Who pays for the AI build-out was again the day’s biggest thread on 25 October 2026, by our grouping of the news. Elon Musk wrote on X on 25 October: “SpaceX closed $40B of debt for GPUs, but we’re investment grade and have Starlink & launch revenue.” SpaceX, Goldman Sachs and Apollo had not published a statement on the close that we could find. Meta’s preliminary information statement, filed with the SEC on 21 October, says its financial adviser told a board committee in June that the most Meta could raise in equity was about $700 billion over a number of years without Mark Zuckerberg’s voting power falling below 50%.
The threads below are who pays for the build; sold out, and priced to rise; power, materials and the supply chain; agents and the race to reach users; and platforms, data and content. Each item links to the full piece, where the primary source is cited. The week as a whole is in our weekly roundup.
Business Pill · FINANCIAL CONDITIONS
A short Business Engineer explainer of financial conditions. It teaches the general idea only and says nothing about the items in this roundup.
AI Daily Briefing · video
Today’s stories in about a minute (25 Oct 2026).
The key insight: As we read it, the day’s coverage was about who finances the build: Musk wrote that SpaceX closed $40B of debt for GPUs, and Meta’s filing says its adviser put the most it could raise in equity without Zuckerberg’s voting power falling below 50% at about $700 billion.
In the same post, Musk wrote that debt markets “have almost entirely shut off access to money for any company that doesn’t have substantial revenue independent of AI data centers,” and that “the 200+ neoclouds are going to see guppy-level death and consolidation.” Two days earlier Bloomberg had reported the $40 billion deal “roiled by recent market turmoil.” Read: Musk on SpaceX’s debt. On 23 October, Barron’s reported that Meta could raise about $20 billion within weeks, according to a FinanceFeeds account of that report; Meta has not announced an offering. Read: Meta’s $700 billion filing.
Masayoshi Son of SoftBank Group “chose locations in Malaysia for $50 billion in technology investments,” Malaysia’s prime minister said, Bloomberg reported on 25 October. Read: SoftBank in Malaysia. On 14 October the US Department of Agriculture announced a $3.3 billion loan to Oglethorpe Power for a gas plant in Monroe County, Georgia, the largest in its Rural Utilities Service’s history, it said. Reuters reported that the loan “could potentially benefit data centers,” according to USDA’s head of rural development; Oglethorpe told Reuters the plant “is not dedicated to serving a data center.” Read: the USDA loan.
Worldwide spending on AI is forecast to reach $2.67 trillion in 2026, according to Gartner research reported by Techstrong, which reports that models account for only about 1% of it. Read: the Gartner split. Apollo chief economist Torsten Slok wrote on 24 October that AI “has delivered zero financial payoff to corporate America outside of tech.” Read: Slok’s note.
People in the US and China who joined the world’s 500 richest in 2026 through AI-related businesses added about $87 billion to their combined wealth in the first nine months of the year, according to a chart the Kobeissi Letter posted, crediting Bloomberg. Read: the new AI billionaires. Andreessen Horowitz “employs 11 of the top 100 individual AI investors,” Stanford’s Ilya Strebulaev wrote in a preview of the Strebulaev-Jackson ranking. Read: the AI investor ranking.
Five dollar figures from the day’s coverage, reported 14 to 25 October 2026, as each linked piece reports them. They are different kinds of money (an adviser’s estimate of Meta’s equity ceiling in Meta’s filing, SoftBank’s Malaysia plan per Bloomberg, SpaceX’s GPU debt per Musk, a possible Meta raise per Barron’s, and USDA’s loan) and not like-for-like.
Sold Out, and Priced to Rise
Amazon chief executive Andy Jassy said of AWS computing capacity, Reuters reported on 23 October: “2027 is largely sold out.” Broadcom chief executive Hock Tan named a different limit, capital, Reuters reported from a panel in Singapore on 12 October. Read: the suppliers’ horizons.
Silicon Data chief executive Carmen Li said its GPU forward curve “became contango” this year, and that “a market consensus view is the price could go up in the future, next 36 months.” Read: the GPU forward curve. Silicon Data’s B300 rental index stood at 151.9 on 22 October, up 51.9% since 31 May, while its LLM Token Index stood at $0.94 per million tokens on 21 October; DRW’s Don Wilson calls the margin between renting GPUs and selling tokens the “crunch spread.” Read: the crunch spread.
