Stunning, Musk Joins Forces with Huang Jen-hsun to Deploy the Stellar Brain! AI Computing Center Launched into Space

marsbitPublié le 2026-08-05Dernière mise à jour le 2026-08-05

Résumé

SpaceX, in its first public earnings report, has announced a partnership with NVIDIA to launch Starmind, an orbital AI computing network. This initiative aims to move AI's computational foundation into space to solve the critical challenges of energy consumption and heat dissipation faced by terrestrial data centers. The core concept involves deploying data centers equipped with NVIDIA's advanced GPUs (like the upcoming Rubin) into sun-synchronous orbit. Key advantages include near-continuous solar power for energy, the vacuum of space for highly efficient radiative cooling, and the absence of terrestrial constraints like land use and grid capacity. The first computational node, the Starmind AI1 satellite, is described as a massive vehicle with a 75-meter wingspan, generating 210 kW of solar power to run a 150 kW computing payload—equivalent to a full NVIDIA GB300 supercomputer rack. These "compute satellites" will connect to users via SpaceX's existing Starlink constellation using high-speed laser links, promising low-latency access from anywhere on Earth. SpaceX's ability to realize this vision is underpinned by its Starship rocket, designed for low-cost, high-frequency launches to transport thousands of satellites, and massive planned manufacturing facilities like the Gigasat factory and the Terafab chip plant. Elon Musk framed the project within the context of the Kardashev Scale, proposing Starmind as humanity's first substantive step toward becoming a Type II civili...

SpaceX released its first public financial report.

In the report, SpaceX officially announced a collaboration with NVIDIA to launch the orbital AI computing network — Starmind.

This time, Musk plans to use a massive satellite matrix to build an infinitely expandable, infinitely energized, absolute-zero cooling 'Space AI Brain' beyond Earth's atmosphere!

Starmind directly relocates the computational foundation of AI from the Earth's surface to space.

Musk intends to use this interstellar computing network to completely solve AI's power consumption and cooling anxieties. Indeed, Earth can no longer contain him.

Everyone thought Musk's ultimate goal was Mars, but on the way to Mars, he is incidentally reshaping the entire landscape of human AI computing power.

This image is AI-generated.

Escaping Earth: Breaking the 'Deadlock' of AI Computing Power

The data in this leaked financial report is impressive: revenue surged by 92%, reaching $7.8 billion. Starlink users have more than doubled, breaking the 12 million mark.

However, none of this data is as captivating as the interstellar computing network.

Today's AI development faces a deadlock—shortage of electricity, water, and land.

A large AI data center consumes as much electricity as a medium-sized city. Earth's energy grid is almost unable to support AI's evolution.

Musk says: Since Earth can't contain it, let's put it in space!

In SpaceX's latest blueprint, the Starmind project proposes a genius solution—

Directly launch data centers equipped with NVIDIA's most advanced chips into sun-synchronous orbit!

Why space? Musk gave a bold explanation in an interview.

First, space always has sunlight.

70% of Earth's surface is water, and even on the remaining land, high-quality areas for solar panels are scarce.

"If aliens came here, they'd probably wonder why it's called Earth when it's mostly water," Musk quipped.

But in sun-synchronous orbit in space, there is no night, no weather interference, and no atmosphere to weaken the sunlight. AI satellites can continuously bathe in pure solar energy 24/7, receiving energy densities several times higher than on Earth.

The second reason is perfect vacuum cooling.

Data centers on Earth spend astronomical amounts of energy on liquid cooling and air conditioning systems. In space, the cooling problem is solved—it can directly radiate waste heat into the deep, cold vacuum of the universe!

Thus, Starmind satellites don't require Earth's energy-intensive cooling towers, fans, or dry coolers.

Even more impressive, leveraging its extensive experience operating vehicles in space, SpaceX has designed deployable liquid radiators covering an area of up to 160 square meters, directly reducing the cooling energy consumption of computing clusters by an order of magnitude.

Additionally, building a large computing center on Earth requires land approvals, dealing with environmental protests, and the grid might not even support it.

But in space, Musk smiles—no land leasing fees, no one protesting at your doorstep. Solar energy is free, vacuum cooling is free, and operating costs are nearly zero!

Starmind AI1: A 'Space Computing Behemoth' with a 75-Meter Wingspan

Is a space data center just an Earth server room stuffed into a metal shell and launched? Obviously not.

The first computing node jointly created by SpaceX and NVIDIA—the Starmind AI1 satellite—is a veritable masterpiece of space industrial art!

According to the published technical details, this satellite is an absolute behemoth.

