Voters and markets both read the midterms as a vote on AI’s physical build-out on 4 November 2026, by our reading of the news. On unofficial returns, limits on data centers lead in 40 of the 41 local measures on Ballotpedia’s 3 November list, by our recount, with about 74% of counted votes on the limiting side. By 1:09 p.m. ET, Amazon, Microsoft, Alphabet and Meta had added about $653 billion of market value while five chipmakers had lost about $149 billion net, by our arithmetic from Nasdaq.com prices.
As we read it, the rest of the day’s coverage ran on two more threads. One was what AI costs, as Apple cut some engineers’ daily AI budgets and the labs’ cheapest main models came to 1% of the top price. The other was who owns the tools around coding agents, as Meta bought Conductor and OpenAI agreed to buy Pierre. The threads are the ballot; the market split; AI money in the races; the cost squeeze; the coding-agent land grab; money and deals; and models and labs. Each item links to the full piece, which names its sources.
Business Pill · GRID COST ALLOCATION
A short Business Engineer explainer on how grid costs are allocated between customers. It teaches the general idea only and does not discuss the elections or the companies in this roundup.
AI Daily Briefing · video
Today’s stories in about a minute (4 Nov 2026).
The key insight: As we read it, the day’s coverage treated the midterms as a vote on AI’s physical build-out: limits on data centers lead in 40 of the 41 local ballot measures, by our recount of unofficial returns, while Amazon, Microsoft, Alphabet and Meta added about $653 billion of market value by 1:09 p.m. ET and five chipmakers lost about $149 billion net, by our arithmetic from Nasdaq.com prices.
Limits on data centers lead in 40 of the 41 local measures on Ballotpedia’s 3 November list, by our recount of unofficial returns at 16:19Z, and about 74% of the votes counted went to the limiting side, by our arithmetic. Goldman Sachs warned that Democratic wins at the state and local level could trigger temporary construction moratoria affecting 45% of planned US data center capacity, according to a summary of its note posted on X; we have not read the note. Read: the 40 of 41 recount, and our explainer on how data center politics works.
In Ohio, limits passed on 17 of 21 measures, by our count of the Cincinnati Enquirer’s lists, in what Fortune reported were at least 14 municipalities passing bans or other restrictions. In NBC News’ national exit poll, 56% of voters said AI and data centers had been crucial or a major factor in their vote for Congress, and 76% opposed a new data center in their own community. Read: Ohio’s 21 measures and the exit poll.
AP called the Texas governor race for Gina Hinojosa at 04:17 UTC. Her campaign site says: “We need an immediate moratorium on data centers to last until the legislature can meet to put in place real protections”. ERCOT’s September board materials show about 194 GW of large loads conditionally kept in its Batch Zero study, 2.1 times the grid’s record peak demand of 91,134 MW, by our arithmetic.
SemiAnalysis estimated on 1 November that 30.3 GW of planned 2028 capacity would be at risk if every pause candidate won, 16.7 GW of it in Texas. Read: the ERCOT queue and the governor races.
In New York, Mayor Zohran Mamdani’s one-year school AI moratorium covers nearly 600,000 students in 2-K through 8th grade, against a maximum of 50,000 in five high school pilots, the city’s release says: about 12 for every pilot seat, by our arithmetic. Read: the school moratorium.
Limits on data centers lead in 40 of the 41 local ballot measures on Ballotpedia’s 3 November 2026 list, by our recount of unofficial returns at 16:19 UTC on 4 November, with about 74% of the votes counted on the limiting side, by our arithmetic.
Markets Split the Build-Out
At 1:09 p.m. ET, Amazon, Microsoft, Alphabet and Meta had added about $653 billion of market value since the 3 November close, 4.4 times the roughly $149 billion net that Nvidia, Micron, Broadcom, Intel and AMD had lost, by our arithmetic from Nasdaq.com prices; Microsoft’s gain alone, about $248.0 billion, was larger than the chipmakers’ combined net loss. These are intraday prices. Read: the $653 billion split.
Earlier in the session, the four hyperscalers averaged a 5.1% gain while ten data-center builders and power suppliers were down an average of 11.8%, by our arithmetic. Six neocloud stocks fell an average of 36.0% from the 2 October close to the 3 November close, against 8.2% for the four hyperscalers, and CoreWeave’s 9.625% notes due 2032 last traded at 78.50 cents on the dollar on FINRA’s TRACE data. Read: builders and power names and the neocloud drawdown.
Tesla and SpaceX each ended the after-hours session on 3 November 1.75% below their regular-session closes, about $61 billion of market value on the share counts in their 10-Q cover pages, by our arithmetic. In Seoul, SK hynix and Samsung Electronics regained about $94 billion between their intraday lows and the close, by our arithmetic on Korea Exchange data. Read: Tesla and SpaceX and Samsung and SK hynix.
