<?xml version="1.0" encoding="UTF-8"?><rss version="2.0" xmlns:content="http://purl.org/rss/1.0/modules/content/"><channel><title>Kardashev Weekly</title><description>A five-minute Monday read on the race to put AI computers in orbit.</description><link>https://kardashevweekly.com</link><language>en-us</language><managingEditor>hello@kardashevweekly.com (William Aldrich)</managingEditor><item><title>Kardashev Weekly</title><link>https://kardashevweekly.com/issues/01</link><guid isPermaLink="true">https://kardashevweekly.com/issues/01</guid><description>A five-minute Monday read on the race to put AI computers in orbit.</description><pubDate>Mon, 23 Nov 2026 00:00:00 GMT</pubDate><content:encoded>&lt;p&gt;Welcome to issue one. The name comes from the &lt;a href=&quot;https://www.cnn.com/2026/07/10/science/elon-musk-kardashev-scale&quot;&gt;Kardashev scale&lt;/a&gt;, a 1960s idea that ranks civilizations by the energy they can harness. A Type I civilization uses a planet’s energy. A Type II civilization uses its star’s. When SpaceX asked the FCC for up to a million orbital data-center satellites, it called the plan a “&lt;a href=&quot;https://docs.fcc.gov/public/attachments/DA-26-113A1.txt&quot;&gt;first step towards becoming a Kardashev II-level civilization — one that can harness the Sun’s full power&lt;/a&gt;.” This newsletter will follow the companies trying to do it, and the money and physics that will decide whether they can.&lt;/p&gt;
&lt;h2 id=&quot;filings-watch&quot;&gt;Filings Watch&lt;/h2&gt;
&lt;p&gt;None of the big applications has been granted.&lt;/p&gt;
&lt;p&gt;SpaceX’s &lt;a href=&quot;https://www.fcc.gov/document/sb-accepts-filing-spacexs-application-orbital-data-centers&quot;&gt;SAT-LOA-20260108-00016&lt;/a&gt;, &lt;a href=&quot;https://docs.fcc.gov/public/attachments/DA-26-113A1.txt&quot;&gt;accepted for filing on February 4&lt;/a&gt;, seeks authority for up to one million satellites between 500 and 2,000 kilometers, linked by lasers through Starlink, with Ka-band radio as backup for tracking and control.&lt;/p&gt;
&lt;p&gt;Blue Origin’s &lt;a href=&quot;https://spacenews.com/blue-origin-joins-the-orbital-data-center-race/&quot;&gt;Project Sunrise&lt;/a&gt;, filed March 19, asks for up to 51,600 data-center satellites in sun-synchronous orbits from 500 to 1,800 kilometers. A September amendment, &lt;a href=&quot;https://fcc.report/IBFS/SAT-AMD-20260917-00370&quot;&gt;SAT-AMD-20260917-00370&lt;/a&gt;, is still pending review.&lt;/p&gt;
&lt;p&gt;Starcloud of Redmond, Washington, has &lt;a href=&quot;https://docs.fcc.gov/public/attachments/DOC-419509A1.txt&quot;&gt;SAT-LOA-20260202-00073&lt;/a&gt;, accepted in March, for up to 88,000 satellites as a “distributed datacenter in space” between 600 and 850 kilometers. On November 12, the &lt;a href=&quot;https://space.commerce.gov/osc-announces-the-space-commerce-certification-pilot-phase-cohort/&quot;&gt;U.S. Office of Space Commerce named Starcloud&lt;/a&gt; one of four companies in a certification pilot for novel missions, including orbital data centers — a process, not a license to fly.&lt;/p&gt;
&lt;h2 id=&quot;money-moves&quot;&gt;Money Moves&lt;/h2&gt;
&lt;p&gt;&lt;a href=&quot;https://stockanalysis.com/stocks/spcx/&quot;&gt;SpaceX stock (SPCX)&lt;/a&gt; closed Friday, November 20, at $127.79, down 7.69% and below its &lt;a href=&quot;https://techcrunch.com/2026/06/11/spacex-officially-prices-shares-at-135-in-the-largest-ipo-ever/&quot;&gt;$135 IPO price of June 11&lt;/a&gt;. The 52-week range is $104.83 to $225.64. The drop came in a broader technology selloff, a day after reports that lenders on a roughly &lt;a href=&quot;https://arstechnica.com/space/2026/11/spacex-lenders-dont-want-elon-musk-launching-their-collateral-into-space/&quot;&gt;$43 billion “Project Saturn” debt package&lt;/a&gt; — more than $30 billion of it sold November 19, including a $25 billion bond backed by a five-year Nvidia-chip lease — had sought to keep that collateral on Earth. SpaceX has said those chips are for terrestrial data centers. Whether a no-orbit clause made the final documents is unclear.&lt;/p&gt;
