AI Wiki

SpaceX Starmind

AI HardwareAI InfrastructureData CentersNVIDIA
18 min read
Updated
Suggest editHistoryTalk19 citationsFact-checked
RawGraph

SpaceX Starmind is SpaceX's planned constellation of solar-powered artificial intelligence compute satellites, intended to function as orbital data centers that run AI workloads in space and beam results back to Earth through the Starlink network. On August 4, 2026, SpaceX announced that it is partnering with NVIDIA to design the compute payload of the program's first satellite, designated AI1, writing that "each of the Starmind satellites will include NVIDIA Rubin GPUs and Vera CPUs for datacenter class space compute."[1] The announcement extends the NVIDIA Vera Rubin data center platform into orbit, and it capped a six-month sequence in which SpaceX filed with the U.S. Federal Communications Commission to deploy up to one million orbital data center satellites, absorbed the AI developer xAI in the SpaceX-xAI merger, unveiled the AI1 satellite design and a dedicated "Gigasat Factory" in Bastrop, Texas, and named the constellation Starmind.[4][5][6][7]

SpaceX describes the program's goal as "capturing solar energy in space to power low-cost, high-performance AI compute for Earth," and says it plans to mass-produce "thousands of AI satellites starting as soon as late 2027."[2] Those deployment targets, along with the associated chip fab and factory buildouts, are company plans and projections rather than demonstrated capabilities; as of August 2026 no Starmind satellite had been launched. On SpaceX's first earnings call as a public company, held August 4, 2026, Musk went beyond the partnership announcement, saying "we're exclusive to Nvidia" for AI compute and that Starmind satellite launches will begin in 2027.[16]

Key facts

AttributeDetails
ProgramStarmind, SpaceX AI satellite constellation (orbital data centers)
OperatorSpaceX (post-merger parent of the SpaceXAI division; see SpaceX-xAI merger)
First satelliteAI1, prototype design unveiled June 8, 2026[6]
Compute payloadNVIDIA Rubin GPUs and Vera CPUs, per the partnership announced August 4, 2026[1]
Power per satelliteUp to 250 kW peak, 175 kW average, per the official Starmind page in August 2026[2]
Earlier disclosed payload120 kW average, 150 kW peak compute payload, per Musk's June 2026 presentation[6]
OrbitSun-synchronous orbit for near-continuous solar illumination[2]; FCC filing covers 500-2,000 km at 30-degree and sun-synchronous inclinations[5]
CoolingDeployable liquid radiators rejecting heat to vacuum[2]
ConnectivityLaser intersatellite links, results relayed to Earth via the Starlink constellation[2]
ManufacturingGigasat Factory, Bastrop, Texas (planned, about 11 million sq ft)[7]
Constellation cap filedUp to 1,000,000 satellites (FCC application, January 30, 2026)[5]
Stated ramp target"Thousands of AI satellites starting as soon as late 2027"[2]; first Starmind launches expected in 2027, per Musk on the August 4, 2026 earnings call[16]
SpaceX AI segment compute (currently terrestrial)1.4 GW nameplate as of June 30, 2026; Musk targets over 2 GW by end of 2026 and close to 10 GW by end of 2027[17][16]

Background

From FCC filing to merger

SpaceX's orbital compute ambitions became a formal program at the end of January 2026, when the company applied to the U.S. Federal Communications Commission for an "Orbital Data Center System" of up to one million solar-powered satellites at altitudes between roughly 500 and 2,000 kilometers, in 30-degree-inclination and sun-synchronous orbits, connected by optical intersatellite links with Ka-band backup for telemetry and command. The application asked the FCC to waive its standard deployment milestones, which would otherwise require half the constellation to be in orbit within six years.[5]

Three days later, on February 2, 2026, Elon Musk announced that SpaceX would acquire xAI in an all-stock transaction, explicitly framing orbital data centers as a rationale for the combination; the merged company went public on the Nasdaq in June 2026.[14][15] In a February 2026 company update, Musk wrote that "by directly harnessing near-constant solar power with little operating or maintenance costs, these satellites will transform our ability to scale compute," and described a million-satellite constellation of orbital data centers as "a first step towards becoming a Kardashev II-level civilization, one that can harness the sun's full power, while supporting AI-driven applications for billions of people today and ensuring humanity's multi-planetary future."[4]

