> ## Content Index > Fetch the complete content index at: https://nexi.fund/llms.txt > Use this file to discover other available public pages before exploring further. # Orbital Raises $5M Pre-Seed to Build AI Data Centers in Space - URL: https://nexi.fund/orbital-data-center-2026/ - Published: 2026-07-01T08:30:31.000Z - Updated: 2026-07-13T12:26:36.000Z - Description: LA-based Orbital raised $5M pre-seed from a16z Speedrun to build AI inference data centers in low Earth orbit: fridge-sized satellites with 100 kW each, NVIDIA GPUs, and a 2027 demo on SpaceX Falcon 9. - Author: Nexi.fund Labs - Tags: Space & Expansion, #mode-5, #hook-character, #track-E, #brand-heavy The last time Euwyn Poon built distributed infrastructure, it was electric scooters parked on city sidewalks. Spin, the startup he co-founded in 2017, deployed thousands of dockless vehicles across US cities before Ford acquired it a year later. The logic was simple: distribute the asset, solve a local problem, scale horizontally. Poon is applying the same logic to something considerably harder to park: AI data centers. Only this time the "sidewalk" is low Earth orbit, and the fleet runs on NVIDIA GPUs cooled by the vacuum of space. Orbital, the Los Angeles-based startup Poon founded earlier this year, announced a $5 million oversubscribed pre-seed round led by a16z Speedrun. The company plans to build and operate a constellation of AI inference data centers in low Earth orbit: fridge-sized satellites, each generating 100 kW of compute power, linked by optical intersatellite connections, and powered by continuous solar energy. 🎯 **Orbital's pre-seed round funds a 2027 SpaceX-hosted demo and the first purpose-built AI compute satellite for 2028.** The company's distributed constellation approach uses many small nodes instead of one football-field-sized satellite. It aims to sidestep the structural and manufacturing constraints that have kept orbital data centers in the realm of conference slides. But the gap between $5 million and a 10-gigawatt orbital compute network is measured in billions, not millions. ## The energy ceiling Terrestrial data centers are hitting a wall that no amount of engineering can easily move. The International Energy Agency projects global data center electricity consumption will more than double by 2030, reaching 945 terawatt-hours annually, roughly Japan's entire yearly demand. In the United States, grid interconnection queues are ballooning, cooling water is becoming a political liability in drought-prone regions, and permitting timelines for new data center construction stretch past five years in key markets. AI inference, the fastest-growing segment of compute demand, amplifies every one of these constraints. Unlike training, which can be scheduled and batched, inference must be available on demand, everywhere, at low latency. That makes it a distributed infrastructure problem, which needs power at the edge, not just in Virginia and Oregon. The thesis is that the most abundant source of distributed power available today happens to be 1,200 miles above our heads. In low Earth orbit, solar irradiance is continuous. No night, no weather, no atmospheric filtering. A satellite in sun-synchronous orbit can collect solar energy at more than five times the density of ground-based panels, around the clock. And the cooling problem that consumes billions of gallons of water on Earth becomes an asset in space, where radiative heat rejection into near-absolute-zero vacuum costs nothing. ## A different architecture What sets Orbital apart from the growing roster of companies pursuing orbital data centers is the shape of the thing itself. SpaceX filed an application with the FCC in January for a non-geostationary satellite system of up to one million satellites to operate as the "SpaceX Orbital Data Center System." Google's Suncatcher project plans two prototype satellites by early 2027\. Starcloud, which counts NVIDIA among its backers, already launched a GPU-equipped satellite in November 2025 and ran a version of Gemini in orbit. Axiom Space is integrating ODC nodes into its commercial station plans. Blue Origin is sketching a 51,600-satellite constellation. Most of these concepts share a premise: build big. A single large satellite with football-field-sized solar arrays, assembled in orbit by robots, hosting hundreds of GPUs in one structure. Poon went the other direction. Orbital's architecture distributes compute across many small, independently deployable satellites, each about the size of a refrigerator with tennis-court-sized solar arrays. The constellation acts as a virtual supercomputer for AI inference, with optical intersatellite links keeping every node in contact with the ground and with each other. Each satellite is designed around NVIDIA's Space-1 Vera Rubin GPU architecture, announced at GTC 2026 specifically for orbital deployment. "When I heard about orbital data centers, I immediately wrote it off," Poon told Payload Space. "Thinking about building a football field size satellite from a maintenance standpoint, physics standpoint, and the cost of having robots construct the thing in space, that seems like a pipe dream." 