Cowboy Space joins the FCC queue for grid-free orbital compute forecast to scale in the 2030s

The Stampede filing is getting forwarded this week, usually to someone holding a utility interconnection date that keeps moving. The note on top asks whether the plan changes. In the application, Cowboy Space COO Joseph Yaffe wrote that Stampede can "bypass the costs and delays associated with building data centers on Earth" ^[1].

The filings pitch hyperscale capacity that never touches the grid, while a company that sells orbital compute expects a smaller first market. "Over the next three to five years, I expect early revenue to be driven primarily by defense and government use cases, with commercial adoption following as the infrastructure matures," Sophia Space co-founder and CEO Rob DeMillo told DCD ^[9]. He put the multi-billion-dollar version on "a five- to ten-year horizon" and compared it to the way "terrestrial cloud evolved from specialized workloads into general-purpose infrastructure" ^[10].

The most specific date comes from Andrew Cavalier, principal analyst at ABI Research. He groups SpaceX, Starcloud and Blue Origin as orbital hyperscalers aiming for "tens of thousands of satellites" ^[11]. "Based on historical execution timelines, we expect gigawatt-scale capacity will be reached in the early 2030s," he said ^[12]. Even on that forecast, gigawatt scale arrives after the decade Altman was describing ^[12][6].

DCD's own analysis calls the feasibility of orbital data centers "highly questionable," even as investment and political will push the idea into "iteratively more pragmatic shapes than the astral data centers initially imagined" ^[7]. Satellite counts in the filings vary widely. SpaceX's is 50 times Stampede's ^[8]. The passages of Cowboy Space's application that DCD quotes include a satellite count and no power figure ^[2][1]. A planner books against power.

A planner can sort this by asking whether the capacity is needed before 2030, and whether the buyer is a government agency with a specialized workload. General AI compute needed before 2030 belongs in the utility plan, and the filings leave that plan where it was ^[12]. If the general compute is needed after 2030, the filings go on a watch list, checked against Cavalier's early-2030s date ^[12]. A government buyer with specialized work is the customer DeMillo expects to pay first, on either timeline, and that box is the only one where a pilot conversation makes sense now ^[9]. Commercial specialized work before 2030 has to wait for the commercial adoption DeMillo places behind government ^[9].

I'd keep orbital compute out of any power plan due this decade. Teams with government workloads bear the tradeoff. By DeMillo's account they are the first market, and a planner who dismisses the filings outright gives up the earliest pilots ^[9].