This week marks a definitive pivot in the narrative of space-based infrastructure. We have moved past the era of speculative "what-ifs" and entered the era of hardware procurement and industrial-scale planning. The centr
Learning to build the future, together · cognitoriverdelta.ai
Space-based compute & power - 2026-W46
Category rollup2026-11Week of November 9, 2026Space-based compute & power · week 2026-W46: Nov 07 - Nov 13, 2026 · 1 subtopic(s) covered · 2122 words · expanded
Weekly Notebook: The Orbital Compute Race
Overview
This week marks a definitive pivot in the narrative of space-based infrastructure. We have moved past the era of speculative "what-ifs" and entered the era of hardware procurement and industrial-scale planning. The central storyline is the emergence of SpaceX’s "StarMind" initiative, which is no longer a mere concept but a burgeoning industrial program characterized by specific hardware designs—the "Kessel" rack—major supply chain movements, such as the €60M+ SINGULUS solar deal—and an aggressive roadmap that seeks to decouple AI scaling from the increasingly volatile and politically sensitive terrestrial energy landscape.
The week's developments reveal a profound tension between two competing visions of the future: the SpaceX "vertical integration" moonshot and the "terrestrial skepticism" represented by established cloud giants. While SpaceX is positioning itself as the sole provider of the critical infrastructure needed for an automated AI era—attempting to build a complete, vertically integrated stack from the launch vehicle (Starship) to the specialized AI hardware (Nvidia-integrated racks)—incumbent players like Google and AWS are signaling deep skepticism regarding the economic and technical viability of moving massive compute loads into orbit.
This tension is being driven by a convergence of three critical factors: the physical limitations of the terrestrial power grid, the growing political resistance to ground-based data centers (specifically the "NIMBY" pressure in regions like Texas), and the unprecedented cost-reduction curve of heavy-lift launch vehicles. Whether the "orbital cloud" becomes a reality depends on whether the logistical math of Starship—specifically its full reusability and "chopstick" catch success—can ever outrun the massive energy and thermal requirements of the next generation of AI models.
Space-based AI compute and power (orbital datacenters)
The transition from "proof of concept" to "industrial scale" is best exemplified by the technical specifics emerging this week regarding the StarMind program. StarMind is not merely a satellite project; it is a specialized architectural response to what analysts are calling the "terrestrial grid bottleneck." As AI models demand unprecedented levels of power, the mounting difficulty of scaling on Earth is driving a search for alternative environments where energy and compute can be decoupled from land-based constraints.
The Hardware: The "Kessel" Rack and Nvidia Integration The specifications being discussed for the StarMind system move well beyond traditional satellite bus designs. At the heart of this architecture is the "Kessel" rack, a custom-designed unit reportedly built to house the Nvidia Vera Rubin architecture. These racks are expected to be highly specialized, with reports indicating a 72-GPU configuration per rack to meet the extreme demands of AI inference and training. Analysts at Wccftech have even projected that by 2028, the industry could see as many as 10,000 of these Nvidia GPU racks deployed in orbit.
This level of compute density brings intense technical challenges, specifically regarding thermal management. Because space is a vacuum, traditional air cooling is impossible; the StarMind architecture must rely on advanced radiative cooling to manage the heat generated by high-density GPU clusters. The goal is to achieve massive scale, with some projections suggesting the StarMind system could eventually involve as many as 1 million orbital data center satellites to achieve "terawatt" scale capacity.
The Power: Solving the Energy Crisis via Space-Based Solar The "Power" component is the most critical and contested element of the orbital compute equation. For these datacenters to be viable, they require an astronomical energy input that the terrestrial grid is struggling to provide. SpaceX is moving aggressively into the procurement phase to address this, having reportedly signed a €60M+ deal with Germany's SINGULUS for solar cells.
The scale of these solar ambitions is staggering. Reports indicate an order for 100 gigawatts (GW) of solar equipment per year. This is intended to leverage the fact that solar power in space is significantly more consistent and efficient than on Earth. Elon Musk has made bold claims regarding this capacity, suggesting that 100GW of space-based solar capacity would match the output of approximately 800GW on Earth. This massive energy deployment is the primary answer to the "energy crisis" facing terrestrial AI developers, aiming to provide the continuous, high-density power required for sustained AI workloads.
The Competition: "Muscle" vs. the "Nervous System" The industry is dividing into two distinct strategic camps: those building the "muscle" (power and specialized racks) and those building the "nervous system" (connectivity and data transfer).