AI1 reveal and the Starmind name

On June 8, 2026, three days before SpaceX priced its initial public offering, Musk laid out the first detailed design of the AI1 satellite in a roughly 30-minute video posted to SpaceX's X account. As presented then, AI1 carried a compute payload averaging 120 kW with a 150 kW peak, at a density of about 70 kW per ton, operating at roughly 600 km altitude, with a deployed wingspan of about 70 meters, slightly wider than a Boeing 747-8. Musk compared one satellite's payload to roughly one NVIDIA GB300 rack, which draws around 140 kW on the ground. The design used an interchangeable compute-module architecture so that processors could come from any chipmaker.[6]

The constellation's name came later that month. On June 23, 2026, Musk confirmed that the AI satellite megaconstellation would be called "Starmind," extending the naming family of Starship, Starlink, Starshield, Starfall, Starbase and Starfactory; Space.com noted that, if built as filed, Starmind would be roughly 100 times larger than the then-current Starlink constellation of nearly 10,700 active satellites.[4] On July 9, 2026, Musk posted "Starmind" on X with a link to a new official page at spacex.com/spacexai/starmind, which has since served as the program's primary public description.[3]

The NVIDIA partnership

The August 4, 2026 announcement

At 19:31 UTC on August 4, 2026, SpaceX posted on X: "SpaceX is partnering with @Nvidia to design the Starmind AI1 satellite compute payload. Each of the Starmind satellites will include NVIDIA Rubin GPUs and Vera CPUs for datacenter class space compute," linking to the official Starmind page and attaching renderings and a video of the satellite.[1] The announcement landed the same afternoon that SpaceX released its first quarterly financial results as a public company, with an earnings webcast scheduled for 4:30 p.m. Eastern that day.[8]

The Vera Rubin platform is NVIDIA's 2026 flagship data center generation, pairing Rubin GPUs with Vera CPUs in rack-scale systems. The Starmind selection extends that hardware family into orbit. NVIDIA had been preparing space-qualified versions of the platform for months: at its GTC conference on March 16, 2026, the company launched a space computing initiative built around the NVIDIA Space-1 Vera Rubin Module, which it said delivers "up to 25x more AI compute for space-based inferencing" compared with an H100 GPU, alongside its IGX Thor and Jetson Orin edge platforms. "Space computing, the final frontier, has arrived," NVIDIA CEO Jensen Huang said in that announcement. "As we deploy satellite constellations and explore deeper into space, intelligence must live wherever data is generated."[9] The companies named as users of NVIDIA's space computing platforms in the March release were Aetherflux, Axiom Space, Kepler Communications, Planet Labs, Sophia Space and Starcloud; SpaceX's adoption was announced separately in August.[9][1]

Neither company disclosed financial terms, order volumes, or which specific Rubin-generation parts (Rubin, Rubin Ultra, or the Space-1 module) will fly on AI1. Two weeks before the announcement, on July 20, 2026, Musk had dismissed as "fake news" a report in Taiwanese media that SpaceX had placed a $52 billion order for roughly one million NVIDIA GB300 GPUs through Foxconn.[10]

Vendor agnosticism

The partnership sits alongside a deliberately supplier-neutral architecture. The official Starmind page states: "We are AI chip vendor agnostic. Our system architecture supports compute modules from any provider."[2] This is consistent with the interchangeable-payload design Musk presented in June 2026, which was explicitly intended to let whichever vendor ships the most competitive silicon supply the processors, and with SpaceX's own plans to eventually fabricate chips in-house.[6] Hours after the announcement, however, Musk framed the relationship very differently: on SpaceX's August 4, 2026 earnings call he said the company had "decided to build exclusively on Nvidia" (see below), even as the official Starmind page continued to carry the vendor-agnostic language as of August 6, 2026.[18][2]

The first earnings call and NVIDIA exclusivity

SpaceX held its first earnings call as a public company in a live webcast beginning at 4:30 p.m. Eastern on August 4, 2026, hours after the partnership announcement.[17] On the call, Musk said SpaceX had "decided to build exclusively on Nvidia," calling the Vera Rubin architecture the best available: "We think it's the best AI computer, and we greatly value our close cooperation and partnership on many levels with Nvidia. So we're exclusive to Nvidia."[16][18] He added that SpaceX's "understanding with Nvidia is that we will receive a very significant percentage of their GPUs next year."[18] The next day Musk repeated the commitment in a post on X, writing, per Business Insider, that "SpaceX has committed to using Nvidia GPUs exclusively because they are the best"; the publication noted that an exclusive commitment cuts out rival chipmakers such as AMD, Intel and Broadcom.[19]

