SpaceX's first earnings report as a public company landed the same day it announced it's putting Nvidia's newest GPUs into orbit.
On August 4, 2026, SpaceX reported its first quarterly results since its June 2026 IPO: $7.8 billion in Q2 revenue, up 92% year-over-year. Hours later, SpaceX and Nvidia announced they're co-designing the compute payload for Starmind, SpaceX's planned orbital data-center satellite network — each satellite carrying Nvidia Rubin GPUs and Vera CPUs. Two announcements, one story: SpaceX is now simultaneously a launch company, a telecom company, and — increasingly literally — a data-center operator.
TL;DR
| Question | Direct answer |
|---|---|
| Q2 revenue? | $7.8B, up 92% YoY, ~$1B ahead of Wall Street estimates |
| Net loss? | $541M, narrowed from $1B a year earlier |
| Starlink subscribers? | 12 million, doubled year-over-year |
| ARPU trend? | $66, flat QoQ but down from $85 a year ago |
| AI segment revenue? | $2.56B, up 247% YoY (xAI, X, cloud services) |
| Backlog? | $47.5B |
| Nvidia deal? | Co-designing Starmind satellite compute payload: Rubin GPUs + Vera CPUs |
| Stock reaction? | Fell 8%+ intraday on capex concerns despite the earnings beat |
The earnings, unpacked
This was SpaceX's first earnings report as a public company, following its June 2026 IPO — meaning it's also the first time investors have gotten a full quarterly breakdown rather than relying on leaked figures or S-1 filings. The headline number cleared expectations by a wide margin: $7.8 billion in revenue against Street estimates roughly $1 billion lower, a 92% year-over-year jump.
Starlink drove 55% of sales, and subscriber count reached 12 million, double the prior year. But the more interesting number sits underneath that growth: average revenue per user held flat at $66 this quarter and is down sharply from $85 a year ago — the arithmetic of Starlink's expansion into lower-price-point international markets and its Mobile service, trading ARPU for volume.
The AI segment — which SpaceX groups as xAI, the X platform, and cloud services — is the fastest-growing line: $2.56 billion, up 247% year-over-year. TechCrunch reported that Anthropic and Google compute deals were named contributors to that growth — a notable detail given Anthropic's existing $1.25B/month xAI compute arrangement covered earlier this year.
Net loss narrowed meaningfully, from $1 billion to $541 million year-over-year, and the order backlog reached $47.5 billion — a forward-demand signal that matters more than any single quarter's revenue for a capital-intensive business like this one.
Why the stock fell on a beat
Despite clearing estimates on every major line, SPCX fell more than 8% intraday. The read from coverage is straightforward: capital expenditures spiked sharply as SpaceX scales Starlink, Starship, and now Starmind simultaneously — three capital-intensive programs running in parallel, each competing for the same balance sheet. Beating a revenue estimate doesn't offset investor concern about a materially larger capex trajectory, particularly for a company that only went public two months earlier and is still establishing its public-market financial discipline story.
Notably, Nvidia's own stock gained on the same day, on the Starmind partnership news specifically — a reminder that this is as much an Nvidia hardware-demand story as a SpaceX earnings story.
Starmind AI1: what's actually being built
The Nvidia partnership targets Starmind, SpaceX's proposed network of orbital data-center satellites first detailed via an FCC filing and the original AI1 satellite unveiling earlier in 2026. The new detail is the compute silicon: each Starmind satellite is planned to carry NVIDIA Rubin GPUs and Vera CPUs — the same architecture family Nvidia sells for terrestrial data centers, adapted for orbital deployment and solar power.
Update — August 5, 2026: NVIDIA confirmed the exact configuration directly: the Starmind AI1 compute payload runs on NVIDIA Vera Rubin NVL72 — the same rack-scale, liquid-cooled Grace-Blackwell-successor system NVIDIA sells as its flagship terrestrial AI factory building block, repackaged for orbital deployment. NVIDIA's own framing leaned hard into the space-marketing angle ("the next chapter of AI infrastructure boldly goes where no AI compute has gone before"), while the top reply on the announcement — "You simply cannot keep building data centers on Earth. Orbital compute is the future." — captured the split reaction: genuine enthusiasm for the infrastructure bet alongside skepticism about whether "NVL72 in orbit" is actually a near-term deployable configuration versus a marketing rendering of the 2027 prototype target.
Musk added a detail that extends this beyond a pure space story: the same Starmind V1 satellite design, minus the solar array and radiator, is also being deployed on the ground in SpaceX's own data centers, which he described as "a major improvement in data center efficiency." That's a notable reversal of the usual direction of technology transfer — normally space hardware inherits terrestrial data-center design, not the other way around. Here, a satellite compute design is informing SpaceX's ground infrastructure too.
