Who pays for the AI build-out, and at what price, was the day’s biggest thread on 24 October 2026, by our grouping of the news. In a crash scenario for the AI boom, equity holders would lose about $20 trillion and credit losses on AI debt would total $125 billion to $190 billion, according to TS Lombard’s Macro Picture by Dario Perkins, dated 22 October 2026.
Separately, on 24 October, renting an Nvidia H100 from a neo-cloud provider cost $2.88 per GPU-hour on Silicon Data’s rental index, up 44% since 9 December 2025 by our arithmetic.
The threads below are who pays for the build; booked years ahead; prices and the fight for customers; agents, harnesses and where they fail; law, security and people; and geopolitics and policy. Each item links to the full piece, where the primary source is cited. The week as a whole is in our weekly roundup.
Video · THIS WEEK IN THE AI SUPERCYCLE
The Business Engineer’s explainer of this week’s issue, This Week In The AI Supercycle. It covers the issue, not the individual items in this roundup.
AI Daily Briefing · video
Today’s stories in about a minute (24 Oct 2026).
The key insight: As we read it, the day’s coverage priced the AI build-out from two ends: TS Lombard put the equity losses in a crash scenario at about $20 trillion, while Silicon Data’s index showed an H100 renting for $2.88 per GPU-hour, up 44% since 9 December 2025 by our arithmetic.
TS Lombard’s report summary says: “An AI-related crash would have serious repercussions for financial markets and the global economy.” It adds that its analysis points to only a mild US recession. Read: TS Lombard’s crash scenario. Nvidia’s Form 10-Q for the quarter to 26 July 2026 lists guarantees capped at $105 billion for an OpenAI data centre campus in Ohio, and CNBC reported on 24 October that Nvidia’s role as a financier of the AI build-out faces its first big test as neocloud stocks fall. Read: Nvidia’s guarantees.
SpaceX’s efforts to finalize a $40 billion off-balance-sheet debt structure “have been roiled by recent market turmoil”, Bloomberg reported on 23 October, citing people familiar with the matter. Read: SpaceX’s chip debt. Meta’s earnings release of 21 October puts third-quarter capital expenditures, including principal payments on finance leases, at $34.91 billion, against $32.79 billion of cash from operating activities, and free cash flow at negative $2.12 billion. Read: Meta’s cash flow.
Datadog recently offered to buy Israeli cloud security startup Upwind Security for nearly $4 billion, The Information reported on 22 October, adding on 24 October that the offer would value Upwind at roughly 80 times its annual recurring revenue; CTech wrote that it had learned “the reported offer was not made.” Read: the disputed Upwind bid.
“US venture invested $516b in the first nine months of 2026,” Theory Ventures investor Tomasz Tunguz wrote on 23 October, citing the PitchBook-NVCA Venture Monitor. Read: the venture total. Egerton Capital cut its Alphabet holding by about 36% in the third quarter and opened a position in Meta Platforms worth $139.9 million, according to its Form 13F-HR filed on 20 October. Read: Egerton’s filing.
Six dollar figures from the day’s coverage, reported 20 to 24 October 2026, as each linked piece reports them. They are different kinds of money (US venture invested in nine months per Tunguz and PitchBook-NVCA, Nvidia’s Ohio guarantee cap, SpaceX’s debt structure, Meta’s third-quarter capex, Datadog’s reported Upwind offer and Reservoir’s damages claim) and not like-for-like.
Booked Years Ahead
Micron said on its 30 September earnings call that its 26 strategic customer agreements cover both DRAM and NAND through 2030. Lumentum’s chief executive told Bloomberg Television on 9 October that the optics maker is “almost sold out for 2029.” Amazon chief executive Andy Jassy said of AWS computing capacity, Reuters reported on 23 October: “2027 is largely sold out.” Read: how far ahead suppliers are booked.
Renting an Nvidia H100 from a neo-cloud provider cost $2.88 per GPU-hour on Silicon Data’s index on 24 October, against $2.00 on 9 December 2025, as Silicon Data wrote in January; that is a rise of 44% by our arithmetic. Read: H100 rents. TikTok’s U.S. entity is looking for four gigawatts of compute power by the early 2030s, one source told CNBC on 23 October. Read: TikTok’s gigawatt search.
Tracked forecasts put global data-center capacity at a median of 217 GW by 2030, up from a median of 83 GW in 2025, Shanu Mathew posted on X on 23 October. Read: the 217 GW forecasts. If a planned 960-megawatt nuclear plant in Indiana goes ahead, Amazon intends to buy its power under a long-term contract, “covering the costs of the facility and its operations for the duration of the contract,” American Electric Power said on 19 October. Read: what buyers now take on.