Musk said SpaceX plans to deploy NVIDIA's Vera Rubin NVL72 rack-scale systems, which connect 72 Rubin GPUs with 36 Vera CPUs, both on the ground and in space, described the Starmind satellite as "essentially an optimized Vera Rubin NVL72 computer," and said the company expects to begin launching Starmind satellites in 2027.[16][18] He told investors that SpaceX expects to end 2026 with more than 2 gigawatts of compute capacity and to reach close to 10 gigawatts by the end of 2027.[16][18] SpaceX's earnings release put the AI segment's nameplate compute at 1.4 GW as of June 30, 2026, up from 1.0 GW at the end of the first quarter and 0.4 GW a year earlier.[17]

The results themselves gave the first public financial picture of the AI business Starmind is meant to serve. AI segment revenue was $2.56 billion for the second quarter of 2026, up 213% sequentially and 247% year over year, growth SpaceX attributed to new Cloud Services Agreements, totaling $14.1 billion in contracted sales, and to increased Grok and X subscription revenues; the segment posted a $1.26 billion operating loss on $15.8 billion of capital expenditures in the quarter.[17] Yahoo Finance reported that SpaceX had entered multibillion-dollar leasing agreements under which Google and Anthropic will rent SpaceX data center capacity to power their AI services.[16]

Markets treated the exclusivity declaration as an NVIDIA endorsement: NVIDIA shares rose about 4% on August 5, 2026, their fifth consecutive session of gains, while SpaceX shares, which had fallen more than 7% in after-hours trading as investors focused on the company's spending, dropped more than 10% on August 5 before recovering part of the decline.[16][18][19] MarketWatch quoted Wolfe Research analyst Chris Caso estimating that NVIDIA could see roughly $35 billion in revenue per installed gigawatt of its chips, which would make SpaceX's stated 2027 capacity target alone exceed the incremental data-center growth Wall Street then expected for NVIDIA in 2027.[18]

AI1 satellite design

SpaceX describes AI1 as an "orbital satellite with localized compute that takes advantage of readily available power and cooling in space and beams back data via high-bandwidth lasers to the Starlink constellation."[2] The official page organizes the design around three constraints it says limit ground-based data centers: power, cooling, and connectivity.[2]

Power. The page lists satellite power of "up to 250 kW peak / 175 kW average," a substantial increase over the 150 kW peak and 120 kW average compute payload Musk described in June 2026.[2][6] In sun-synchronous orbit, the page says, AI1 satellites capture continuous solar power with no weather or atmospheric loss, avoiding Earth's grid, land, and cooling constraints. The high-efficiency solar arrays are described as "derived from the technology already powering thousands of Starlink satellites" and manufactured by SpaceX in Bastrop, Texas.[2]

Cooling. Deployable liquid radiators reject heat into the vacuum of space; SpaceX says that without the chillers, cooling towers, fans, or dry coolers of terrestrial data centers, its AI satellites "reduce cooling power overhead by an order of magnitude."[2] The June 2026 design showed up to 110 square meters of deployable liquid radiators with redundant pumped loops and micrometeoroid shielding. For scale, Tom's Hardware noted that the International Space Station's external thermal control system rejects roughly 70 kW across 422 square meters of radiator, about half the heat load of a single 140 kW GB300 rack.[6]

Connectivity. High-speed laser links interconnect the satellites and hand results to the Starlink constellation for the trip back to ground stations, which SpaceX characterizes as a low-latency, high-bandwidth connection.[2] The company argues that satellites with localized AI compute in sun-synchronous orbit "are the fastest way to scale given the growing difficulty of finding electrical power sources on Earth."[2]

SpaceX also emphasizes manufacturing heritage and orbital sustainability, stating that its AI satellites "use proven technology and are more streamlined than the Starlink satellites we already manufacture at scale," and that the constellation is being designed with orbital-debris safety in mind.[2] Starship's reusable architecture and payload capacity are presented as critical for deploying the large, heavy satellites.[2]

Manufacturing: the Gigasat Factory

SpaceX is building what it calls the Gigasat Factory in Bastrop, Texas, "to manufacture AI satellites at massive scale," with the stated aim of enabling "rapid production and deployment of thousands of AI satellites starting as soon as late 2027."[2] As announced on June 8, 2026, the facility is planned at about 11 million square feet on a roughly 1,000-acre site, more than ten times the size of Starfactory, SpaceX's Starship production complex. The plan vertically integrates much of the satellite supply chain on one campus: solar ingots and wafers, solar cells, printed circuit boards, silicon electronic components, user terminals, gateways, and final AI1 assembly. At the time of the announcement, Musk said the solar manufacturing buildings were already under construction and the AI satellite production building was about to break ground.[7]

