — though Musk has been promising a Tesla ride-hailing network since 2019. Optimus is in early production on the Fremont line, but Musk himself supplied the caveat: “There’s still quite a lot of work to do with regard to the AI and the robot for it to be truly useful for open-ended tasks.” Volume deliveries to outside companies might come by the end of 2027. Might.
This is not to say Tesla is doomed or SpaceX is vindicated. It is to say something plainer: Tesla’s AI narrative currently rests on robotaxis, robots and merger discussions — announced products and future plans — while its quarter’s numbers rest on a car business that shrank. That is a gap between a keynote and an income statement, and the income statement is the one auditors sign.
Meta: inventing money to fund the future
Meta’s quarter was the strangest of the lot, and not because of the revenue. The company reported $64.60 billion in revenue — genuinely strong — alongside a $9.89 billion quarterly charge for the child-safety settlement with US attorneys general, part of a $17.1 billion settlement agreed in August. Capex guidance was raised to $135–145 billion. And then came the financial engineering: a new class of non-voting shares, trading from 2027, announced explicitly as a funding mechanism for AI capex, acquisitions and employee compensation. CFO Susan Li framed it as wanting to “expand our funding tools and give us flexibility to invest behind the unprecedented momentum we’re seeing across every part of our business driven by AI.”
Meta says it is a funding mechanism for AI capex, acquisitions and employee compensation. Whether that amounts to inventing a new kind of money to fund an endless capex treadmill — and what say current shareholders retain over the spending it enables — are questions the announcement does not answer. Whether that is bold capital allocation or an admission that the capex treadmill never stops depends on your priors — but it is, at minimum, a number with a story behind it.
Meta did have one AI-linked figure to show: its AI agent Muse, launched in September, has hit 10 million US downloads, Meta takes a fee on commercial transactions, and Zuckerberg says it will be “highly profitable” in a few years. The research supplies no current profit figures for Muse — only Zuckerberg’s expectation that it becomes “highly profitable” in a few years. The quarter also showed the cost of moving fast: one incident in which an agent shared a user’s address with a stranger on Marketplace, and a paused conversation-sharing feature after private chats started appearing publicly on the web.
The audit’s missing line item
Here is where the audit gets awkward — for Meta and Tesla in particular. The thesis of this season is that the numbers separate companies posting AI-linked revenue (SpaceX’s SI segment) from companies whose AI story still rests on announcements (Tesla’s robotaxis, Optimus, merger talks). But that split, honestly drawn, is between disclosed results and announced plans — not between measured AI revenue and promises. Neither Meta nor Tesla breaks out how much of its revenue is actually attributable to AI activities. The most important line item in the AI economy — how much of their revenue is actually AI-linked — remains the one their figures do not report.
And the market, for its part, is auditing the auditors. Semiconductor stocks fell 15–30% this month even as analysts’ earnings forecasts for next year rose — Micron’s forecast went up 14% while its shares fell 17%. The shares fell not because the companies will earn less, but because investors will pay less for the same earnings. Microsoft’s quarter fed the same unease: revenue of $61.9 billion, up 6.1%, but the operating margin contracted to 35.2% against 37.9% expected, blamed on higher AI cloud costs, and price targets for Adobe, Salesforce and Snowflake were cut by roughly 12%. The Nasdaq Composite slipped 1.4% on the day.
The fair counterargument: this season arguably validates the AI spend rather than indicting it. SpaceX’s SI grew 830%. Meta grew revenue to $64.6 billion with real underlying strength beneath a one-off settlement charge. Tesla’s flat quarter may be a transition while capex converts into robotaxi and robot revenue. Markets punishing laggards and rewarding leaders is the system working, not a verdict on the promise.
The AI earnings season ledger is in
Which leaves the season’s honest verdict: the gap between the keynote slides and the income statement is now measurable, and it runs in both directions. Some companies are posting the revenue. Some are posting the spending and asking you to wait for the revenue. And some are inventing new shares — announced as a funding mechanism, with the question of when the spending stops unanswered.
Earnings seasons are recurring audits, and the next one will ask the question this one could not answer: what share of any of this revenue is actually AI — and will anyone ever break it out? Until then, read the income statements, enjoy the keynotes, and mind the gap between them.
