SpaceX Reveals $13 Billion ARR Boost From New Hosting Deal, Targets Orbital Compute Satellites Launch Next Year
Goldman Sachs Communacopia + Technology Conference, September 10, 2026
SpaceX CFO Bret Johnsen used a fireside chat at Goldman Sachs' Communacopia + Technology Conference to disclose fresh detail on the company's AI infrastructure ambitions, confirming that a compute hosting deal closed earlier this month adds roughly $13 billion in incremental annualized revenue starting December 1. The disclosure strengthens the company's already-stated goal of reaching $100 billion in ARR by year-end 2026, a target Johnsen said he now has "even more conviction" in hitting. The new contract translates to $1.11 billion a month once it kicks in, underscoring how quickly SpaceX's compute business has scaled from what was, only a couple of quarters ago, a much smaller piece of the balance sheet.
Orbital Compute Timeline Pulled Forward
The most consequential new information from the session centers on orbital compute, which Johnsen argued the market continues to underestimate. "Most, if not everybody, other than us, think it's many, many years away," he said, contrasting SpaceX's plan to fly its first orbital compute satellites next year. The vehicles will use the same V3 satellite bus already flying for Starlink, modified with larger solar arrays to support compute payloads. Johnsen framed this as a direct extension of Starship's reusability, arguing that once both stages of the vehicle are being reflown routinely, the cost curve for orbital compute inverts relative to terrestrial infrastructure. "In terrestrial, everything is getting actually more expensive... you're actually on the wrong curve," he said, pointing to power, cooling, and construction cost inflation. His expectation is that orbital compute could reach cost parity with terrestrial data centers as soon as next year, then continue declining as launch cadence and satellite efficiency improve — a claim that, if realized, would be a genuine competitive differentiator versus any land-based hyperscaler model.
Terrestrial Compute Buildout: 5-10 Gigawatts Next Year
On terrestrial infrastructure, Johnsen reiterated guidance of 5 to 10 gigawatts of deployed compute capacity next year, up from a little over 2 gigawatts by the end of this year — a steep ramp that he said is underwritten by line-of-sight to power supply the company is largely standing up itself. He acknowledged the industry-wide bottleneck shifting from chips to power and permitting, a dynamic Elon Musk had flagged as early as last year. SpaceX's NVIDIA-exclusive relationship was reaffirmed as central to the compute strategy, with Johnsen noting the value of being seen as a partner that actually deploys allocated compute at scale rather than simply hoarding it.
Compute Deal Structure and Monetization Economics
Johnsen offered unusually specific detail on how SpaceX structures its compute hosting agreements: nearly all are effectively 6-month commitments, structured as 90-day terms with a 90-day out, including the latest $13 billion deal. He tied this cautious structuring to internal conviction in SpaceX's own AI products, saying the company does not want to be "constrained" by giving away compute long-term. On monetization, he confirmed guidance of $30 to $50 per watt for next year, with the company currently tracking at the high end of that range — a figure he suggested compares favorably to what peers like Anthropic and OpenAI are likely generating, though he acknowledged he lacks visibility into their actual financials. Johnsen also noted a "less than one-year payback" dynamic on GPU capital expenditure, which he argued is opening up a new asset-backed financing market for compute infrastructure, distinguishing it from multi-decade infrastructure assets like launch towers or hangars.
Cursor Acquisition and Model Product Momentum
Johnsen pointed to the recent acquisition of Cursor as a turning point in market perception of SpaceX's AI product ambitions, alongside the viral rollout of Grok Bot and incremental model improvements from Grok 4.5 to 4.6, with 4.7 imminent. He described the pace of product iteration as core to justifying a shift away from SpaceX's historically organic growth strategy. "This year, we certainly did some M&A... it's because when you're faced with a product cycle as amazing as the AI product cycle, and time to market becomes so critical, you then look at how do I get to not just the frontier, but how am I winning at the frontier," he said. He argued the strategic logic rests on SpaceX's unique access to lower-cost compute via reusable rocketry translating into structurally cheaper token costs versus competitors.
Starship Flight 13 Learnings and Flight 14 as First Revenue Flight
Johnsen detailed technical outcomes from Starship's Flight 13, which he called a "huge milestone," including a Raptor engine relight, delivery of demonstration V3 Starlink satellites, and what he described as the softest, most precise splashdown yet achieved for the second stage. Recovery teams towed the vehicle back to shore, allowing engineers to inspect heat shield performance directly — data he said is already informing Flight 14, expected later this month. Notably, Flight 14 will be the first Starship flight carrying revenue-generating cargo, flying production V3 Starlink satellites rather than demonstration payloads. Johnsen flagged that subsequent flights will target recovery of both first and second stages, a milestone that would materially accelerate the reusability economics underpinning both terrestrial and orbital compute plans.
Direct-to-Device Mobile: Full 5G Service Targeted for First Half of 2028
On mobile connectivity, Johnsen laid out a two-phase rollout. The current direct-to-device service, live in the U.S. through partners such as T-Mobile's "T-Satellite" offering, is limited to text and light voice — useful for emergencies and dead zones but not a full replacement for terrestrial 5G. A next-generation satellite constellation, using the same V3 bus as Starlink and the AI compute satellites, is slated to begin launching next year, with full 5G-quality service targeted to go live in the first half of 2028. SpaceX has acquired mid-band spectrum from EchoStar to support the U.S. rollout and has FCC approval for both direct-from-space and terrestrial hybrid capability. International rollout will likely require country-by-country regulatory approval, mirroring the path Starlink broadband took to reach over 170 countries. Johnsen suggested incumbent carriers view the offering as a competitive threat, remarking dryly that "the big 3 operators were a little concerned about us and wanted to create a cartel maybe."
Starlink Enterprise and Mobility Demand
Johnsen said Starlink has moved decisively from a consumer-focused, "better than nothing" service in its 2020-2021 launch phase to an enterprise-grade product now capable of 99.9%-99.99% uptime commitments and formal SLAs. He cited a large backlog in aviation, with three of the four largest U.S. airlines now on board (he declined to name the holdout, calling it "kind of embarrassing"), alongside expanding demand from cruise lines, yachts, and trains. He framed physical AI — robotics, autonomous vehicles — as the next major connectivity demand driver, arguing terrestrial networks cannot meet the always-on, high-bandwidth requirements these devices will need.