Key Moments
Building the First Data Centers in Space
Y CombinatorWant to know something specific about what's covered?
We've already dissected every moment. Ask and we will deliver (with timestamps).
Key Moments
Space-based data centers are emerging as a solution to Earth's AI energy crisis, but early missions are facing extreme engineering challenges and have cost $2 million for the first prototype.
Key Insights
StarCloud aims to deploy 88,000 satellites, providing on the order of 20 gigawatts of new compute capacity, dwarfing Earth's largest data centers (around 1 gigawatt).
The company's first prototype, StarCloud 1, cost $2 million, a fraction of the estimated $75-$100 million cost quoted by prime contractors for a similar mission.
StarCloud is developing a deployable radiator for thermal management that is 10 times less mass per watt and 500 times less cost per watt than the ISS radiator.
The company's first launch was booked just one day after founding, acting as a crucial forcing function for product development.
StarCloud's initial strategy shifted from space-based solar (losing 95% of energy in transmission) to data centers, identifying a launch cost breakeven of $500/kg.
NVIDIA is collaborating with StarCloud on a new space-specific chip, the Reuben space Reuben one, and StarCloud has heavily modified H100 GPUs for the space environment.
Addressing the terrestrial energy bottleneck for AI
The rapid expansion of artificial intelligence is creating an unprecedented energy demand on Earth, leading to constraints on building new energy projects and data centers. StarCloud's ambitious vision is to circumvent these terrestrial limitations by building data centers in space, leveraging the abundant and low-cost solar energy available in orbit. This approach is becoming increasingly viable due to the rapidly declining costs of space launches, driven by innovations like SpaceX's Falcon 9 and the forthcoming Starship. By harnessing solar power in space, StarCloud aims to provide the massive computational resources required for future AI development without contributing to Earth's energy and environmental challenges.
The pivot from space-based solar to data centers
The initial concept for StarCloud, then named Lumin Orbit, was space-based solar power, a decades-old idea involving large solar arrays beaming energy to Earth. However, analysis revealed that approximately 95% of the energy is lost during transmission from space to the ground. Even with the declining launch costs, this inefficiency made the concept economically unfeasible. The company re-evaluated its approach, noting that most new energy projects on Earth were being built for data centers. This led to a pivot: instead of beaming power down, they would bring the data centers themselves to space. This shift significantly improved the economic breakeven point for launch costs, from an estimated $50/kg for space-based solar to $500/kg for in-space data centers, a more attainable target with current and future launch capabilities.
StarCloud 1: A low-cost, high-risk prototype in orbit
The company's first mission, StarCloud 1, launched in November 2025, was a testament to startup agility and cost-effectiveness. While prime contractors quoted $75-$100 million for a similar endeavor, StarCloud achieved its mission for just $2 million, including launch. This involved significant engineering ingenuity, such as submerging an NVIDIA H100 GPU and other components in phase change material for thermal management, a novel approach for space applications. The team conducted makeshift thermal cycling tests, dunking components in ice baths and using heat guns, highlighting the unconventional methods employed under tight deadlines. Despite these challenges, StarCloud 1 successfully trained the first large language model in space, proving the core concept of running data center-grade GPUs in orbit. The mission's separation video was even featured by NVIDIA CEO Jensen Huang at the GTC conference.
Engineering challenges: Heat and radiation
Operating data centers in space presents two primary engineering hurdles: thermal management and radiation tolerance. Unlike Earth, space lacks atmospheric circulation for cooling, requiring innovative solutions for heat dissipation. StarCloud is developing a lightweight, low-cost deployable radiator designed to be 10 times less massive per watt and 500 times cheaper per watt than the International Space Station's radiator. Simultaneously, the chips must withstand the harsh radiation environment of space, which can flip bits and cause system failures. StarCloud conducts extensive ground testing at particle accelerators to understand component failure points and develops shielding and software mitigations. They claim to be uniquely positioned with data on how H100 and B200 GPUs fail under high-velocity proton and heavy ion bombardment.
Leveraging commercial off-the-shelf components
To reduce costs and accelerate development, StarCloud largely avoids specialized, expensive space-rated components. Instead, they opt for off-the-shelf automotive-grade parts and rigorously test them for performance in harsh environments, similar to SpaceX's strategy. This approach extends to component selection for SSDs, power delivery systems, and converters. By using components with established supply chains and performing their own radiation testing, they can achieve significant cost savings compared to traditional aerospace practices. This philosophy is crucial for making space-based data centers economically viable.
Scaling up with StarCloud 2 and beyond
Following the success of StarCloud 1, the company is planning a phased approach to scaling. StarCloud 2, a 10-kilowatt spacecraft, will serve as a product for government and military customers, offering compute services for existing satellites. The next major milestone is StarCloud 3, a 200-kilowatt, 3-ton spacecraft designed for hyperscale data center clients. These will be launched in batches of 50 per Starship, offering approximately 10 megawatts of compute capacity per launch. The company has filed with the FCC for a constellation of 88,000 such satellites, projecting a total compute capacity of around 20 gigawatts, which is significantly more than the largest terrestrial data centers (around 1 gigawatt). They estimate that up to 10 terawatts of compute capacity could be deployed in sun-synchronous orbits, vastly exceeding current global demand.
Investor skepticism and the shift in deep tech funding
StarCloud faced considerable investor skepticism initially, with many deeming the idea too sci-fi or dependent on unproven low-cost launch capabilities. The company was rejected by over 100 VCs before securing its seed funding and faced further rejections after its YC demo day. However, the landscape has shifted dramatically. The increasing difficulty of building data centers on Earth due to energy and regulatory concerns, coupled with the progressing reality of low-cost space launch, has made space-based data centers more attractive. Benchmark's investment, despite conflicts arising from SpaceX's similar ambitions, signals a growing appetite for hard tech, particularly in space, as investors perceive software moats to be diminishing.
