Elon Musk said “It takes us ~2 months working 24/7 to get a large compute cluster working reliably” at SpaceX, and added: “I’d guess closer to 6 months for others.” By his own two figures that is 3 times faster, by our arithmetic. He made the claim in a post on X at 19:01 UTC on 11 November, arguing that “actually usable” compute is “an important metric”.
He was replying to the account @zerohedge, which had shared a chart of how much of each developer’s data-centre pipeline Bernstein counts as credible. Across the ten developers on that chart, the credible figures add up to 111 GW, by our arithmetic, with Amazon at 33 GW, about 19 times the 1.7 GW of compute SpaceX’s AI segment reported at the end of September.
The full post opens: “SpaceX has a chance of going about half the GW next year of actually usable compute.” Musk then calls “actually usable” “an important metric” and gives the two- and six-month figures. The six-month figure is his own guess about other operators, and he names none of them.
The first sentence is the harder one to read. Musk does not say what “the GW” refers to: the chart he replied to, a single developer on it, or a figure of his own. Nor does he say whether “half the GW” means half a gigawatt. We do not turn it into a SpaceX capacity number.
Each developer’s pipeline in the Aterio database against the part counted as credible, as shown in a chart shared by @zerohedge and attributed to Aterio and Bernstein. SpaceX’s 1.7 GW is installed compute from its 10-Q. The 111 GW total and the 19x ratio are our arithmetic.
Business Pill · Lead Time
An evergreen Business Engineer explainer on lead time: why the wait is part of the price. It teaches the general idea only and does not discuss SpaceX or the developers named here.
The key insight: As we read it, Musk is moving the measure from gigawatts announced to gigawatts that work. He says SpaceX takes about two months to get a large cluster running reliably, against his guess of about six for others. The chart he answered counts 111 GW of credible pipeline across ten developers, by our arithmetic, all of it capacity that its footnote says is not yet active.
What Was Posted on X
@zerohedge SpaceX has a chance of going about half the GW next year of actually usable compute.
“Actually usable” is an important metric. It takes us ~2 months working 24/7 to get a large compute cluster working reliably. I’d guess closer to 6 months for others.
Elon Musk’s reply to @zerohedge at 19:01 UTC on 11 November says SpaceX has a chance of going about half the GW next year of actually usable compute, and that it takes SpaceX about 2 months to get a large compute cluster working reliably, against his guess of closer to 6 months for others. This piece quotes it and does not turn ‘half the GW’ into a SpaceX figure.
How much of each developer’s data center pipeline actually gets built
Share of forward pipeline ranked by Bernstein as credible: META, MSFT, GOOGL, QTS, Vantage are top 5
The post Musk replied to, from @zerohedge at 18:15 UTC, charts the share of each developer’s forward pipeline that Bernstein counts as credible, with a pipeline average of 34%.
Data Centers in pipeline, by company: Amazon 33GW credible Google 20GW credible Meta 15GW credible MSFT 10GW credible QTS 9GW credible Stack Infra 7GW credible Vantage 6GW credible Digital Realty 4GW credible etc
An earlier @zerohedge post at 18:14 UTC lists credible pipeline by company, from Amazon at 33 GW to Digital Realty at 4 GW, with a chart sourced to Aterio, Bernstein Analysis and Estimates. We did not find Bernstein’s report itself, so the piece attributes these figures to the chart.
The Pipeline Chart He Replied To
Zerohedge’s first post listed “Amazon 33GW credible”, “Google 20GW credible”, “Meta 15GW credible” and “MSFT 10GW credible”, followed by QTS at 9 GW, Stack Infra at 7 GW, Vantage at 6 GW and Digital Realty at 4 GW. Its chart, headed “In Dataset & Credible Capacity Top 10 Developers”, gives its source as “Aterio, Bernstein Analysis and Estimates” and adds Prologis at 4 GW and Crusoe at 3 GW.
That chart counts only capacity that is under construction, announced or in a land bank. Its footnote says it “Excludes Active and Cancelled”. Amazon has 46 GW in the database and 33 GW counted as credible, a completion rate of 72%. Meta’s rate is 86%, Microsoft’s 85% and Google’s 83%, while Crusoe’s is 45%.
A second zerohedge post ranked the share of each pipeline “Bernstein counts as credible” and marked a “Pipeline average: 34%”. The lowest of the ten largest low-credibility pipelines it showed was O’Leary Ventures, at 3%. We did not find Bernstein’s report itself, so these figures are as shown in the charts shared by zerohedge, attributed there to Bernstein and Aterio.
Where SpaceX Sits Against It
Amazon alone is about 30% of the 111 GW the ten developers have in credible pipeline, by our arithmetic, and Amazon, Google, Meta and Microsoft together are 70%.
SpaceX is not on the chart. Its AI segment, SI, reported 1.7 GW of nameplate compute at the end of September, up from 0.6 GW a year earlier, 2.8 times in a year, as we set out in SpaceX vs CoreWeave: Both Near 1.7 GW, 2.1x the AI Revenue. CoreWeave reported over 1.7 GW of active power in its results release of 10 November.
The two sets of numbers measure different things. Bernstein’s figures, as charted, are future capacity that has not yet been switched on. SpaceX’s 1.7 GW is installed compute. Musk’s point is about a third stage: how long it takes, once the hardware is in place, before a cluster runs reliably.
The Structural Read
As we read the exchange, the chart and the post measure three different stages: what developers have announced, what Bernstein counts as likely to be built, and, in Musk’s framing, how soon installed hardware becomes usable. Only the last is SpaceX’s claim, and it rests on his own figures.
“It takes us ~2 months working 24/7 to get a large compute cluster working reliably. I’d guess closer to 6 months for others.”
Three Implications
FOR SPACEX Its AI segment reported 1.7 GW of nameplate compute at the end of September, 2.8 times a year earlier, by our arithmetic; the two-month figure is Musk’s, not a filed number.
FOR THE LARGEST BUILDERSAmazon has 33 GW counted as credible, about 30% of the ten developers’ 111 GW, by our arithmetic, at a completion rate of 72% in the chart.
FOR ANYONE READING PIPELINE NUMBERS The second chart zerohedge shared marks a pipeline average of 34%, with O’Leary Ventures at 3%.
The framework says: “Companies that solve the power equation will dominate the AI era; those that don’t will watch their chips sit idle.”
As we read it, Musk’s two-month claim is about the same idle-chip risk, seen from inside the data hall rather than at the grid connection. The framework predates the post and does not discuss it.
What Is Not Established
Musk gives no source for the six-month figure, and no definition of when a cluster counts as working reliably. He does not say which clusters the two-month figure covers. We have not seen either figure in a SpaceX filing.
We could not read Bernstein’s report itself. We did not contact SpaceX, Bernstein or the developers named for this piece.
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.
Musk says SpaceX gets a large cluster working reliably in about two months, against his guess of about six for others, 3 times faster by his own figures. He made the claim in reply to a chart, attributed to Bernstein and Aterio, on which ten developers hold 111 GW of credible pipeline, by our arithmetic.
A note on sourcing. We read Elon Musk’s 11 November post on X in full, and the two posts by @zerohedge it replies to, with the charts they carry. The developer figures are as shown in those charts, which attribute them to Aterio and Bernstein; we did not find Bernstein’s report itself. SpaceX’s 1.7 GW and CoreWeave’s capacity are from the filings and release covered in our earlier piece. The totals, shares and ratios are our arithmetic. We did not contact SpaceX, Bernstein or the developers named.
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.
