Orbital AI Compute Gets Its First Institutional Series B, What Cowboy Space's $275M Raise Changes for Infrastructure...

The $355M number is the one that matters.

Not the $275M in this round. The cumulative $355M in total outside funding that SiliconANGLE confirmed when Cowboy Space Corp. disclosed its Series B. That’s a number large enough to actually build and launch orbital hardware. It’s not a proof-of-concept budget. It’s a deployment budget, and the investors who wrote the checks, Index Ventures leading, with NEA and IVP alongside, have underwritten that interpretation.

Cowboy Space builds AI data centers designed for low-earth orbit. The power source is orbital solar: direct sunlight, continuous exposure, no grid connection required. CEO Baiju Bhatt co-founded Robinhood Markets, which means he’s demonstrated he can take a technically complex, regulation-adjacent infrastructure product from concept to mass adoption. Index Ventures doesn’t need to be told who Bhatt is. They wrote the check knowing the pedigree.

Why Orbit? The Power Argument Made Concrete

Epoch AI’s 2026 tracking shows hyperscaler capex trending toward $770B annually, a figure that assumes terrestrial power infrastructure can keep pace. It probably can’t, at that speed.

The U.S. electrical grid permitting process for large-scale data centers now runs 3-7 years in most jurisdictions. Power purchase agreements for renewable energy are oversubscribed in every major data center market. Water cooling requirements are generating regulatory pushback in drought-affected regions. These aren’t hypothetical constraints, they’re the reason Meta signed a 1GW space-based solar agreement and the reason fuel cell deployments are accelerating across the hyperscaler ecosystem.

Orbital solar doesn’t route through any of those bottlenecks. A module in LEO generates power continuously, at predictable output, without land permitting, grid interconnection queues, or water. The latency tradeoff is real, LEO introduces round-trip communication delays that make orbital compute unsuitable for applications requiring sub-10ms response times. But for batch inference, model training runs, and asynchronous agentic workloads, the latency is tolerable. Cowboy Space is positioning for exactly that segment of the market.

Three Models, Updated for May 2026

The framework for evaluating alternative AI compute power models has three branches: grid (traditional utility power, renewable PPA where available), fuel cell (on-site hydrogen or natural gas generation, bypassing grid interconnection), and orbit (LEO solar, bypassing terrestrial infrastructure entirely). A prior brief on these three models laid out the comparative landscape in April.

What’s changed since then:

*Grid:* Still the lowest-cost option at scale for established hyperscaler campuses where power is already contracted. The constraint is new capacity, permitting timelines mean new grid-connected data centers that break ground in 2026 won’t be operational until 2029 or later in most U.S. markets.

*Fuel cell:* Gaining traction as a bridge solution. Faster permitting than grid in many jurisdictions, lower cost than orbit, but still dependent on hydrogen or gas supply chains. Several hyperscaler campuses are now deploying fuel cell capacity specifically for AI workload overflow.

*Orbit:* Before today, institutionally interesting but pre-Series B. After today, $2B valued and institutionally backed. That doesn’t make orbit cost-competitive with grid power per-watt-hour, it isn’t, not yet. What it does is move orbit from speculative to evaluable. Infrastructure investors and enterprise architects can now model against a real company with a funded deployment roadmap.

The $355M total doesn’t include launch costs for operational scale, and Cowboy Space hasn’t disclosed its deployment timeline publicly. Those are the two variables that will determine whether orbital compute can actually compete with fuel cell on cost by 2028.

The Competitive Landscape

Cowboy Space isn’t alone in the orbital compute category. True Anomaly, which closed a reported new round in May 2026 coverage, operates in the defense-adjacent orbital compute space. Overview Energy, which signed the reported 1GW space-based solar agreement with Meta, is in the infrastructure supply chain for orbital power. The category is forming around three distinct layers: compute hardware in orbit (Cowboy Space’s play), orbital power generation for terrestrial transmission (Overview Energy’s play), and defense-specific orbital compute (True Anomaly’s play). These aren’t direct competitors, they’re different bets on which part of the orbital value chain captures the most value.

Index Ventures chose the compute hardware layer. That’s a specific read on where the margin lives.

What the Baiju Bhatt Factor Actually Means

It’s worth naming directly: Bhatt’s Robinhood pedigree doesn’t guarantee Cowboy Space succeeds, but it meaningfully reduces the execution risk discount that early-stage deep tech usually carries. Robinhood went from zero to regulated financial infrastructure at consumer scale. The operational complexity of that, regulatory navigation, infrastructure reliability, user trust, is legitimately transferable to the challenge of deploying AI compute hardware in orbit.

The investor thesis isn’t just “orbit is interesting.” It’s “this is the operator who could actually pull it off.” That’s a different investment than backing a technically credentialed team with no prior scaling experience.

Compute Specs: What’s Verified and What Isn’t

TechCrunch reported, and the Wire included, that Cowboy Space modules are designed to provide approximately 1 megawatt of computing power via roughly 800 onboard GPUs. That figure didn’t appear in the SiliconANGLE page content that forms the primary verification basis for this brief, and TechCrunch wasn’t in the source verification report. Treat the 1MW/800-GPU figure as company-stated specifications, not independently verified hardware performance. The distinction matters for enterprise architects evaluating this as a serious option: vendor-stated compute density and actual delivered performance in LEO are different numbers, and no independent evaluation exists yet.

What Investors and Infrastructure Teams Should Track

The Series B validation is done. The open questions are operational.

First deployment date is the critical milestone. Everything about Cowboy Space’s investment thesis depends on demonstrating that orbital AI compute actually works at the module specs described. A successful first deployment, hardware in LEO, compute workloads running, power output confirmed, creates a comps anchor that every subsequent orbital infrastructure raise will price against.

Cost-per-token benchmarking against terrestrial alternatives is the second milestone. Once modules are operational, the question becomes whether orbital compute can reach cost parity with fuel cell solutions for async workloads. If it can by 2028, the addressable market expands significantly. If it can’t, orbital compute stays a premium niche for constrained geographies or defense workloads.

The third variable is the hyperscaler response. If Microsoft, Google, or Amazon evaluate orbital compute as a credible grid-alternative for overflow AI workloads, the market validation is complete. A hyperscaler infrastructure partnership announcement would be the signal that orbital compute has graduated from “serious startup” to “category.”

Don’t bet on a 2027 hyperscaler partnership. Do watch for the first operational deployment confirmation, that’s what converts this from a funded thesis to a proven infrastructure category.