SpaceX is clearly the leader of the "muscle" approach, prioritizing massive solar power generation and specialized hardware like the Kessel rack to drive raw compute capacity. In contrast, Alphabet (Google) is pursuing its "Suncatcher" project, which appears to be a strategic play for high-speed, low-latency connectivity. Suncatcher is described as an 81-satellite cluster designed for extreme-proximity laser data exchange, with satellites positioned as close as 100-200 meters apart. The goal of Suncatcher is to solve the "data transmission bottleneck," aiming for link speeds in the tens of terabits per second. This suggests that while SpaceX wants to own the capacity to process, Google is focused on the infrastructure to move the resulting data.
Economic and Political Drivers The economic viability of these projects remains the industry's greatest "fault line." There is a massive disconnect between the optimistic projections of SpaceX supporters and the warnings from terrestrial incumbents. On one side, analysts like Jo Bhakdi argue that space-based compute will be cheaper than terrestrial options by 2028, provided Starship launch costs drop toward the $100/kg mark. On the other side, Google's energy leadership has issued a stark warning: space-based data centers could ultimately cost 100 times more than their Earth-based counterparts.
This economic debate is being accelerated by a shift in the political landscape. In the United States, the election of Governor Gina Hinojosa in Texas has brought rising political opposition to the construction of massive, energy-hungry terrestrial data centers. This "NIMBY" (Not In My Back Yard) pressure—and the threat of potential regulatory moratoriums—is acting as a massive tailwind for the orbital compute movement. If the path to scaling AI on Earth is blocked by local politicians and grid constraints, the "high ground" of orbit becomes a strategic necessity.
Finally, this is an international race. The transition to space is becoming a cornerstone of national AI strategies globally. South Korea has announced plans for its own orbital AI data center tests by 2028, and China’s "Threebody" constellation is reportedly already deploying AI processing capabilities in orbit, with claims of reaching 5 Peta-operations per second.
Cross-cutting themes
The most significant cross-cutting theme this week is the convergence of the "Energy-Compute-Transport" Triad. In traditional industrial sectors, energy, computing, and logistics are treated as separate value chains. In the orbital compute sector, these are being fused into a single, verticalized stack. To build an orbital datacenter, a provider cannot simply purchase a chip; they must simultaneously solve for power generation (via SINGULUS solar procurement), thermal management (radiative cooling), high-density hardware (Kessel/Nvidia racks), launch logistics (Starship reusability), and high-speed connectivity (Suncatcher/laser links). This level of vertical integration is the hallmark of the SpaceX approach and is likely the only way the economics could ever approach parity with Earth.
Another major theme is the shift from "Resource Scarcity" to "Deployment Speed" as the primary competitive metric. On Earth, the primary bottlenecks for AI are the availability of land and the capacity of the power grid. In the orbital sector, the bottleneck shifts entirely to "Launch Cadence" and "Deployment Scale." The viability of the entire sector is now tethered to the success of the SpaceX Starship program. If Starship achieves full reusability and the "chopstick" catch success, the cost of deploying compute drops precipitously. In this new paradigm, the winner is not necessarily the company with the best algorithm, but the company that can achieve the highest launch frequency to deploy hardware at scale.
Where sources agree
- The Terrestrial Grid Crisis: There is a broad consensus across analysts and industry leaders that the terrestrial power grid, combined with mounting political and regulatory constraints (NIMBYism), is becoming a fundamental obstacle to the continued scaling of AI.
- The Starship Dependency: Almost all sources agree that the economic feasibility of orbital compute is fundamentally contingent on the success, reusability, and cost-reduction of the SpaceX Starship launch vehicle.
- The Solar Advantage: Sources agree that the ability to leverage high-capacity, continuous solar energy in space—unhindered by atmospheric interference—is a primary driver for moving compute out of the atmosphere.
- The Internationalized Race: There is agreement that the race for orbital compute is a global endeavor involving significant activity from the US, China, and South Korea.
- The Strategic Necessity of Vertical Integration: Analysts agree that the move to space is being driven by a need to bypass the physical and political constraints currently choking terrestrial AI expansion.
Where sources disagree
- The Economics of Space: There is a massive, unresolved disagreement regarding the cost-efficiency of orbital vs. terrestrial compute. Some analysts, such as Jo Bhakdi, argue space will be cheaper by 2028; conversely, Google's energy head asserts that space-based centers will be 100x more expensive than Earth-based ones.
- Deployment Timelines: Timelines for "meaningful" deployment are wildly inconsistent. Estimates range from highly optimistic (Q3 2027 or as early as 2028) to deeply skeptical projections that suggest meaningful deployment won't occur until 2040 or 2050.
- The Scale of Ambition: Sources differ significantly on the projected scale of the industry. Some analysts focus on specific hardware counts (e.g., 10,000 GPU racks by 2028), while others discuss astronomical scales, such as the "Starmind" proposal for 1 million satellites or Jeff Lutz's prediction that 99.99% of all digital compute will move to space within 10 years.
