SpaceX “plans to purchase chip packages from Nvidia and procure the rest directly, bypassing Nvidia’s L6 compute boards”, for a custom rack called Kessel, research firm SemiAnalysis wrote on X on 11 November. SemiAnalysis said that “would give SpaceX more control over supplier selection and component sourcing”.
SemiAnalysis said Kessel “specifies an 84-layer M10 midplane, versus 156 layers of M9 planned for the earlier Kyber design”, in the same thread. That is 72 fewer layers, or 46% fewer, by our arithmetic. SpaceX has not commented, and the firm’s full note is behind its institutional paywall; we have not read it.
Business Pill · Vertical Integration
An evergreen Business Engineer explainer on vertical integration and stacked margins. It teaches the general idea only and does not discuss SpaceX, Nvidia or Kessel.
The key insight: SemiAnalysis says SpaceX plans to buy chip packages from Nvidia and source the rest of its custom Kessel rack directly, bypassing Nvidia’s L6 compute boards. The firm says Kessel specifies an 84-layer M10 midplane, versus 156 layers of M9 planned for Nvidia’s Kyber: 72 fewer layers, or 46% fewer, by our arithmetic. SpaceX has not commented.
The thread calls Kessel “a custom Vera Rubin NVL72 rack that replaces internal NVLink cable cartridges with a passive PCB midplane”. It “keeps Oberon’s 72 GPU packages, 18 compute trays and 9 switch trays, but rearranges them back-to-back around the board”, SemiAnalysis wrote.
The firm called it “mini Kyber”, “with a smaller scale-up domain and a design intended for both terrestrial and orbital deployment”. On the layer count, it said “Fewer layers give it a more feasible manufacturing path, although the design requires a higher-grade material”. Kyber is Nvidia’s Rubin Ultra NVL576 rack, planned for the second half of 2027, as we reported.
PCB midplane layers: 84 (M10) for SpaceX’s Kessel versus 156 (M9) planned for Nvidia’s Kyber, as reported by SemiAnalysis on X on 11 Nov 2026. 72 fewer layers, 46% fewer, is our arithmetic. SpaceX has not commented.
What Was Posted on X
SpaceX is developing Kessel, a custom Vera Rubin NVL72 rack that replaces internal NVLink cable cartridges with a passive PCB midplane. It keeps Oberonโs 72 GPU packages, 18 compute trays and 9 switch trays, but rearranges them back-to-back around the board. Think mini Kyber, with a smaller scale-up domain and a design intended for both terrestrial and orbital deployment. The question is whether this smaller version can make the midplane practical. (1/4)๐งต
SemiAnalysis opened its four-post thread on 11 November by describing Kessel as a custom Vera Rubin NVL72 rack that replaces internal NVLink cable cartridges with a passive PCB midplane, keeping 72 GPU packages, 18 compute trays and 9 switch trays, for terrestrial and orbital use. Later posts in the thread give the 84-layer midplane figure and the plan to buy chip packages from Nvidia, both quoted in this piece.
Who Buys the Parts
The supply-chain change is the new detail. Our 7 November piece covered FundaAI’s report that SpaceX had ordered 6,000 space-specific racks for 2027 from a server ODM, and FundaAI described Kessel as a space version of NVL72 compressed into “a ~9U flat enclosure”. SemiAnalysis now says SpaceX would buy the chip packages from Nvidia itself and source the rest.
At 72 GPU packages per rack, 6,000 racks would mean 432,000 packages, by our arithmetic, matching the 432,000 GPUs FundaAI cited. SpaceX has not confirmed either report. Separately, Bloomberg reported that an Apollo-backed entity would buy about $47.3 billion of Nvidia GPUs and equipment for SpaceX, with a $7 billion SpaceX prepayment, as we reported on 4 November.
For scale on the assembly layer: rack maker Supermicro reported a 12.7% GAAP gross margin on $19.9 billion of fiscal first-quarter revenue, as we reported. As we read it, buying packages and sourcing the rest is the route that lets a large buyer decide who captures that kind of integration margin. SemiAnalysis does not discuss margins in the thread.
Where It Fits in SpaceX’s Compute Plans
SpaceX has ordered a solar cell line from Singulus for its Starmind satellites (our coverage), and a City of Starbase notice puts about $24 billion on a SpaceX gas plant and super-compute facility (our coverage). Its request for up to 1 million orbital data center satellites now carries AWS chief Matt Garman’s view, per Alex Heath of Sources, that space data centers are out of reach for another decade or two.
The Structural Read
As we read it, the change is about who decides the bill of materials. SemiAnalysis says buying packages and procuring the rest would give SpaceX more control over supplier selection and component sourcing; the thread does not say which suppliers gain or lose, and the full note that does is paywalled.
SemiAnalysis, on X, 11 Nov 2026
“Fewer layers give it a more feasible manufacturing path, although the design requires a higher-grade material.”
Three Implications
FOR NVIDIA SemiAnalysis says SpaceX plans to purchase chip packages from Nvidia while bypassing Nvidia’s L6 compute boards; the thread gives no volumes or prices.
FOR RACK AND BOARD MAKERS SemiAnalysis says SpaceX would procure the rest directly; its paywalled note covers which suppliers stand to benefit, and the thread names none.
FOR SPACEX FundaAI reported 6,000 space-specific racks for 2027; at 72 GPU packages per rack that is 432,000 packages, by our arithmetic. SpaceX has confirmed neither report.
The framework says: “the seven-layer AI stack is now governed from the bottom — energy, silicon, compute capacity.”
As we read it, a buyer that specifies its own rack and buys chip packages directly is reaching into the silicon and compute layers of that stack. The framework predates this report and does not discuss SpaceX or Kessel.
What Is Not Established
SpaceX has not confirmed Kessel, the package purchases or FundaAI’s 6,000-rack order. The thread does not name the board makers or suppliers; SemiAnalysis says its paywalled note covers “which suppliers stand to benefit”. Volumes, prices and timing for Kessel are not given in the thread.
This Week In The AI Supercycle
The Great Convergence of Capital, Power, and Geopolitics
This week’s issue: for two years the AI build-out was priced as a supply story, and last week as a credit story. This week it was priced as a permission story, and the answer arrived as a cost, not a ban.
SemiAnalysis says SpaceX’s Kessel rack would use an 84-layer midplane, 46% fewer layers than Nvidia’s planned Kyber by our arithmetic, and that SpaceX would buy chip packages from Nvidia and source the rest itself, bypassing Nvidia’s L6 compute boards. SpaceX has not commented.
A note on sourcing. Every Kessel detail and quotation comes from SemiAnalysis’s four-post thread on X of 11 November 2026; its full note is paywalled and was not read. Other figures come from earlier FourWeekMBA reporting, linked in the text. Ratios are our arithmetic.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