DeepSeek says one of its serving clusters earns US$38 per GPU-hour at valley pricing, in a deployment Q&A it published on 20 October. The median posted on-demand rate for an Nvidia B300 was $7.98 per hour in August, according to AIMultiple’s index as reported by Akash Network. Read: DeepSeek’s GPU-hour. SemiAnalysis’s Dylan Patel says a gigawatt of AI capacity has gone from about $40 billion to more than $60 billion, and would be closer to $20 billion without the chip design houses’ margins. Read: Patel’s gigawatt math.
Musk also wrote on 25 October: “I think we’re going to see a lot of stranded compute in partially complete data centers.” Read: stranded compute. Counterpoint Research said in a January 2026 insight that custom AI chip shipments will cross 15 million in 2028, surpassing data center GPU shipments. Read: custom chips. The H100 rent and token-price figures are set side by side in our cost-and-price piece.
Power, Materials and the Supply Chain
Research firm SemiAnalysis said on 24 October that six Texas data center campuses planning their own on-site power, “8.7GW in all,” need air permits while Governor Greg Abbott’s permit freeze is in force. Read: the Texas permits. The power deals and limits are set beside Musk’s line in our power ledger.
Japan is months away from running out of yttrium, the Financial Times reported on 20 October, according to Mining.com.au; Argus data cited by the FT puts yttrium oxide at around US$1,150 per kilogram outside China and US$9 within China, about 128 times by our arithmetic. Read: the yttrium shortage.
Nvidia confirmed plans with SPIL to expand manufacturing capacity 15 times for optical engines used in co-packaged optics by mid-2027, SDxCentral reported from GTC Berlin on 21 October. Read: the optical engines. By our count, a UBS research figure lists 50 Chinese and 38 overseas names across six layers of a humanoid robot, or 87 distinct companies, since Leadshine appears twice. Read: the humanoid map.
Agents and the Race to Reach Users
OpenAI’s Thibault Sottiaux posted on 24 October: “DM if you want to help scale dots to 1B users and have done similar things in the past.” Meta said on its third-quarter call that Muse has more than 10 million users, TechCrunch reported. Read: the personal agent race. Musk wrote on 25 October that Grok Bot “can now call someone or a business for you”, and TestingCatalog posted that voice calling is gradually rolling out on Muse. Read: agents that make calls.
The seven-day average of tokens processed for Anthropic models on OpenRouter is up 3.6 times since Claude Opus 5.5 launched, Carta’s Peter Walker posted, sharing an OpenRouter chart. Read: Anthropic on OpenRouter. Microsoft Research reports that scaling from 1 to 128 coding agents with no central orchestrator raised the mean final test-pass rate from 19.31% to 28.78% on the five hardest ProgramBench tasks. Read: Agensh.
Sam Altman said in an interview with Matthew Berman, posted on 21 October, that recursive self-improvement is “a very dangerous thing to push on” and that the world will need “speed limits or whatever on how quickly we let RSI go.” Read: Altman’s speed limits. ByteDance postings on Boss Zhipin offer forward-deployed engineers 35,000 to 65,000 yuan a month, The News reported. Read: China’s deployer pay.
Platforms, Data and Content
Erik Newsham, founder of the data broker Grepped, wrote on X that the company’s average deal used to be “around $250-$400k” and is “nowadays … closer to $1.1-$1.3m”; those are his own figures. Read: the data wars. In Vizio’s third quarter of 2024, its Device segment had a gross loss of $6.7 million while Platform+ made $115.8 million of gross profit, according to its SEC filing. Read: the smart-TV economics.
Meta said “nearly 10 million advertisers used at least one” of its generative AI ad creative tools in the third quarter. Read: Meta’s AI ad tools. Musk wrote: “We’re adding long press on like to send someone $1 via X Money.” Read: X Money’s $1 like.