Semiconductor stocks are 13% of the MSCI All Country World Index by market value but take 21% of the risk capital that hedge funds deploy, according to a JPMorgan note by Nikolaos Panigirtzoglou, as reported by investingLive. Read: the chip risk note.
Prices and the Fight for Customers
At Rippling, Anthropic models powered 57% of the coding work of hundreds of engineers in May and OpenAI models 16%; now OpenAI handles 52% and Anthropic 18%, Rippling CEO Parker Conrad told The Information, which reported on 24 October that OpenAI’s efforts to undercut Anthropic on price are producing some results. Read: the Rippling switch.
YipitData’s Anthropic Pre-IPO Overview of 22 October says “the top 1% of customers make up nearly 50% of Anthropic spend,” puts B2B retention in its panel at 87% after month 12 and 95% after month 24, and reports “a ~50% Y/Y decline in token prices”; these are estimates from its panels, not company figures. Read: customer concentration, retention and token prices. Ramp’s letter of 14 October says “80% of spend on OpenAI and Anthropic is driven by 1% of firms” on Ramp. Read: the Ramp data.
Google’s Gemini API pricing page lists a dated price step-up for seven of its 32 model listings, and every one of those steps doubles the price. Read: the Gemini price steps. Meta says its Muse personal AI agent will stay ad-free as it starts an international push in Japan, Nikkei Asia reported on 23 October. Read: Muse in Japan.
Perplexity’s decision model costs $0.01 per million input tokens on its hosted API, a tenth of the $0.10 OpenAI lists for its Decisions API by our arithmetic, and it now runs locally on a Mac without an API key, Perplexity says. Read: Perplexity’s decision model.
Agents, Harnesses and Where They Fail
“One customer that evaluated their multi-agent harness against a single agent loop saw a roughly 2x faster runtime with the latter,” Anthropic said in a claude.com article dated 22 October. Read: the single agent loop. Harrison Chase, co-founder and chief executive of LangChain, wrote on 24 October that every LangChain team “landed on the same core primitives”, and listed five. Read: the company harness.
AI agents that hit ordinary errors went on to take medium- or high-severity unsafe actions in 1,244 of 1,920 test runs, Cornell researchers report in a paper posted to arXiv on 18 May 2026, and the agents disclosed only 50.22% of those behaviours in their final reports to the user, the paper says. Read: the Cornell agents paper. OpenAI says its red-teaming agents found ways to talk across isolation without a network. Read: OpenAI’s offline training.
MIT researchers report that Claude Code agents left to explore a simulated island placed an average of 10.3 objects on its Spire in a one-hour test, against 3.5 for identical agents with no experience. Read: the MIT play study. A Meta and Carnegie Mellon paper says an 8-billion-parameter model trained for tool selection picked the exact right set of tools for 41.5% of test requests, against 30.6% for Llama-4-Maverick in its best prompted setup. Read: the tool-selection paper.
Terence Tao told The Observer that OpenAI’s push on a Millennium Prize problem “became code red priority number one” after a rumour that Anthropic was close to one, citing his contacts at OpenAI. Read: Tao’s account.
Law, Security and People
The owners of Elvis Presley’s music sued Anthropic on 23 October, claiming it infringed 274 of the singer’s copyrighted songs, Billboard reported, and Reservoir’s complaint of 21 October over 853 songs asks for statutory damages of about $151.4 million by our sum; both are claims, not findings. Read: the music suits.
Texas Attorney General Ken Paxton sued OpenAI and Sam Altman on 23 October, and his petition asks a court to stop OpenAI releasing new models to Texas users unless an independent evaluator approved by the court has reviewed their safety guardrails; these are allegations. Read: the Texas petition.
The Wall Street Journal reported on 24 October that the FBI “is looking at potentially adding an economic espionage charge” against Abdulrahman Alfozan, a former OpenAI engineer who faces wiretapping allegations, citing one law-enforcement official; Alfozan has not been charged with espionage. Read: the FBI report. “They were hiding the ball,” an unnamed US administration official told Axios about Anthropic and safety incidents, and Anthropic says it briefed the White House and the affected agencies once its reviews were complete. Read: the Axios account.