Its wingspan reaches 75 meters (nearly the length of a football field).

The solar array outputs a whopping 210 kilowatts of power.

The computing payload's peak power is 150 kilowatts, with an average power of 120 kilowatts.

150 kilowatts perfectly matches the power consumption of an entire NVIDIA GB300 supercomputer cabinet!

Musk has clearly stated: "SpaceX is currently committed to using NVIDIA GPUs exclusively because they are the best."

These satellites will carry NVIDIA's not-yet-fully-deployed Rubin GPUs and Vera CPUs, embarking on a frenzy of computation in space.

Some might ask: Won't network latency be crippling with servers so high up?

This is precisely SpaceX's real moat—the Starlink network.

Starmind satellites don't have the complex phased array antennas of Starlink; they are purely 'computing satellites.' They network directly with the existing Starlink constellation via terabit-level ultra-fast laser links.

"People always worry about high latency with space computing," Musk said dismissively in an interview. "But the speed of light is very fast."

AI satellites are only 600 to 800 kilometers from Earth's surface, and light travels 300 kilometers per millisecond. This means the physical latency from a space computing center to your phone or computer is a mere 3 milliseconds!

At this speed, distance is not an issue.

In the future, you could directly ask xAI's Grok large model a question on your phone. Your query would travel via Starlink to NVIDIA computing clusters in space, complete the inference, and instantly return to your screen.

And all of this operates completely independent of Earth's power grids and fiber optic cables!

Musk's Confidence: Starship and the Trillion-Level Manufacturing Titan

Why can only Musk pull this off? Why are other tech giants stuck competing fiercely on Earth?

SpaceX CFO Bret Johnsen hit the nail on the head: "Without an extremely low-cost, fully and rapidly reusable heavy-lift rocket, a space AI brain is pure fantasy."

The most critical piece of Musk's puzzle is Starship.

In the recently concluded V3 test flight, Starship demonstrated its capabilities. It is the largest, heaviest, and most powerful vehicle ever built by humans (with more than twice the thrust of the Apollo-era Saturn V).

The cost of sending materials into space with traditional rockets is prohibitively high. Starship's goal is to launch, land, and relaunch within a day, just like commercial airliners.

Today, nearly 90% of global mass reaching orbit is delivered by SpaceX's Falcon rockets. With Starship, SpaceX's goal is to increase its annual orbital delivery capacity from 2,500 tons to a staggering 1 million tons per year within about three years!

Without Starship's massive payload capacity, launching tens of thousands of multi-ton AI satellites into orbit would be impossible.

To support this 'space city-building' effort, SpaceX's manufacturing capabilities are undergoing an epic expansion—they are building the Gigasat mega-factory in Bastrop, Texas, planning to mass-produce thousands of AI satellites by the end of 2027.

But that's not enough. To support the annual demand for hundreds of gigawatts (GW) and ultimately one terawatt (TW, equal to 1000 GW) of space AI computing power, SpaceX, in partnership with Tesla, has planned an astonishing project called Terafab (terawatt-level chip fabrication plant).

Its footprint will reach an astonishing 100 million square feet—a full 10 times the size of the massive Tesla Gigafactory Texas.

Musk's business closed loop is vividly displayed here—

Massive cash flow from Starlink -> funds Starship R&D -> Starship launches massive numbers of Starmind satellites into space -> builds the space AI brain to empower xAI -> generates higher-dimensional profits.

A perfect, interlocking cycle. No competitor can replicate it; only he can achieve it.

And what's even more terrifying—every link of this chain is open to competitors.

Competitors can buy SpaceX launch services, use Starlink communication, and even rent computing power on Starmind. But Musk doesn't care, because when you are the infrastructure for all tracks, you *are* the track itself.

Musk's Ambition:

Climbing the Kardashev Scale, Advancing Toward a 'God-Level Civilization'

Is Musk doing this to sell AI services and make money? That's underestimating him.

In internal discussions, Musk proposed a grand, almost sci-fi, even religiously sacred narrative—the Kardashev scale.

This is an objective standard in cosmic sociology to measure a civilization's level of development.

Type I Civilization: Can harness all the energy of its home planet (Earth).

Type II Civilization: Can harness all the energy of its home star (the Sun).

Type III Civilization: Can harness the energy of its entire galaxy.

Musk pointed out: "Where are we on the Kardashev scale today? The answer is, we don't even register. The energy we utilize is less than a trillionth of the Sun's output. If we were an object of study for an alien civilization, our performance would be completely failing."