Future outlook and encouraging hardware founders
StarCloud anticipates significant growth in the space industry over the next five years, driven by projected increases in launch capacity. The company's advice to aspiring founders, particularly in hardware, is to prioritize building a world-class technical team, echoing Y Combinator's core philosophy. They emphasize that for hardware startups, securing exceptional engineering talent early on is paramount. With the increasing geopolitical importance of compute and AI, and the regulatory hurdles on Earth, StarCloud is positioned to become a critical player in ensuring Western nations have access to essential intelligence and computational power. The company also plans to include a Bitcoin mining ASIC and AWS Outpost hardware on future missions, catering to diverse market demands.
Mentioned in This Episode
●Products
●Software & Apps
●Companies
●Organizations
●Concepts
●People Referenced
Break-even Launch Cost Comparisons for Space-Based Services
Data extracted from this episode
| Service | Break-even Launch Cost per Kilogram |
|---|---|
| Space-Based Solar | $50 |
| Data Centers in Space | $500 |
StarCloud Radiator Cost and Mass Comparison vs. ISS
Data extracted from this episode
| Metric | StarCloud Radiator | ISS Radiator |
|---|---|---|
| Mass per Watt of Dissipation | 10x less | N/A |
| Cost per Watt of Dissipation | 500x less | N/A |
StarCloud Data Center Scale Comparison
Data extracted from this episode
| Scale | Capacity | Context |
|---|---|---|
| StarCloud 3 (per Starship) | 10 megawatts | New compute capacity |
| StarCloud 3 Constellation (FCC filing) | 20 gigawatts | New compute capacity |
| StarCloud in Dawn Sun-synchronous Orbit | 10 terawatts | Potential capacity (20x US power grid) |
| Largest Terrestrial Data Center | 1 gigawatt | Current benchmark |
Common Questions
Companies are building data centers in space to access virtually unlimited, low-cost solar energy, overcoming terrestrial energy constraints and the increasing difficulty of building new energy projects on Earth. The rapidly decreasing cost of space launch also makes this economically viable.
Topics
Mentioned in this video
StarCloud 1 included two GPUs from ARM.
The first version of the Gemini model was run on StarCloud 1 in orbit.
StarCloud is partnering with AWS to launch Outpost hardware, which will be useful for their military customers.
Mentioned as an example of a software company whose stock, along with other SaaS stocks, declined, contributing to a shift in investor appetite towards hard tech.
Booked their first launch with SpaceX ride share early in the company's founding, and SpaceX's Starship program is seen as a key enabler for lowering launch costs. They are also a partner for Starlink laser terminals for future satellites.
The company building data centers in space to address AI's energy bottleneck. They focus on leveraging solar energy in space and aim to reduce launch costs.
StarCloud became the fastest growing unicorn in YC history. YC's advice on booking launches early and hiring top talent was crucial for the company's development.
Mentioned as an example of a company that Y Combinator has funded in the asteroid mining space.
StarCloud 1 included NVIDIA GPUs, specifically an H100. NVIDIA is a close partner, and StarCloud is working with them on designing new space-hardened chips like the Reuben space chip.
Developing the Nova rocket, mentioned as one of the new launch vehicles coming online.
Developing the Neutron rocket, mentioned as one of the new launch vehicles coming online.
Developing the New Glenn rocket, mentioned as one of the new launch vehicles coming online.
Co-founder and CEO of StarCloud, discussing the company's mission to build data centers in space, the development of StarCloud 1, and the challenges and opportunities in the space industry.
Mentioned as an early writer in the 1940s who conceived of space-based solar power.
CEO of NVIDIA, who featured the StarCloud 1 separation video on a large screen at the GTC conference.
His estimates for Starship deployment are mentioned, and his past statement about the sun being the primary energy source is cited as a hook for StarCloud's mission.
Led StarCloud's $170 million funding round, marking their first space investment and recognizing the potential of StarCloud's technical team and ambitious vision.
StarCloud conducted radiation testing for their hardware at the particle accelerator at Brookhaven National Lab.
Mentioned as a company that ensures telecommunication satellites work in space, particularly in GEO orbits with higher radiation.
Mentioned as a current launch vehicle that is rapidly driving down the cost of getting to space, and is being phased out in favor of Starship.
A key component of StarCloud's strategy, its development at Starbase Texas and potential for massive, reusable launch capacity is what makes space data centers economically viable. StarCloud plans to deploy its StarCloud 3 spacecraft on Starship.
The most interesting GPU on StarCloud 1, which underwent extreme thermal cycling. It was a key component to prove that data center-grade GPUs could operate in space.
Mentioned as a high-power GPU that StarCloud has data on failure points for in high radiation environments.
Mentioned as a high-power GPU that StarCloud has data on failure points for in high radiation environments.
A chip designed in collaboration with NVIDIA specifically for the space environment.
Mentioned as a future GPU that will be included in StarCloud's offerings.
More from Y Combinator
View all 618 summaries
50 minWaymo Co-CEO Dmitri Dolgov: The Demo Is Only 1% Of The Work
31 minPatrick Collison: "What If You Succeed?"
58 minJeff Dean: The 1% Rule for Building in AI
77 minMulti-GPU Kernels, Intelligence per Watt, Heterogeneous Inference, and More | YC Paper Club
Ask anything from this episode.
Save it, chat with it, and connect it to Claude or ChatGPT. Get cited answers from the actual content — and build your own knowledge base of every podcast and video you care about.
Get Started Free