&lt;p&gt;On November 10, Germany’s &lt;a href=&quot;https://www.singulus.com/2026/11/10/singulus-spacex-supply-contract/&quot;&gt;SINGULUS TECHNOLOGIES signed a contract&lt;/a&gt; worth more than €60 million for a turnkey heterojunction solar-cell line, due in the first half of 2027 and meant to scale to more than 100 gigawatts of equipment at SpaceX’s Gigasat Factory in Bastrop, Texas. That is equipment capacity, not 100 gigawatts of compute next year. In August, &lt;a href=&quot;https://nvidianews.nvidia.com/news/spacexai-adopts-nvidia-vera-cpu-to-accelerate-agentic-ai-at-massive-scale&quot;&gt;NVIDIA said&lt;/a&gt; the first Starmind satellite will be based on an optimized Vera Rubin NVL72 system, adapted for orbit.&lt;/p&gt;
&lt;p&gt;As of November 19, three space ETFs held SpaceX: &lt;a href=&quot;https://stockanalysis.com/etf/ufo/holdings/&quot;&gt;Procure Space (UFO)&lt;/a&gt; at about 15.7%, &lt;a href=&quot;https://stockanalysis.com/etf/arkx/holdings/&quot;&gt;ARK Space &amp;#x26; Defense (ARKX)&lt;/a&gt; at about 12.0%, and &lt;a href=&quot;https://stockanalysis.com/etf/nasa/holdings/&quot;&gt;Tema Space Innovators (NASA)&lt;/a&gt; at 9.0% in public shares plus 17.8% in a SpaceX SPV, or roughly 27% combined.&lt;/p&gt;
&lt;h2 id=&quot;launch-board&quot;&gt;Launch Board&lt;/h2&gt;
&lt;p&gt;&lt;strong&gt;Launched.&lt;/strong&gt; &lt;a href=&quot;https://blog.google/innovation-and-ai/models-and-research/google-research/project-suncatcher-prototype/&quot;&gt;Google’s Project Suncatcher prototype&lt;/a&gt;, built with Planet Labs, flew on SpaceX’s Transporter-18 mission on October 1. On October 30, Google &lt;a href=&quot;https://blog.google/innovation-and-ai/models-and-research/google-research/project-suncatcher-satellite-response/&quot;&gt;ran an onboard Gemma 4 model&lt;/a&gt; and got the first response from a data-center-class TPU in orbit — a haiku, downlinked over Svalbard, Norway. Next: thermal and radiation tests, then two linked satellites in 2027.&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;Promised, not flown.&lt;/strong&gt; SpaceX has not launched a Starmind satellite. The &lt;a href=&quot;https://www.spacex.com/spacexai/starmind&quot;&gt;official AI1 design&lt;/a&gt; has a 75-meter wingspan and up to 250 kilowatts of peak compute (175 kW average). The Gigasat Factory in Bastrop is supposed to start building them in the thousands “as soon as late 2027.” On November 6, Musk &lt;a href=&quot;https://www.investmentnews.com/equities/elon-musk-touts-data-centers-in-space-at-baron-investment-conference/268925&quot;&gt;guessed&lt;/a&gt; first launches “in the 3rd quarter of next year,” then “significant volume probably in ‘28” and “very, very large volume in ‘29.”&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;The rocket the thesis needs.&lt;/strong&gt; Starmind depends on cheap, fully reusable Starship flights. On November 13–14, &lt;a href=&quot;https://aviationweek.com/space/operations-safety/starship-flight-15-ends-first-ship-tower-catch&quot;&gt;Flight 15 caught the upper stage in the tower’s arms&lt;/a&gt; for the first time. The Super Heavy booster catch was aborted after a fuel shortfall.&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;Booked.