The Starmind interstellar computing network is the substantive first step for human civilization toward becoming a 'Type II Civilization.'

Musk's goal is: within two and a half years, deliver 10 gigawatts (GW) of annualized computing power to space; within three and a half years, reach 100 GW.

And his ultimate goal is one terawatt (1 TW, i.e., 1000 GW) of space computing power! To put that in perspective, that's about twice the current annual electricity consumption of the entire United States!

At the end of the interview, SpaceX executive Dan Huot asked Musk a seemingly sci-fi question: "After we've done our absolute best on Earth, what's next?"

Musk's answer stunned everyone: "Why stop at one terawatt? That's too small."

His vision is to establish a mass driver and chip factory on the Moon.

To achieve computing power a thousand times greater than one terawatt, launching from Earth is no longer sufficient. He plans to mine resources locally on the Moon, using lunar surface materials to manufacture solar panels, radiators, and even AI chips on-site.

Because the Moon has no atmosphere and only one-sixth of Earth's gravity. SpaceX would use large linear motors similar to electromagnetic railguns, requiring no rocket fuel, to literally 'shoot' completed AI satellites from the lunar surface into deep space!

"If we can send that much mass to the Moon, it means—anyone can go to the Moon."

This is Elon Musk.

While many are still fearing 'will AI replace humans?', he is already thinking—

How to extend human consciousness throughout the solar system, how to elevate humanity on the Kardashev scale from that 'utterly insignificant' speck of dust to a species at least worthy of a glance from an alien civilization.

Musk, Sowing Civilization into the Cosmos

Whether it's Mars colonization, Neuralink, AGI, or now the 'Space AI Brain,' every goal Musk proposes sounds crazy at first.

But as SpaceX CFO Bret Johnsen said—

After working with Elon for 15 years, the most amazing thing is: he sets those seemingly audacious goals, then he gets down in the trenches with you to solve the hardest technical bottlenecks. Step by step, you find you're actually moving toward that goal. And, everything he's promised, he's delivered.

In 122 days, he built the world's largest xAI supercomputer 'Colossus'; Starship achieved the awe-inspiring 'chopstick' rocket catch; Starlink weaved an impenetrable net across the global sky.

Now, he has unfurled the blueprint for Starmind.

As NVIDIA's computing power takes wing on SpaceX's Starship, the most hardcore, most spectacular revolution in human technological history is already unfolding thunderously beyond the atmosphere.

Earth, perhaps, is truly just the cradle of human civilization.

And Musk is personally building that 'super brain' capable of illuminating the stars and seas for all humanity.

As Musk said: "Humanity, which hasn't even reached one micro-sol yet, is time to start thinking like a civilization worthy of the universe." Starmind is just the first step.

2027, see you in space.

References:

https://x.com/SpaceX/status/2084723854534951218

https://x.com/openlabxorg/status/2084732002645987402

https://www.spacex.com/spacexai/starmind

Editor: Aeneas

This article is from the WeChat public account "New Zhiyuan", author: ASI Apocalypse

Questions liées

QWhat is the Starmind project, and which two companies are collaborating on it?

AThe Starmind project is an orbital AI computing network, announced as a collaboration between SpaceX and Nvidia.

QWhat are the three main advantages of locating AI data centers in space according to the article?

AThe three main advantages are: 1) Constant, unobstructed solar power in sun-synchronous orbit, 2) Perfect vacuum for passive radiative cooling, eliminating massive cooling energy costs, and 3) No land acquisition, environmental permitting, or local opposition issues.

QWhat is the critical enabling technology that makes the Starmind concept feasible, as explained by SpaceX's CFO?

AThe critical enabling technology is an extremely low-cost, fully and rapidly reusable heavy-lift launch vehicle—specifically, SpaceX's Starship rocket.

QWhat is the ultimate power goal for the Starmind network mentioned by Elon Musk in the article?

AElon Musk's ultimate goal is to achieve one terawatt (1 TW) of computing power in space, which is equivalent to twice the current annual electricity consumption of the United States.

QHow does the article connect the Starmind project to a broader, long-term vision for human civilization?

AThe article connects Starmind to the Kardashev Scale, a measure of a civilization's technological advancement. Starmind represents humanity's first concrete step toward becoming a Type II civilization, capable of harnessing the total energy output of its star (the Sun), by expanding computing infrastructure into space.

Lectures associées

Arthur Hayes en 10 000 mots : Quand la vague de l'IA atteindra-t-elle son sommet ? La bulle va-t-elle éclater ? Pourquoi je rachète massivement de l'ETH ?