The Grok Imagine account posted a 7-minute, AI-generated first episode of an “Odyssey” series on 23 October, and Variety reported that Musk has “partially followed through” on his July promise. Read: the Odyssey episode. Semafor reported that Musk “has invested roughly $1 billion in Skydance, according to people familiar with the matter.” Read: the Skydance report. Alex Wissner-Gross announced Chain Theory, with a candidate for the fine-structure constant he says matches its measured value to ten significant digits. Read: Chain Theory.
Microsoft and Alphabet report after the market closes on Wednesday 28 October, according to Microsoft’s release and Nasdaq’s earnings calendar, and Amazon on Thursday 29 October, its release says. The three companies’ latest filings show $1.71 trillion of commitments between them, by our arithmetic. Read: the week ahead.
Dylan Patel’s estimates, per a Fireside Alpha transcript of a CASPA panel, as our piece reports them. Estimates, not filings.
The Structural Read
Amazon chief executive Andy Jassy said of AWS computing capacity, Reuters reported: “2027 is largely sold out.” Silicon Data chief executive Carmen Li said the company’s GPU forward curve “became contango” this year.
DeepSeek says one of its serving clusters earns US$38 per GPU-hour at valley pricing, and SemiAnalysis’s Dylan Patel says a gigawatt would cost closer to $20 billion without the chip design houses’ margins, against more than $60 billion now.
As we read it, the agent coverage was about reach: OpenAI’s Thibault Sottiaux posted about scaling Dots to 1B users, Meta said Muse has more than 10 million users, and Musk wrote that Grok Bot can now call a person or business for you.
Elon Musk, on X, 25 October 2026, as quoted in our piece
“SpaceX closed $40B of debt for GPUs, but weโre investment grade and have Starlink & launch revenue.”
Three Implications
LENDERS AND INVESTORS Musk wrote that debt markets “have almost entirely shut off access to money for any company that doesn’t have substantial revenue independent of AI data centers,” and Bloomberg had reported the $40 billion deal “roiled by recent market turmoil.”
TEAMS RENTING COMPUTE Silicon Data’s B300 rental index stood at 151.9 on 22 October, up 51.9% since 31 May, and its LLM Token Index stood at $0.94 per million tokens on 21 October.
COMPANIES BUILDING ON MODELS The seven-day average of tokens processed for Anthropic models on OpenRouter is up 3.6 times since Claude Opus 5.5 launched, Carta’s Peter Walker posted, sharing an OpenRouter chart.
The analysis says: “The $1T capex cycle is being underwritten by a $1T concentrated demand commitment from two counterparties whose combined free cash flow is deeply negative.”
As we read it, Sunday’s coverage was about the financing side of that cycle: SpaceX’s GPU debt and Meta’s equity ceiling. We do not say one caused the other. The analysis was published in May 2026 and does not discuss these reports.
What Is Not Established
This roundup adds no new reporting and selects from the day’s coverage. Every figure above is taken from the linked FourWeekMBA piece, each built from its own primary source. The five threads are our grouping, not a finding.
We do not claim that these stories are connected or that one caused another. The SpaceX close rests on Musk’s posts; the $700 billion is an adviser’s estimate in a filing, and Meta has not announced an offering; Grepped’s figures and Wissner-Gross’s result are their own claims; and the Skydance report rests on people familiar with the matter.
This Week In The AI Supercycle
Execution Fault Lines, Concentrated Demand, and Geopolitical Firewalls
This week’s issue: the AI build is not short of demand for its parts. It is short of cheap money, and its end revenue rests on a narrow base of large, often venture-funded buyers.
Musk wrote that SpaceX closed $40 billion of debt for GPUs, and Meta’s filing puts the most it could raise in equity without Zuckerberg losing majority voting power at about $700 billion over a number of years.
The day’s coverage also includes AWS saying 2027 is largely sold out and a GPU forward curve that now prices rents to rise. The Business Engineer’s analysis of where AI capex goes reads stories like these together: read the analysis.
A note on sourcing. This roundup draws on FourWeekMBA’s news coverage of 24 and 25 October 2026, linked above; every story was built from its own primary source, which it cites and links. We re-read each piece on 25 October 2026 and took every figure here from it. The five threads and the figures in our charts are our own grouping and selection. Nothing here is a forecast, and nothing here is financial or investment advice.
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.
Scroll to Top
Discover more from FourWeekMBA
Subscribe now to keep reading and get access to the full archive.