The development timeline is early-stage: prototype testing in early 2027, with mass production targeted for later that year if development stays on schedule. This is a multi-year hardware program, not a near-term operational capability.
The interchangeable-chip question
SpaceX's original AI1 filing described the satellite's compute payload as interchangeable — designed so SpaceX wouldn't be locked to a single chipmaker across the life of the constellation. The Nvidia partnership confirms Rubin GPUs and Vera CPUs as the payload for the AI1 generation specifically, not necessarily a permanent single-vendor commitment for every future Starmind satellite. That distinction matters for reading this announcement correctly: it's a concrete hardware decision for the near-term prototype and production run, not evidence that SpaceX has abandoned the multi-vendor flexibility it originally described. Whether later Starmind generations stay on Nvidia silicon or diversify will depend on how the Rubin/Vera-equipped AI1 prototypes actually perform once they reach orbit in 2027.
What people are asking
Is this the same AI1 satellite from the January FCC filing? Yes — Starmind AI1 is the specific satellite design SpaceX detailed earlier in 2026 (see explainx.ai's full AI1 technical breakdown). The Nvidia partnership is new: it specifies the actual compute silicon going into that satellite design, which wasn't confirmed at the time of the original filing.
Does SpaceX's AI revenue growth depend on xAI's own fundraising? Partly — the AI segment groups xAI, X, and cloud services together, so its growth reflects both xAI's own compute demand and third-party deals like the Anthropic and Google arrangements TechCrunch reported. It's not purely organic Starlink-adjacent growth; it's tied to the broader AI compute market SpaceX is now selling into.
Why would anyone put a data center in orbit instead of just building more on the ground? The pitch — consistent with SpaceX's original AI1 filing — is continuous solar power without a terrestrial grid connection or land footprint, at the cost of much harder cooling, maintenance, and connectivity engineering. The ground deployment of the same V1 design suggests SpaceX sees efficiency gains in the underlying satellite compute architecture even independent of the orbital use case.
Should Nvidia investors care more than SpaceX investors here? Both, but for different reasons — Nvidia gets a new, high-margin data-center-silicon customer at a moment when hyperscaler capex growth is a constant market question; SpaceX gets a compute program that's still years from revenue, layered on top of Starlink and Starship capex that already worried the market this quarter.
The pattern across Musk's companies
This announcement rhythm — earnings and a major infrastructure partnership landing the same day — isn't new for Musk-run companies, but it's notable here because it ties SpaceX's public-market debut directly to the AI compute narrative rather than treating Starlink and Starship as the whole investment story. SpaceX going public was, in large part, a bet that public investors would fund the capital intensity of the AI compute buildout alongside the traditional launch and satellite-internet business — and pairing the first earnings call with a marquee Nvidia partnership is a direct signal to that audience about where growth capital is headed next.
It also puts SpaceX in a small group of companies straddling both sides of the AI infrastructure market: a buyer of compute (via xAI's own model training) and, if Starmind ships, a seller of compute capacity to other labs — the same dual role Anthropic's existing xAI rental arrangement already hints at. Whether that becomes a genuinely differentiated business line or stays subordinate to Starlink's core economics is the multi-year question this quarter's numbers can't yet answer.
The takeaway
SpaceX's first earnings report as a public company delivered exactly what a growth story needs — revenue beating estimates by a wide margin, subscriber growth, and a fast-growing AI segment — and the market still marked it down on capex concerns. That combination is worth sitting with: a beat-and-raise quarter that still trades down is usually a signal the market is pricing in the next several quarters of spending, not just the one just reported, and SpaceX's simultaneous Starlink, Starship, and Starmind buildout gives investors plenty of forward capex to price in. The Nvidia Starmind partnership announced the same day is the more speculative, longer-horizon story: a 2027 prototype timeline riding alongside a business already spending aggressively on Starlink and Starship. Whether "data center in orbit" becomes a real revenue line or stays a hardware R&D program is a multi-year question; what's confirmed today is that Nvidia's Rubin and Vera silicon is now explicitly part of SpaceX's compute roadmap, on the ground and off it.
Related on explainx.ai:
- SpaceX's Solar-Powered Orbital AI Datacenters: The AI1 Satellite
- Elon Musk, Starmind, and xAI: Bio and Meaning
- NVIDIA Alpamayo 2 Super: Open Reasoning Model for Robotaxis
- NVIDIA Cosmos 3: Open Physical AI World Models
- Elon Musk on AI Compute: Space vs. Local Tokens
Official: SpaceX × Nvidia Starmind announcement · SpaceX Q2 2026 earnings coverage — TechCrunch
Financial figures reflect SpaceX's Q2 2026 earnings release on August 4, 2026; Starmind AI1 timelines are SpaceX's stated development targets and may shift.