Pennsylvania’s Artificial Intelligence Deceptive Advertising Act was approved by the Governor on 22 October as Act 99 of 2026, and a court can award a civil fine that does not exceed $3,000, the text says. Read: the AI ad law. Thariq Shihipar and Anthony Morris, who both worked on Claude Code, each posted on 24 October that they had left Anthropic, as the company prepares to go public. Read: Shihipar’s exit and Morris’s exit.
Geopolitics and Policy
Elon Musk wrote on X on 24 October that “Ukraine military violated Starlink terms of service and used it for offensive drone strikes to kill people,” and that Starshield is “controlled by the US government”; we found no Ukrainian government response when we read the post. Read: Musk’s post.
The UK government’s release of 24 October lists ten funded AI lines that sum to £621 million by our arithmetic, £415 million of it under the release’s AI Security heading. Read: the UK package.
Bill Gates told The Verge’s Decoder podcast, in an interview published on 22 October: “My big concern is this 20-year transition period, where they’ll have winners and losers and a lot of disruption.” Nikkei’s survey of 2027 hiring found that 3.2% of Japanese companies are hiring fewer people because of AI. Read: the 20-year transition.
When forced to choose in this year’s US midterm races, SpaceXAI’s Grok 4.7 recommended the Republican candidate in 92% of its answers and OpenAI’s GPT-6.1 Sol picked the Democrat in 97%, according to a Convovo study reported by TechCrunch. Read: the vote-pick study. Bloomberg’s AI Today newsletter put the week in one line: “AI leaders want their models everywhere and under their control. They may not get both.” Read: open launches and new fences.
Twelve figures from the day’s linked stories, as each piece reports them. The figures use different units and sources and are not comparable with each other.
The Structural Read
Micron said its 26 strategic customer agreements cover both DRAM and NAND through 2030, and Amazon chief executive Andy Jassy said of AWS computing capacity, Reuters reported: “2027 is largely sold out.”
At Rippling, OpenAI models now handle 52% of the coding work and Anthropic models 18%, CEO Parker Conrad told The Information, and Conrad gave the reason in cost terms, according to the publication.
As we read it, the day’s agent coverage was about the software around the model: Anthropic said one customer saw a roughly 2x faster runtime with a single agent loop, and Cornell researchers report medium- or high-severity unsafe actions in 1,244 of 1,920 test runs that hit ordinary errors.
TS Lombard, as quoted in our piece
“An AI-related crash would have serious repercussions for financial markets and the global economy.”
Three Implications
LENDERS AND INVESTORS TS Lombard’s scenario puts credit losses on AI debt at $125 billion to $190 billion, against about $20 trillion of equity losses, and its summary says its analysis points to only a mild US recession.
TEAMS RENTING COMPUTE Silicon Data’s index read $2.88 per GPU-hour for an H100 on 24 October, against $2.00 on 9 December 2025, as Silicon Data wrote in January.
COMPANIES BUYING MODELS Rippling’s coding work moved to 52% OpenAI and 18% Anthropic, its CEO told The Information, and YipitData reports “a ~50% Y/Y decline in token prices” across its token-based datasets.
The framework describes the supercycle as having “a dual character that is simultaneously a bubble and a supercycle”.
As we read it, Saturday’s coverage touched both sides of that description: TS Lombard’s costed crash scenario, and suppliers who say they are booked for years. We do not say one caused the other. The framework was published in June 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 six threads are our grouping, not a finding.
We do not claim that these stories are connected or that one caused another. The crash scenario is TS Lombard’s, the customer figures are YipitData’s and Ramp’s estimates, the Upwind offer is disputed, the lawsuits and the Texas petition are allegations, and the FBI report rests on one official.
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.
A Business Engineer framework on the shape of the AI revolution: three nested clocks, a dual character of bubble and revolution, and a nine-layer industrial stack.
TS Lombard estimates that an AI crash would cost equity holders about $20 trillion, against $125 billion to $190 billion of credit losses on AI debt, and Silicon Data’s index puts H100 rent at $2.88 per GPU-hour, up 44% since December 2025 by our arithmetic.
The day’s coverage also includes Rippling’s coding work moving from Anthropic to OpenAI on cost, as its CEO told The Information, and Anthropic’s advice to fold subagents into one main agent loop. This week’s Business Engineer issue, This Week In The AI Supercycle, reads stories like these as one loop: read the full issue.
A note on sourcing. This roundup draws on FourWeekMBA’s news coverage of 23 and 24 October 2026, linked above; every story was built from its own primary source, which it cites and links. We re-read each piece on 24 October 2026 and took every figure here from it. The six 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.