100 kW compute per satellite ↑ 5× ISS power per node #### Orbital ODC Constellation Specs Each satellite: fridge-sized, 100 kW compute, NVIDIA Space-1 Vera Rubin GPU, optical ISL, radiative cooling. Target: 100,000+ satellites delivering 10+ GW of orbital compute. First demo (Pathfinder) on SpaceX Falcon 9 in 2027\. *Source: Orbital press release, Jun 2026* ## Roadmap and reality The timeline is ambitious by any measure. The company plans a Pathfinder mission in 2027: a hosted GPU payload on a SpaceX Falcon 9 rideshare that will validate radiation tolerance, thermal management, and data downlink in orbit. If that succeeds, Orbital-1, the first purpose-built compute satellite, follows in 2028\. Factory-1, a satellite assembly and testing facility in the South Bay area of Los Angeles, is under development to support scaled production. The company has filed with the FCC for a full constellation license and is positioning for a larger financing round after the pre-seed, as it moves from technology validation to scaled manufacturing. The runway from $5 million to even a single operational satellite is tight. Hardware-intensive space startups typically need Series A rounds of $20 to $40 million to reach orbit. But the oversubscribed round and the breadth of syndicate participation (17 investors across venture, angel, and strategic capital) suggest strong institutional appetite for the thesis. The broader ODC market is moving from PowerPoint to payload. Starcloud has demonstrated that an H100 GPU can operate in orbit under radiation. SpaceX's FCC filing, whether or not it represents a serious near-term product, established orbital compute as a regulatory category. Google's Suncatcher prototypes will fly in 2027\. The question now is which architecture and which business model will first cross the chasm from demonstration to commercial viability. Orbital's bet is that a distributed, horizontally scalable constellation funded by inference compute demand that is growing faster than terrestrial supply will reach that viability sooner than the mega-satellite approach. It is a bet on manufacturing cadence and launch economics rather than on-orbit assembly. If launch costs continue their trajectory toward $200/kg by the mid-2030s, the unit economics of deploying tens of thousands of small compute nodes become calculable in a way that a single bespoke football-field satellite never does. ## What needs to go right The technical hurdles are real. Operating GPUs in space means radiation hardening at a scale never attempted commercially. Thermal cycling between sun and shade in LEO can swing 200°C on a single orbit. Downlinking inference results from 100,000 satellites requires a ground segment architecture that does not yet exist. And the capital required to scale from a single demo to a commercially meaningful constellation runs to billions of dollars, years of execution risk in a startup that today has less than a dozen employees. Poon is candid about the gap. "We're tackling some of the hardest engineering challenges in the industry," he has said, "at a satellite size and constellation scale never attempted before." The proponents of the large-satellite approach make a mirror argument: that building one big thing is actually simpler than orchestrating 100,000 small ones, and that on-orbit assembly, while unproven, eliminates the constellation-scale launch burden. Neither camp yet has revenue to settle the argument. What both camps share is a conviction that terrestrial data center constraints will eventually force compute off the planet. 📊 **Key signals to track** Pathfinder mission (2027): GPU operation, radiation tolerance, thermal performance validated in orbit FCC license decision: constellation authorization signals regulatory viability Series A round size and lead investor: real institutional conviction beyond accelerator stage Launch cost trajectory: sub-$500/kg makes ODC unit economics calculable Terrestrial data center power pricing: the faster grid constraints bite, the faster orbital alternatives become competitive --- *Orbital's story sits at the convergence of two trends we have been tracking closely. As we wrote in* [*June*](https://nexi.fund/orbital-ai-data-centers-2026/)*, the orbital data center category moved from concept to regulatory filing this year. The question now is which company turns that filing into revenue, and whether the answer is a distributed constellation of 100,000 small satellites or something closer to a traditional big-sat approach. Pathfinder 2027 will be the first real data point.* [ Orbital Raises Oversubscribed $5M Pre-Seed Round Orbital's official press release with full funding details, investor list, and mission timeline. Orbital ](https://orbital.inc/press/orbital-raises-5m-preseed.html?ref=nexi.fund) Primary source: funding announcement, technical specs, and CEO statement. [ Orbital raises $5 million to join orbital data center race SpaceNews coverage of Orbital's pre-seed round with competitive context and industry analysis. SpaceNews ](https://spacenews.com/orbital-raises-5-million-to-join-orbital-data-center-race/?ref=nexi.fund) Industry perspective: where Orbital fits in the ODC competition. [ Orbital Raises $5M Pre-Seed to Build ODC Constellation Payload Space interview with founder Euwyn Poon covering the distributed architecture philosophy and technical approach. Payload Space ](https://payloadspace.com/orbital-raises-5m-pre-seed-to-build-odc-constellation?ref=nexi.fund) Detailed technical breakdown: satellite specs, architecture decisions, and founder rationale.