- The Valuation Impact: Analysts are divided on how much this affects the parent companies. While Larry Goldberg claims that 80-90% of SpaceX's current valuation is tied to orbital data centers, other analysts view the program as just one of several major revenue drivers.
Numbers and claims to verify
- 100x cost increase: The claim from Google's energy head that orbital data centers will cost 100x more than terrestrial ones.
- 1 million satellites: The claim by Farzad Mesbahi regarding the scale of the "Starmind" system.
- 100GW solar capacity: The claim regarding the scale of the SINGULUS/SpaceX solar equipment orders.
- 100GW space solar matches 800GW on Earth: Elon Musk's claim regarding the comparative energy output of space-based solar.
- <$100/kg launch cost: The target launch cost that would serve as the economic catalyst for the sector.
- 99.999% reliability: The technical reliability claim for orbital compute.
- 10,000 NVIDIA GPU racks by 2028: The specific capacity projection from Wccftech analysts.
- 5 Peta-operations per second: The technical specification for China's "Threebody" constellation.
- $4.22 billion market reach by 2035: The market sizing projection from openPR.com.
- 9-10 GW computing capacity by 2027: The capacity projection from BestInTESLA.
Investment and strategic implications
The transition to orbital compute represents a fundamental shift in the definition of AI infrastructure. If successful, the "moat" for the next generation of AI companies will move from proprietary algorithms to the ownership of the physical, orbital assets that power them. For investors, this means that the "compute race" is increasingly becoming a "launch and energy race."
Strategically, companies are looking at space-based compute as a "hedge" against terrestrial instability. As power grids face increasing strain and political opposition to data centers grows—particularly in critical tech hubs like Texas—the ability to move compute to an environment with high energy density and minimal regulatory friction becomes a strategic necessity.
However, the extreme cost disparity noted by terrestrial incumbents (the "100x" problem) suggests that this remains a high-risk "all or nothing" play. The winners will likely be those who can achieve total vertical integration—controlling everything from the launch vehicle and the solar procurement to the specialized GPU rack design—to drive costs down faster than the terrestrial grid can be upgraded. This creates a potential "unassailable monopoly" for any player that can successfully master the Starship-to-StarMind pipeline.
What to watch next week
- Starship Flight Milestones: Any updates regarding the "chopstick" catch or Starship reusability testing will have immediate implications for the "orbital compute" economic model and the feasibility of the $100/kg target.
- Supply Chain Movements: Further news regarding the €60M+ SINGULUS solar deal or other massive procurement orders will serve as a proxy for how much SpaceX is actually investing in the StarMind program's immediate phase.
- Alphabet/SpaceX Relations: Given the reported conflict of interest and the recent exit of Google from the SpaceX board, watch for any further signs of tension or direct competition between the Suncatcher (connectivity) and StarMind (compute) initiatives.
- Regulatory Signals: Any new developments regarding data center moratoriums or political opposition in Texas will provide critical context on the "political driver" behind the move toward space-based infrastructure.
Sources
- AI on the Ballot | Stratechery by Ben Thompson — Ben Thompson, Nov 08
- AWS CEO Throws Cold Water on AI Data Centers in Space — Business Insider (via Google News), Nov 13
- Agnikul to set up Andhra Pradesh facility for orbital data centre units — Business Standard (via Google News), Nov 13
- Another WORLD FIRST for Tesla 🤯 Cybertruck might get a second lifeline 👀 NEW FSD Approval — BestInTESLA, Nov 12
- Blueshift outlines thermal plan for orbital data centres — DataCenterNews Asia Pacific (via Google News), Nov 12
- Could The Texas Freeze Actually Help SpaceX? — Brighter with Herbert, Nov 10
- Does SpaceX’s $2Trillion Bet have a Physics Problem? — Jo Bhakdi, Nov 07
- Elon Just Reposted This After Tesla's 11% Crash — Brighter with Herbert, Nov 08
- Elon Musk & Ron Baron Stun (highlights from today) — Steven Mark Ryan, Nov 07
- Elon Musk & SpaceX’s Next-Gen Space AI Satellites Could Boost Peak Compute To 400kW, Says — Wccftech (via Google News), Nov 10
