Cowboy Space’s Giant Seattle Lease Turns Orbital Data Centers Into a Manufacturing Bet

Cowboy Space has leased a 291,035-square-foot facility near Seattle, despite having no operational rocket or commercial orbital data center. The deal moves its ambitious plan beyond fundraising decks and into an expensive manufacturing phase. It also creates a clear test: can the startup build rockets and AI infrastructure together before its capital, schedule, or physics catches up with it?

The Kent, Washington, building will support rocket and spacecraft production for Cowboy Space’s planned Stampede constellation. Each vehicle would carry a solar-powered data center into low Earth orbit. The company expects to begin launching its own rockets as early as 2028, according to lease details reported by GeekWire.

That schedule puts Cowboy Space against more than other orbital computing startups. It places the company in direct competition with established launch providers, especially SpaceX and Blue Origin. The startup believes owning its rocket solves a shortage of launch capacity. That solution also forces it to master one of the technology industry’s most difficult manufacturing businesses.

The Kent Lease Makes the Commitment Physical

Cowboy Space has acquired room for a factory before proving the machines that factory must eventually produce.

The industrial building stands at 7650 South 228th Street in Kent, approximately 20 miles south of Seattle. It previously operated as a Costco distribution and delivery center. Newmark, the broker representing landlord CenterPoint Properties, described the transaction as the Puget Sound region’s largest industrial lease of 2026 through the deal date.

Cowboy Space plans to convert the logistics property into a specialized production facility for rockets and space hardware. Newmark says the operation should create 300 jobs. The company had 46 Kent-based openings listed when the lease became public, offering an early indication of its hiring needs.

The site gives Cowboy Space access to an unusually concentrated aerospace labor market. Boeing has operated in Kent for decades, including work connected to NASA’s Apollo lunar rovers. Blue Origin maintains a major presence there, while Stoke Space operates nearby with a team that includes former Blue Origin engineers.

Cowboy Space already has ties to this talent pool. Warren Lamont, its head of launch and propulsion, previously worked at Blue Origin and quantum computing company IonQ. Former SpaceX launch director Tyler Grinnell has also joined the company.

The lease follows several steps intended to establish a full launch program. Cowboy Space disclosed a propulsion testing agreement involving NASA’s Stennis Space Center in Mississippi during July 2026. Lamont has said the team is targeting its first engine hot-fire test in 2027.

An engine hot-fire test involves operating a rocket engine under controlled conditions while measuring thrust, combustion behavior, pressure, and thermal performance. It is an essential development milestone. However, it remains far removed from launching an integrated vehicle safely and repeatedly.

This distinction matters because a large factory can accelerate a mature design, but it cannot substitute for one. The Kent facility must eventually support engine production, stage integration, avionics, testing, and spacecraft assembly. Each function introduces its own equipment, supply chain, certification, and staffing requirements.

The property therefore represents both capacity and exposure. Rent, conversion work, machinery, and payroll begin well before launch revenue. A startup that occupies nearly 300,000 square feet must turn that fixed commitment into a functioning production system.

Cowboy Space is not merely expanding an established satellite business. It is preparing to manufacture several technologies that remain under development. That makes the lease important, but it does not make the company’s orbital data center plan operational.

Why Cowboy Space Wants to Own the Rocket

The company’s central argument is that orbital computing cannot scale if its launch schedule depends on somebody else’s priorities.

Founder and CEO Baiju Bhatt started the company in 2024 under the name Aetherflux. Its original objective was collecting solar energy in orbit and transmitting that energy to Earth. Bhatt later shifted toward putting computing hardware beside the orbital power source.

That change removed the losses and complexity involved in transmitting energy to terrestrial customers. It also created a new logistical problem. Large amounts of computing hardware, power equipment, communications gear, and thermal hardware would still need transportation into orbit.

Bhatt told launch capacity reporters that discussions with multiple providers did not reveal enough future availability. He also concluded that available launch economics would struggle to compete with terrestrial data centers.

