SpaceX's Q2 2026 Earnings and Exclusive Nvidia Starmind AI Partnership

SpaceX and Nvidia are building the Starmind AI1 satellite's compute payload around a simple physical advantage: satellites in sun-synchronous orbit receive constant, uninterrupted solar power, without depending on Earth's electrical grid or fighting for scarce power-plant capacity ^[1]. Nvidia's contribution is a Space-1 Vera Rubin module rated at up to 25 times the AI processing performance of an H100 GPU, with commercial shipments arriving later in 2026 ^[2]. The catch is regulatory, not technical: SpaceX has asked the FCC for approval to fly up to one million of these orbital data-center satellites at 500-2,000 km altitudes, linked by petabit-scale lasers, and that filing has not yet been ruled on ^[2]. One outlet's framing is blunt - this is currently a chip order without a license ^[3]. Musk's own targets are aggressive regardless of that gap: SpaceX is aiming for 2 gigawatts of AI compute online by the end of 2026, scaling toward 10 gigawatts by the end of 2027 ^[4]. That timeline assumes both the hardware and the regulator cooperate - and only one of those two is fully within SpaceX's control.

AMD's Q2 2026 numbers were objectively strong: revenue of $11.5 billion, up 50% year-over-year, with non-GAAP earnings per share of $1.66 beating the roughly $1.61 consensus ^[5]. Its shares fell 6-9% anyway ^[5][6], because the same 24 hours carried Musk's on-air declaration that 'going forward, we have decided to build exclusively on Nvidia, because we think the Blackwell architecture is the best architecture' ^[6]. AMD had also roughly doubled its own capital spending, to $808 million from $282 million a year earlier, positioning for exactly the kind of hyperscaler and space-compute business the Starmind announcement now signals it will not get ^[6]. Nvidia, meanwhile, rose 3-4% on the same news ^[7][8]- a reminder that in AI-infrastructure headlines, markets increasingly price the announcement of a customer relationship as heavily as either company's actual quarterly results.

Musk's 'we're exclusive to Nvidia' line reads less like a cold procurement decision and more like the culmination of a relationship that predates SpaceX's IPO by a decade. Musk was among the team that received Nvidia's first DGX-1 supercomputer directly from Jensen Huang in 2016, and in October 2025 Huang personally delivered a DGX Spark 'petaflop' system to Musk at SpaceX's Starbase facility ^[9]. The financial ties run deeper than handshakes: Nvidia took an equity stake in xAI as part of a $20 billion funding round in January 2026, and those shares later converted into SpaceX stock once xAI was absorbed into the company - meaning Nvidia was already a SpaceX shareholder before Starmind was announced ^[10]. Musk's on-call framing - 'we think the Vera Rubin architecture is the best architecture...we greatly value our close cooperation and partnership on many levels with Nvidia' ^[4]- is consistent with that history. The exclusivity announcement reads as a hardware relationship formalizing a financial one that already existed.

The number that spooked investors wasn't revenue - it was capital expenditure. SpaceX's Q2 2026 capex came in at roughly $18.4 billion, about 39% above the $13.22 billion analysts had modeled ^[11]. S&P Global's Melissa Otto put the reaction plainly: 'The stock is down because the capex for the AI segment was more than double what was expected. It's ambitious and remains a show-me story' ^[12]. That skepticism has a track record behind it - SpaceX's AI unit (the former xAI) reportedly lost $6.4 billion on just $3.2 billion of revenue the prior year ^[3]. Layered on top is the pending $60 billion all-stock acquisition of Cursor, whose roughly $4 billion in annualized recurring revenue is already baked into SpaceX's forward AI-revenue guidance before the deal has even closed ^[13]. Investor forums went further than Otto's cautious framing, describing a circular-financing concern in which capital flowing between Nvidia's AI-lab investments and the compute those labs rent from SpaceX effectively funds the same GPU purchases Nvidia is selling - a dynamic worth watching as SpaceX's capex keeps compounding faster than its disclosed AI revenue.

The loudest reactions to Starmind split into three distinct readings. On X, SpaceX's and Nvidia's own announcement posts framed the news as a straightforward infrastructure milestone - AI compute 'going to orbit' - and drew overwhelmingly positive, excited replies treating it as a natural extension of the Starlink playbook. Independent analyst commentary took a more corporate-strategy view, tying Starmind to SpaceX's broader move toward Wall Street legitimacy - a newly public company simultaneously pursuing a credit rating upgrade and index inclusion - and arguing that AI compute, not launch services, is now the more consequential growth line to watch. Investor and engineering communities elsewhere were considerably more skeptical, running two parallel threads: a financial one questioning whether the Nvidia-SpaceX-AI-lab capital relationships amount to vendor financing rather than organic demand, and a technical one debating whether GPUs can actually survive un-serviced for five to ten years in orbit, whether radiators can be sized to shed enough heat, and whether orbital latency limits the hardware to inference rather than the training workloads the compute buildout is pitched around. None of the three camps is obviously wrong: the underlying regulatory picture still has SpaceX's satellite-constellation filing awaiting an FCC ruling ^[3], which means the celebratory, the strategic, and the skeptical readings are, for now, all bets on execution that hasn't happened yet.