Space Exploration Technologies Corp. SPCX

Key data

The business

At its core this is still SpaceX: Falcon and Starship put mass into orbit cheaper than anyone alive, and Starlink sells the internet back down from that orbit, a dish on a roof and a subscription around $120 a month that renews every month whether a rocket flies or not. That recurring layer is the cash engine. In 2026 the company swallowed xAI, the Grok chatbot, and paid $60 billion in stock for Cursor, the coding tool, welding an AI segment onto the hardware. It now reports three: Connectivity (Starlink, and all of the product revenue), Space (launch and government work), and AI. Services are now 94% of sales. Two customers matter: Customer A, spread across all three segments and almost certainly the US government, at 18.3%, and Customer B, brand new, sitting entirely in AI, at 19.5%. The moat is real on the hardware, a reusable-launch near-monopoly and a first-mover satellite network with spectrum nobody else holds, and absent on the AI, which was bought, not built into a lead.

The numbers

The quarter is the whole story in one line: revenue up 92% year over year while the operating loss collapsed from nearly a billion to $143 million. Back that out and incremental operating margin ran 22%, the operating leverage the case for owning it is built on.

Q1's -$1.19 print carried a heavy merger charge and its net loss is not disclosed in this filing; the inflection is Q2, when the AI customer landed. The half-year shows where the growth actually sits, and it is not the rockets.

Q2's $0.09 loss on a $541 million net loss backs out to about 6.0 billion diluted shares, still half the roughly 13.1 billion the $1.9 trillion market cap counts, and that was before Cursor's 389 million new shares landed in August. The balance sheet is the other half of the bet, and it moved violently in six months.

Here is the compounding math that governs everything. First-half revenue was $12.5 billion. The full-year 2026 consensus is $43.3 billion, which means the second half has to nearly triple the first to about $31 billion, and 2027 has to double that again to $104 billion. That entire ramp is the AI segment consolidating, and almost all of it rests on Customer B, one buyer, first seen this quarter. The variant is simple: the market is pricing the AI as if it already works, and the print that settles it is Q3, the first full quarter with xAI and Cursor inside the numbers, showing whether AI revenue broadens past a single 19.5% customer or leans harder on it.

Management

You are underwriting Musk's judgment, not a board's. Class B shares carry ten votes each, so control is absolute, and the capital allocation reads as empire-building funded by high-priced stock: $60 billion for Cursor, plus share repurchases handed to current and former xAI employees as liquidity rather than value. Musk sold $1.2 million of stock in April, plan status not disclosed, a rounding error against his holding and the only insider transaction on the tape; there were no buys. The record to judge is the mergers, and they were priced for a future that has to arrive.

How it fails or surprises you

The AI ramp does not show up. The 91x forward multiple assumes revenue climbs from a $25 billion annual run-rate toward $104 billion by 2027. If second-half 2026 fails to triple the first half, there is nothing under the price. The tell arrives fast: Q3 total revenue and any standalone AI-segment disclosure.

The balance sheet tightens under three capital sinks at once. Long-term debt doubled to $38.3 billion in six months while launch and Starlink stay capital-hungry and the AI bet needs years of funding. The $100 billion war chest looks bottomless until three moonshots draw on it together. Watch free cash flow and the debt line next quarter.

Starlink and launch alone are the surprise (right tail). Under the AI noise sits a recurring services base that is 94% of revenue, throwing 22% incremental margin, with deferred revenue at $14.3 billion and contracts booked out to 2029. If direct-to-cell and launch cadence scale, the hardware justifies much of the price by itself and the AI comes as a free option the market is currently paying for anyway.

Closing thoughts

Q3 2026 is the print that settles it. That quarter consolidates the full AI roll-up for the first time and shows whether Customer B's 19.5% share was the start of broad adoption or a single buyer carrying the whole bet. Second-half revenue has to triple the first half to reach the $43.3 billion full-year target, and 2027 has to double that again. If Q3 shows AI revenue spreading past that single customer and total revenue tracking toward $12 billion, the 91x forward multiple starts to make sense and the right tail stays live. If the AI segment still leans on Customer B alone and revenue falls short, most of the $1.9 trillion above the hardware value is a premium on hope, and the downside is real.

The bet is still that one man runs three moonshots in one company, and that the AI he bolted on earns real money fast. What breaks it is second-half revenue failing to ramp toward the full-year number while the AI line still leans on a lone 19.5% customer. If Q3 revenue does not clear roughly $12 billion with the AI segment broadening, the ramp is a hope, not a business.