Cowboy Space responded by starting its own heavy-lift rocket program. The company expects an initial launch before the end of 2028. Rather than carrying a conventional satellite as a separate payload, its rocket’s upper stage would become the data center after reaching orbit.

An upper stage is the section that takes a payload from the edge of space into its intended orbit. Most missions discard or deactivate that stage after deployment. Cowboy Space wants to fill it with computing equipment and retain it as useful infrastructure.

The company says this design removes redundant structures and avionics. It targets a satellite mass between 20,000 and 25,000 kilograms, with one megawatt of electrical capacity and just under 800 graphics processing units. Those figures describe design goals, not independently demonstrated hardware.

A vehicle capable of carrying that mass would sit above SpaceX’s Falcon 9 in projected performance, although below Starship. That is an extraordinary first rocket project for a company founded only two years ago. Even experienced launch businesses have needed repeated tests and significant capital before achieving dependable orbital service.

Cowboy Space argues that specialization will simplify development. Its rocket does not need to accommodate a broad catalog of commercial payloads. Engineers can optimize the vehicle, upper stage, power system, communications, and computing hardware as one integrated product.

Yet specialization also concentrates risk. A failure in the rocket delays the data center. A thermal problem weakens the mission economics. A communications bottleneck limits useful computing work, even if every other component functions.

The integrated architecture prevents Cowboy Space from changing launch providers easily. That dependence is the strategy’s point, but it is also its central weakness. The company is replacing an external bottleneck with several internal ones.

The Orbital Data Center Race Is Splitting Into Two Routes

Cowboy Space is betting on vertical integration while its rivals rely more heavily on external launch systems and staged demonstrations.

Starcloud, based in Redmond, Washington, represents the closest regional comparison. It has pursued smaller orbital computing demonstrations and plans a production site in nearby Woodinville. The company has also sought authorization for a constellation containing as many as 88,000 satellites.

Starcloud placed an Nvidia H100-class GPU in orbit aboard a 60-kilogram demonstration satellite in November 2025. That mission provided an early test of AI hardware in the orbital environment. It did not validate a commercial data center at meaningful scale.

The company’s longer-term concept involves attaching many data center modules to a large structure with extensive solar arrays. Starcloud has described a multikilometer facility assembled through numerous launches. Its founder has acknowledged that large-scale docking remains years away.

Cowboy Space takes a different route. It proposes free-flying data centers created directly from rocket upper stages. Each launch would add another independent megawatt-class node to the Stampede constellation.

The design reduces the need to assemble one enormous structure in orbit. It also avoids turning the rocket into a disposable transportation expense. However, every node must still operate as a reliable spacecraft, power plant, communications terminal, cooling system, and computing cluster.

SpaceX creates pressure from another direction. Its existing Falcon 9 business has achieved a launch cadence no startup can match. It is also developing Starship while proposing a vast AI-processing satellite network of its own.

If SpaceX reserves future capacity for first-party systems, Cowboy Space’s concern about launch availability becomes credible. If SpaceX instead sells abundant capacity at lower costs, building an independent rocket becomes harder to justify.

Blue Origin adds further competitive tension. Its New Glenn rocket has faced a long development path, illustrating how much time and capital heavy launch systems require. At the same time, Blue Origin possesses manufacturing resources and financial backing far beyond those of most startups.

Google is following a slower research path through Project Suncatcher. Its concept uses tightly coordinated satellite clusters with optical links between spacecraft. Public planning has pointed toward the mid-2030s rather than a near-term commercial service.

These approaches reveal two competing assumptions. One holds that launch will become a widely available transportation layer. The other assumes major orbital infrastructure operators must control their own routes to space.

Cowboy Space has chosen the second assumption. Its Kent factory turns that view into a substantial commitment before the launch market’s future shape becomes clear.

Solar Power Does Not Solve the Heat Problem

Persistent sunlight helps generate electricity, but it does not remove the thermal, communications, maintenance, and replacement constraints of orbital computing.

Terrestrial data centers consume large amounts of power and often depend on water-intensive cooling systems. Grid connections can take years, while local opposition can delay new sites. Orbital proposals promise to bypass land constraints and collect solar energy for longer periods.