&lt;/strong&gt; On November 19, Google &lt;a href=&quot;https://blog.google/innovation-and-ai/models-and-research/google-research/blue-origin-project-suncatcher-partnership/&quot;&gt;partnered with Blue Origin&lt;/a&gt; for Suncatcher’s first heavy-lift launch: New Glenn, targeted for the first quarter of 2029, carrying satellites “each with roughly the capacity of a data center rack.” Starcloud &lt;a href=&quot;https://investors.rocketlabcorp.com/news-releases/news-release-details/rocket-lab-lands-over-515m-neutron-launch-deals-constellation&quot;&gt;booked a Rocket Lab Neutron flight for 2029&lt;/a&gt; and a &lt;a href=&quot;https://investors.fireflyspace.com/news-releases/news-release-details/firefly-aerospace-signs-starcloud-commercial-customer&quot;&gt;Firefly&lt;/a&gt; lunar-orbit demo no earlier than 2028, following a recent low-Earth-orbit AI demonstration. Project Sunrise has no public launch date.&lt;/p&gt;
&lt;h2 id=&quot;deep-dive-can-you-really-cool-a-gpu-in-space&quot;&gt;Deep Dive: Can you really cool a GPU in space?&lt;/h2&gt;
&lt;p&gt;Space is cold, which makes “free cooling” sound obvious. It isn’t. There is no air in orbit, so there is no breeze. Heat can leave a satellite in only one useful way: as infrared light, radiated from a surface into the void.&lt;/p&gt;
&lt;p&gt;That follows the Stefan-Boltzmann law: the power you can reject scales with the radiator’s area and with its temperature to the fourth power. Area is expensive: every square meter has to be launched. &lt;a href=&quot;https://spectrum.ieee.org/orbital-data-centers-heat&quot;&gt;IEEE Spectrum&lt;/a&gt; estimated that a single 700-watt chip like an Nvidia H100 would need about 1.4 square meters of radiator to stay at 60°C, if the panel faces deep space at 3 kelvins. A roughly 40-kilowatt AI rack would need about 80 square meters — roughly a pickleball court. After five years in low Earth orbit, coating damage can push that to about 2.0 square meters per chip, a 40% penalty you have to launch on day one.&lt;/p&gt;
&lt;p&gt;SpaceX &lt;a href=&quot;https://www.spacex.com/spacexai/starmind&quot;&gt;says&lt;/a&gt; that with no chillers, cooling towers, or fans, its AI satellites “reduce cooling power overhead by an order of magnitude.” Google, already cooling TPUs in a vacuum chamber and now in orbit, calls this “&lt;a href=&quot;https://blog.google/innovation-and-ai/models-and-research/google-research/google-project-suncatcher-facts/&quot;&gt;a crucial research challenge&lt;/a&gt;” and is using heat pipes to move heat to radiators. Space saves the electricity a terrestrial data center spends on mechanical cooling. It does not abolish heat. It turns waste heat into area, mass, and launch cost.&lt;/p&gt;
&lt;h2 id=&quot;my-take&quot;&gt;My Take&lt;/h2&gt;
&lt;p&gt;The constraint is energy, not slogans. Earth is already arguing over every new gigawatt of data-center power. Space offers continuous sunlight and a vacuum that radiates heat away — if you can lift the radiators and fly the rockets cheaply enough that the “physics tax” does not eat the savings. That is why last weekend’s Starship catch matters more than Friday’s stock drop, and why a €60 million solar-cell line may tell you more than a filing for a million satellites. Google has a real TPU in orbit. SpaceX has the factory, the chips, and the launch system. Neither has a data center in the sky yet. Watch the companies that treat cooling, launch cost, and licenses as engineering problems, not press releases.&lt;/p&gt;
&lt;hr&gt;
&lt;p&gt;This is not financial advice. Do your own research.&lt;/p&gt;
&lt;p&gt;Subscribe at &lt;a href=&quot;https://kardashevweekly.com&quot;&gt;kardashevweekly.com&lt;/a&gt;.&lt;/p&gt;</content:encoded><author>hello@kardashevweekly.com (William Aldrich)</author></item></channel></rss>