Dans un article, Arthur Hayes explore la nature de la bulle de l'intelligence artificielle (IA) et son impact potentiel sur les marchés. Il soutient que les dépenses d'investissement (CAPEX) massives en infrastructures d'IA ne relèvent pas de la "technologie" mais d'un nouveau type d'immobilier, un investissement en "béton et acier" pour des centres de données. Contrairement à la bulle Internet de 2000 (axée sur les bénéfices), Hayes affirme que la bulle actuelle de l'IA est une bulle de crédit similaire à la crise financière de 2008. Le point de rupture critique ne surviendra pas lorsque les géants de l'IA cesseront de gagner de l'argent, mais lorsque la croissance des dépenses d'investissement ralentira, exposant ainsi les risques de crédit et le surendettement des entreprises et des institutions financières. Hayes prévoit que ce ralentissement de croissance pourrait commencer vers mi-2027, suivi d'une période de surchauffe du crédit. Les banques continueront à prêter massivement pour plusieurs raisons : c'est rentable (grâce à une politique de taux favorable), les gouvernements le souhaitent pour des raisons de "sécurité nationale", et elles anticipent un sauvetage étatique en cas de problème. Cette dynamique entraînera une mauvaise allocation du capital à grande échelle. Pour éviter un effondrement, Hayes estime que les gouvernements, en particulier celui des États-Unis, interviendront en imprimant massivement de l'argent, potentiellement via des fonds souverains achetant directement des actions d'entreprises d'IA. Il lie cette perspective monétaire future au cours du Bitcoin. Après avoir absorbé la liquidité initiale, le Bitcoin se stabilisera et entamera une tendance haussière majeure lorsque l'excès de liquidité créé pour soutenir la bulle de l'IA se déversera dans les marchés non traditionnels. Il évoque un objectif à long terme de 1 million de dollars pour le Bitcoin. Enfin, Hayes partage sa vision personnelle sur l'Ethereum (ETH), qu'il considère comme le principal bénéficiaire de la tokenisation des actifs financiers (RWA). Il fixe un objectif de prix approximatif de 5000 dollars pour fin 2026.

Odaily星球日报Il y a 37 mins

Arthur Hayes en 10 000 mots : Quand la vague de l'IA atteindra-t-elle son sommet ? La bulle va-t-elle éclater ? Pourquoi je rachète massivement de l'ETH ?

Odaily星球日报Il y a 37 mins

Criant d'acheter, pourquoi Goldman Sachs reste optimiste sur Samsung et SK Hynix

Malgré une baisse récente de 23% et 35% des actions de Samsung Electronics et SK Hynix, Goldman Sachs réitère sa recommandation d'achat. La banque mise sur trois facteurs pour soutenir sa position optimiste. Premièrement, Goldman anticipe une remontée significative des prix du HBM (High Bandwidth Memory), composant clé pour l'IA, d'ici 2027. Le prix moyen par gigabit devrait atteindre environ 2,9$, retrouvant ainsi un écart de prix par rapport à la DRAM traditionnelle. Cette hausse s'explique par une demande d'IA supérieure à l'offre, une complexité de fabrication accrue et une consommation de capacités de production plus importante. Deuxièmement, la signature d'accords d'approvisionnement à long terme (LTA), d'une durée de 3 à 5 ans, avec de grands clients comme les centres de données, devrait stabiliser le cycle. Ces contrats couvriraient une part importante de la capacité de production prévue (jusqu'à 70% pour Samsung) et incluraient des mécanismes comme des prix planchers et des acomptes, réduisant les risques pour les fabricants. Troisièmement, les stocks des principaux fournisseurs et clients restent à un niveau bas (2-4 semaines), limitant le risque d'un retournement soudain du marché à court terme. De plus, la demande croissante de SSD d'entreprise pour serveurs IA devrait compenser partiellement la faiblesse des marchés grand public comme les smartphones, maintenant le marché NAND sous tension. Enfin, Goldman considère que l'expansion de fabricants comme CXMT ne menace pas à court terme l'équilibre du marché mondial de la DRAM haut de gamme et du HBM, en raison d'écarts technologiques persistants. Les risques persistent (demande d'IA inférieure aux attentes, conditions des LTA moins favorables), mais pour Goldman, la combinaison de la hausse des prix du HBM, des contrats à long terme et des faibles stocks soutient la visibilité des revenus et justifie l'optimisme.

marsbitIl y a 1 h

Criant d'acheter, pourquoi Goldman Sachs reste optimiste sur Samsung et SK Hynix

marsbitIl y a 1 h

Trading

Spot
活动图片