A dawn-dusk sun-synchronous orbit follows the boundary between Earth’s illuminated and dark sides. Spacecraft in this orbit can receive sunlight during most of each trip around the planet. That makes the route attractive for systems requiring consistent solar power.

However, space is not automatically cold in a way that conveniently cools computer chips. Heat cannot leave through surrounding air or circulating water. A spacecraft must radiate thermal energy through large surfaces designed for that purpose.

A 2026 engineering analysis of orbital economics modeled a representative one-megawatt system. Its base case required approximately 5,640 square meters of photovoltaic area and 2,500 square meters of radiator area.

The study estimated that solar generation, energy storage, and radiators alone would contribute 29.4 kilograms for each kilowatt of delivered computing power. Adding fixed spacecraft components increased the range to between 34 and 59 kilograms per kilowatt.

These estimates do not describe Cowboy Space’s finalized design. They show why a one-megawatt target cannot be evaluated through GPU count alone. The mass and area needed to power and cool those GPUs directly affect launch economics.

Communications create another constraint. Many AI workloads depend on moving large datasets into a cluster and returning results. A spacecraft cannot deliver the same fiber connectivity available inside a terrestrial campus.

Optical links can transmit large quantities of information using lasers. They still depend on accurate pointing, ground infrastructure, weather conditions, orbital visibility, and network coordination. Useful capacity also depends on how much data a workload must exchange for each unit of computation.

That favors workloads generated in space. An Earth-observation satellite can process sensor data locally, then transmit a smaller result instead of sending every raw image. General cloud computing for terrestrial users demands a much more difficult communications model.

Radiation adds reliability concerns. Commercial processors must operate through energetic particles that can corrupt memory or damage components. Shielding raises mass, while redundancy increases equipment needs and power consumption.

Repair presents a harsher problem. A failed accelerator in a ground facility can be replaced by a technician. Servicing a free-flying orbital data center requires robotic capability, a crewed mission, or acceptance that the damaged hardware has reached the end of its useful life.

Computing equipment also becomes obsolete quickly. A satellite designed around current accelerators may still work after newer terrestrial hardware delivers much more output per watt. Cowboy Space must recover its deployment cost during a finite useful period.

Its own technical disclaimer describes published capabilities, schedules, and performance targets as objectives rather than demonstrated outcomes. That distinction should frame every assessment of the Kent expansion.

The facility shows the company intends to pursue the engineering challenge. It does not answer whether a complete orbital system will outperform a ground data center economically.

The Funding Supports a Factory, Not a Finished System

Cowboy Space has enough capital to begin serious hardware development, but its agenda spans several capital-intensive programs at once.

The company announced a $275 million Series B round in May 2026 at a reported post-money valuation of $2 billion. Index Ventures led the financing, with participation from Breakthrough Energy Ventures, Construct Capital, IVP, and SAIC.

Cowboy Space had previously raised $80 million. The latest round therefore brought its disclosed funding total to $355 million. That capital supports expansion, hiring, rocket development, and work on the data center constellation.

The investor group provides more than money. SAIC brings experience with government technology programs. Breakthrough Energy Ventures focuses on climate and energy systems. Cowboy Space also presents its technology through energy security and national security arguments.

The company has identified Nvidia as a collaborator and says it plans to deploy Nvidia Space-1 Vera Rubin Modules. That association strengthens the computing side of its story, but it does not validate Cowboy Space’s integrated spacecraft or launch schedule.

A processor partnership cannot close the full system. The startup still needs engines, tanks, structures, flight software, launch infrastructure, ground stations, solar arrays, radiators, optical communications, and operational licenses.

Rocket development regularly consumes capital faster than initial projections suggest. Test failures are part of the process, but each failure affects hardware, facilities, staffing, and schedules. A company that develops both launch vehicles and payloads carries overlapping technical programs.