The production targets attached to the factory are aggressive and, as of August 2026, entirely prospective. Musk said SpaceX would try "to get to roughly an annualized rate of a gigawatt per year by the end of next year at Space AI compute," then scale "by an order of magnitude per year," reaching an annualized 10 GW within about two and a half years and possibly 100 GW within about three and a half.[7] At roughly 150 kW of compute per satellite as then specified, a 1 GW annual rate implies launching on the order of 6,000-plus AI1 satellites per year, versus about 10,500 active Starlink satellites accumulated since 2019.[7][4]

Chip supply and the Tesla fab project

The Starmind page states that "SpaceX plans to produce advanced AI chips at scale in its new fab facility, a joint project with Tesla using a modular architecture that enables seamless integration and continuous upgrades of next-generation processors."[2] The page does not name the fab, but Tom's Hardware identified the joint Tesla venture as TeraFab, the semiconductor project Musk announced on March 22, 2026: a chip fabrication plant, reportedly carrying an initial price tag of about $20 billion, to be built on Tesla's campus in eastern Travis County near Austin, Texas.[2][6][11]

In the March announcement livestream, Musk said the fab would put logic, memory, packaging, testing, and lithography-mask production under one roof, and that it would make two classes of chips: an edge-inference part for Tesla vehicles and Optimus robots, and what the publication described as a higher-power chip hardened for the space environment, designed to run hotter than terrestrial parts to minimize radiator mass on satellites. He framed the project as a response to chip supply constraints ("We either build the TeraFab, or we don't have the chips, and we need the chips, so we build the TeraFab"), estimating global AI chip output at roughly 20 gigawatts per year against an eventual need he put at 100 to 200 gigawatts per year of terrestrial chips and "on the order of a terawatt of chips in space." He gave no timeline for production, and said Tesla, SpaceX, and xAI would keep buying from TSMC, Samsung, and Micron in the meantime. SpaceX's own IPO filing had warned that the company could not yet secure as many chips as its orbital plans require.[11][6]

Timeline

DateEvent
January 30, 2026SpaceX files FCC application for up to 1,000,000 orbital data center satellites at 500-2,000 km[5]
February 2, 2026SpaceX announces all-stock acquisition of xAI, citing orbital data centers as a rationale[14]
March 16, 2026NVIDIA launches space computing platforms at GTC, including the Space-1 Vera Rubin Module[9]
March 22, 2026Musk announces the TeraFab chip fab, a Tesla-SpaceX joint project in Austin, Texas[11]
June 8, 2026Musk unveils the AI1 satellite design (120 kW average / 150 kW peak payload) and the 11-million-sq-ft Gigasat Factory in Bastrop[6][7]
June 23, 2026Musk confirms the constellation name "Starmind"[4]
July 9, 2026Musk links the official Starmind page on spacex.com[3]
August 4, 2026SpaceX announces the NVIDIA partnership for the Starmind AI1 compute payload (Rubin GPUs, Vera CPUs); official page lists power of up to 250 kW peak / 175 kW average[1][2]
August 4, 2026On its first earnings call as a public company, Musk says SpaceX is "exclusive to Nvidia," plans Vera Rubin NVL72 systems on the ground and in space, and expects first Starmind launches in 2027[16][18]
Late 2027Stated start of mass deployment, "thousands of AI satellites" (company plan)[2]

Context and competing efforts

Starmind is the largest of several announced attempts to put AI compute in orbit. Starcloud, a startup that flies NVIDIA hardware, launched its first satellite, Starcloud-1, carrying an NVIDIA H100 GPU in November 2025, raised a $170 million Series A at a $1.1 billion valuation in March 2026, and is designing a 200 kW spacecraft sized for launch on SpaceX's Starship.[12] Google announced Project Suncatcher in November 2025, "a research moonshot" to scale machine learning in space using its TPU chips on interconnected solar-powered satellites, with two prototype satellites planned with Planet for launch by early 2027.[13] NVIDIA's March 2026 space computing initiative named a further set of partners working with its orbital platforms, including Axiom Space, Kepler Communications, Planet Labs and Sophia Space.[9]

The economics and engineering of orbital data centers remain unproven at anything near the announced scale, and the concept has prominent skeptics; Tom's Hardware noted that OpenAI CEO Sam Altman had called orbital data centers "ridiculous" earlier in 2026, while Musk responded to thermal-management critiques by saying it is "safe to say SpaceX knows how to do heat rejection in space," pointing to the more than 10,000 Starlink satellites SpaceX operates.[6] SpaceX has nonetheless begun signing compute customers ahead of deployment: Tom's Hardware reported that the company's signed compute deals included a $920 million-per-month agreement with Google.[6] The Starmind-specific dates and capacity figures cited in this article, including the late-2027 mass-deployment start, the gigawatt-per-year ramp, and the fab output targets, come from SpaceX itself and should be read as company plans.