- Elon Musk Says SpaceX’s Orbital AI Supercomputers Launching in 2027 Will ‘Blow People’s Mi — Engineerine (via Google News), Nov 08
- Elon Musk Stuns Billionaire — Farzad Mesbahi, Nov 07
- Elon’s Insane Super Intelligence Plan — Farzad Mesbahi, Nov 12
- Google Owns $74 Billion of SpaceX. So Why Leave? — Brighter with Herbert, Nov 11
- Google energy chief says space data centers could cost up to 100 times more — Mezha.net (via Google News), Nov 08
- Google energy head: space data centers would cost up to 100x more than on Earth — Hacker News, Nov 08
- How Google Plans to Build Data Centres in Space Using AI Satellites? — Analytics Insight (via Google News), Nov 07
- How SpaceX Becomes the Most Profitable Company Ever — Farzad Mesbahi, Nov 09
- How big could this really get? #Tesla #TSLA #TeslaStock #TeslaInvestors #StockMarket — Randy Kirk, Nov 11
- Is Nvidia Stock a BARGAIN after Jensen Huang’s resignation? — Jo Bhakdi, Nov 12
- It's Not as Simple as Putting Data Centers in Space (video) — Hacker News, Nov 11
- Just a moonshot? Google energy head claims data centers in space would cost up to 100x mor — TechRadar (via Google News), Nov 08
- Musk Backs Secondhand Page Claim of 99.999% Orbital Compute — FourWeekMBA (via Google News), Nov 08
- Musk Bets on 'Space Data Centers' to Win AI Race, Opening New Fronts With Terafab and Opti — BigGo Finance (via Google News), Nov 08
- NEW INTERVIEW: Elon Just Put Odds on $1 Trillion SpaceX Revenue — Brighter with Herbert, Nov 07
- NVIDIA WITHOUT JENSEN? — Jo Bhakdi, Nov 09
- OwlO Cloud emerges from stealth promising European scalable orbital compute & AI — Data Center Dynamics (via Google News), Nov 13
- SPCX Secures Singulus Deal To Scale Starmind Space Data Center Satellites — Yahoo Finance (via Google News), Nov 10
- STARLINK IS THE UNDERSUNG HERO — Peter H. Diamandis, Nov 07
- Satlyt CEO Rama Afullo: Orbital Data Centers Already Pay Off, Telcos Face 10-Year Extincti — BigGo Finance (via Google News), Nov 12
- South Korea plans first orbital AI data center test for 2028 — digitimes (via Google News), Nov 10
- Space-Based Data Center Market to Reach US$4.22 Billion by 2035 | — openPR.com (via Google News), Nov 09
- SpaceX CEO Elon Musk Says 100GW Of Solar Power In Space Could Match 800GW On Earth For Sta — TradingView (via Google News), Nov 12
- SpaceX CEO Elon Musk Says 100GW Of Solar Power In Space Could Match 800GW On Earth For Sta — Yahoo Finance (via Google News), Nov 12
- SpaceX Orders 6,000 Racks for Space Data Center, Plans 1-GW Deployment — TOKENPOST (via Google News), Nov 08
- SpaceX Rises Over 2% Pre-Market: Musk Says 100GW Annual Capacity Solar Cell Equipment Orde — TradingKey (via Google News), Nov 11
- SpaceX Signs €60M+ SINGULUS Deal for Starmind Satellite Solar Cells - BASENOR — Tesla Accessories (via Google News), Nov 10
- SpaceX Stock Fell Hard. SpaceX Business Kept Winning! — Brighter with Herbert, Nov 09
- SpaceX Taps Germany's SINGULUS for 100 GW-a-Year Solar Line to Power Starmind Orbital Data — BigGo Finance (via Google News), Nov 11
- SpaceX developing custom Nvidia Vera Rubin rack dubbed Kessel for its orbital data centers — Data Center Dynamics, Nov 12
- SpaceX is designing Kessel, a 72-GPU Nvidia Rubin rack for data centers in space, report s — MLQ.ai (via Google News), Nov 11
- SpaceX signs major SINGULUS contract for Starmind solar cell production — Drive Tesla Canada (via Google News), Nov 10
- SpaceX’s AI Satellite Plan Picks Pricey Solar Panels For 100GW Compute Megaplan That Elon — Wccftech (via Google News), Nov 11
- SpaceX’s Texas factory to make solar cells for AI satellites - San Antonio Express-News — Google News, Nov 12
- Tesla Isn’t Pretending Anymore | They're Just Not Doing It — Ryan Shaw, Nov 13
- Tesla and SpaceX Stock Moving Towards Reality Fast — Randy Kirk, Nov 09
- The next wave of mission-critical connectivity: Orbital compute — PwC (via Google News), Nov 09
- Three HUGE Catalysts for Tesla and SpaceX say Larry G — Randy Kirk, Nov 10
- Welcome to November 9, 2026 — Dr. Alex Wissner-Gross, Nov 09
- What Tesla’s Robotaxi And Optimus Do To A Paycheck — Cern Basher, Nov 13
- When Elon Solves This, It's Game Over — Farzad Mesbahi, Nov 12
- Why China's AI Satellites Think Before They Phone Home — Farzad Mesbahi, Nov 12
- Why Google Wants Its AI Satellites Flying 200 Meters Apart — Farzad Mesbahi, Nov 11
Informational analysis synthesized by AI from sourced, dated material, curated by a human. Treat specific claims as unverified until checked. Not financial advice.