The Kent lease increases those demands. Converting a former distribution center into an aerospace factory requires specialized tooling and carefully controlled production areas. Engine and propellant work also impose safety and regulatory requirements unfamiliar to ordinary warehouse operators.

Hiring 300 people would expand the organization substantially. Rapid recruitment can provide needed expertise, especially in a region filled with aerospace veterans. It can also strain management, design control, documentation, and coordination across California, Washington, and Mississippi.

The 2028 launch target is therefore more useful as a testable deadline than as a guaranteed arrival date. Investors and potential customers can watch whether engine testing begins during 2027, whether major design reviews conclude, and whether integrated hardware enters production.

The startup must also clarify its initial customer. Large language model training requires fast communication among thousands of processors. Distributed inference and processing of space-generated data can work with looser coordination.

Cowboy Space often frames Stampede around broad AI demand. Its first commercially credible workloads will probably be narrower. Customers with satellite imagery, defense sensing, or other space-native data may tolerate orbital constraints more easily than ordinary cloud users.

That narrower market can still matter. Early launch providers succeeded by serving specific missions before expanding their ambitions. Cowboy Space does not need to replace terrestrial cloud computing immediately, but it does need a workload that pays for scarce orbital resources.

Environmental claims need similar discipline. Moving computing off Earth can reduce direct pressure on local grids, land, and cooling water. Rocket production, launch emissions, satellite replacement, and orbital debris introduce different costs.

Large constellations also increase collision-management requirements. Spacecraft must track surrounding objects, perform avoidance maneuvers, and leave orbit safely after service. A failed 20,000-kilogram platform would present a different risk from a small demonstration satellite.

The core financial question is not whether orbital computing can operate. Small experiments already show that processors can function beyond Earth. The question is whether useful computation can cover the combined cost of launch, spacecraft construction, communication, operations, and replacement.

What to Watch Before Cowboy Space Reaches Orbit

Three signals will show whether the Kent lease is becoming a production system or merely enlarging the company’s fixed costs.

The first signal is a sustained engine hot-fire campaign in 2027. One successful firing would mark progress, but repeated tests across realistic operating conditions would matter more. Results should demonstrate stable combustion, target thrust, acceptable temperatures, and restart behavior where the architecture requires it.

A delayed or abbreviated campaign would weaken the 2028 launch schedule. Consistent testing would support the argument that Cowboy Space has moved from conceptual design into hardware development.

The second signal is evidence of an integrated thermal and computing prototype. The company needs to show how accelerators, power electronics, batteries, radiators, shielding, and optical communications work as a system. A laboratory demonstration would not prove orbital economics, but it would reduce major technical uncertainty.

Investors should look beyond headline GPU counts. The important measurements include delivered computing power, radiator area, total mass, useful communications capacity, and projected service life. These variables determine how much productive work each launch can support.

The third signal is a defined launch and customer path. Cowboy Space needs regulatory filings, a launch site strategy, mission approvals, and credible users for its initial capacity. A customer focused on space-generated data would strengthen the plan more than a broad promise to serve every AI workload.

Competitor progress will shape that test. Starcloud’s demonstrations can reveal how commercial accelerators survive radiation and thermal cycles. SpaceX’s launch pricing and first-party computing plans can either validate Cowboy Space’s scarcity thesis or make its independent rocket look unnecessarily expensive.

The industry’s early tests already show that storing or processing small amounts of data off Earth is possible. Scaling that work into a commercially competitive megawatt facility remains a separate challenge.

Cowboy Space’s Kent lease matters because it places real people, equipment, and recurring costs behind that challenge. The startup has chosen to control transportation, power, computing, and communications instead of waiting for a mature orbital supply chain.

That decision gives Cowboy Space more control if its system works. It also removes many external explanations if it does not.

For developers and enterprise buyers, the immediate question is not when everyday cloud workloads will leave Earth. It is which workloads gain enough from orbital processing to justify the added constraints. Watch the engine tests, the thermal prototype, and the first customer commitment. Together, those signals will show whether Cowboy Space orbital data centers are becoming practical infrastructure or remaining an unusually large manufacturing bet.