See also

References

  1. ^SpaceX (@SpaceX) on X. "SpaceX is partnering with @Nvidia to design the Starmind AI1 satellite compute payload." August 4, 2026. x.com/...2084723854534951218
  2. ^SpaceX. "Starmind." Official SpaceX website, accessed August 6, 2026. spacex.com/...starmind
  3. ^Elon Musk (@elonmusk) on X. "Starmind." July 9, 2026. x.com/...2075367885438890134
  4. ^Mike Wall. "Another 'Star' is born: SpaceX names AI megaconstellation 'Starmind'." Space.com, June 25, 2026. space.com/...x-names-ai-megaconstellation-starmind
  5. ^SpaceNews. "SpaceX files plans for million-satellite orbital data center constellation." January 2026. spacenews.com/...orbital-data-center-constellation
  6. ^Luke James. "Elon Musk's first-gen orbital data center craft spans wider than a Boeing 747 and runs an interchangeable chip payload." Tom's Hardware, June 9, 2026. tomshardware.com/...ails-its-ai1-compute-satellite
  7. ^Tom's Hardware. "SpaceX unveils 11-million-square-foot Gigasat factory, a new manufacturing facility for space-based data centers." June 9, 2026. tomshardware.com/...-late-2027-from-its-satellites
  8. ^SpaceX Investor Relations. "SpaceX to Post Second Quarter 2026 Results and Host Webcast on August 4, 2026." July 20, 2026. ir.spacex.com/...default
  9. ^NVIDIA Newsroom. "NVIDIA Launches Space Computing, Rocketing AI Into Orbit." March 16, 2026. nvidianews.nvidia.com/...space-computing
  10. ^Teslarati. "Elon Musk debunks $52 billion SpaceX-NVIDIA GPU deal." July 20, 2026. teslarati.com/...-spacex-nvidia-gpu-deal-fake-news
  11. ^Luke James. "Elon Musk unveils $20 billion 'TeraFab' chip project to make chips, memory, and package processors all under one roof." Tom's Hardware, March 22, 2026. tomshardware.com/...0-billion-terafab-chip-project
  12. ^TechCrunch. "Starcloud raises $170 million Series A to build data centers in space." March 30, 2026. techcrunch.com/...-ato-build-data-centers-in-space
  13. ^Google. "Project Suncatcher." The Keyword (Google blog), November 4, 2025. blog.google/...google-project-suncatcher
  14. ^CNBC. "Musk's xAI, SpaceX combo is the biggest merger of all time, valued at $1.25 trillion." February 3, 2026. cnbc.com/...musk-xai-spacex-biggest-merger-ever
  15. ^NPR. "SpaceX blasts off with a record-breaking $75 billion IPO." June 11, 2026. npr.org/...spacex-ipo-price-elon-musk
  16. ^Daniel Howley. "Nvidia stock rises after Musk says SpaceX will exclusively use company's chips." Yahoo Finance, August 5, 2026. uk.finance.yahoo.com/...e-companys-chips-211803293
  17. ^SpaceX. "SpaceX Reports Second Quarter 2026 Results." Exhibit 99.1 to Form 8-K, SEC EDGAR, August 4, 2026. sec.gov/...earningsreleaseq22608042
  18. ^Britney Nguyen. "Nvidia's stock is basking in the glow of a high-profile endorsement." MarketWatch, August 5, 2026. marketwatch.com/...gh-profile-endorsement-b7c48e7b
  19. ^Shubhangi Goel. "SpaceX and Nvidia are taking their relationship exclusive, here's what Musk said about their new status." Business Insider, August 5, 2026. businessinsider.com/...only-buy-from-nvidia-2026-8

Improve this article

Add missing citations, update stale details, or suggest a clearer explanation. Every suggestion is reviewed for sourcing before it goes live.

4 revisions · v5 · 3,547 words · full history

Fact-checks are independent of edits: a reviewer re-verifies the article against its sources and stamps the date. How we verify

Research and drafting on this wiki are AI-assisted, under named human editorial standards. How AI is used here

Reviewer note: Earnings-call section verified against SpaceX's SEC 8-K exhibit and named-reporter coverage; the August 4, 2026 call date confirmed from the SEC filing; financial figures quote-matched to the exhibit; the NVIDIA exclusivity quotes verified across two outlets; the official Starmind page's vendor-agnostic language confirmed still live as of August 6, 2026.

Cite this page: AI Wiki. "SpaceX Starmind." aiwiki.ai, updated 4 Aug 2026, fact-checked 5 Aug 2026. CC BY 4.0. https://aiwiki.ai/wiki/spacex_starmind

Suggest edit