AI Money in the Races
AI and crypto super PACs spent $61.7 million of their general-election money on the losing side of races AP had called by 15:25Z, by our count of FEC filings, 67% of the $92.2 million they spent in called races. In Ohio’s Senate special election, the crypto PAC Defend American Jobs spent $56.6 million for Jon Husted and against Sherrod Brown, and AP called the seat for Brown, by our arithmetic from FEC filings. Read: the $61.7 million.
In Michigan, AP called the Senate race for Abdul El-Sayed, whose plan says: “At an absolute minimum, I’m calling for a pause in AI frontier development”. Musk-funded America PAC reported $9.18 million of the $220.16 million outside groups spent in the race, by our arithmetic, and none of the 10 AI super PACs we track appears in those filings. Read: El-Sayed’s race and the early AP calls.
The Cost Squeeze
Apple lowered the daily spending cap for some engineers using AI models to $200 from $300 last month, Bloomberg reported, a cut of one third by our arithmetic, and is planning further cuts for some teams, “potentially reducing their daily allowances to as little as $100”. Read: Apple’s AI budgets.
In OpenAI’s and Anthropic’s main current lineups, the cheapest model costs $0.10 per million input tokens and $0.50 per million output tokens, against $10 and $50 for the most expensive: 1% of the top price, by our arithmetic from their pricing pages. Tencent lists its new Hy4 at $2.501 per million output tokens, one-eighth of what OpenRouter lists for Claude Opus 5.5, by our arithmetic.
Alibaba sold nearly 100,000 QwenWork monthly passes in 30 minutes, according to its own data as reported by SCMP, about $150 in total by our arithmetic. Read: the 1% price floor, Tencent’s Hy4 and Alibaba’s live stream.
Anthropic’s amended IPO filing shows a net loss of $104.0 billion for the nine months to 30 September; $94.3 billion of it, about 91% by our arithmetic, is the change in fair value of convertible notes, derivative liabilities and SAFEs, against an operating loss of $1.26 billion. Read: the $104 billion loss.
The Coding-Agent Land Grab
Meta has acquired Conductor, “an AI coding startup that lets software engineers manage multiple coding agents simultaneously”, Axios reported; terms were not disclosed, and Conductor’s own post says it will begin deleting Conductor Cloud user data on 20 November. OpenAI announced that it will acquire Pierre Computer Company, the company behind the Code Storage git service, a week after Microsoft said agent-generated pull requests on GitHub “have increased 6X over the past six months”. Read: Conductor and Pierre.
Ricursive Intelligence released Ripple-65B Preview under Apache 2.0, with a self-reported Terminal-Bench 2.1 score of 80.5 against 80.1 for a 501-billion-parameter open model. Cognition launched SWE-2 Flash the same day and posted about it about 91 minutes after Ricursive’s post, by our arithmetic from the two posts’ times; it lists SWE-2 Flash at 89.2 on the same test and says it is “not sharing the weights publicly yet”. Read: Ripple-65B and SWE-2 Flash.
ChatAndBuild says its Living Brain memory layer used 22.6 times fewer tokens per question than a full-context baseline in one internal test. Two users report on Postman’s GitHub tracker that its desktop app installed AI plugins into Claude Code and Codex when they opened it. Read: ChatAndBuild and Postman’s plugins.
Three of the four named Anthropic exits we could source over the last 60 days came from Claude Code, by our count, the latest Cat Wu, who wrote “Friday was my last day at Anthropic”. Similarweb estimates monthly visits to claude.ai/code rose about 12x in a year, by our arithmetic from its chart, and 79 of Anthropic’s 681 open roles name Claude Code, by our count. Read: Cat Wu, the exits, the traffic and the job ads.
Money and Deals
Sequoia Capital says it has raised about $10 billion in new funds, in a memo that also closes its London office, sunsets Arc and Scouts and merges its Early and Growth teams: roughly $1.4 billion per general partner, by our arithmetic. Meta’s first euro bond drew โฌ53 billion of orders for a sale of as much as โฌ9 billion, according to First Squawk, about 5.9 times the maximum size by our arithmetic. Read: Sequoia and Meta’s euro bond.
AMD’s 10-Q values its pending Taalas acquisition at approximately $2.9 billion, including up to approximately $1.4 billion of contingent consideration. The same filing values the warrants AMD issued to OpenAI and Meta at a weighted average of $178.58 a share, about 12.5 times their grant-date fair value, by our arithmetic. AMD’s 8-K filings give each customer a warrant for up to 160 million shares; the 320 million shares equal about 19.5% of AMD’s shares outstanding, also by our arithmetic. Read: Taalas, the warrants and how the AMD deals work.
Riot’s 10-Q names Anthropic as the tenant of its 910-megawatt, roughly $28 billion Corsicana lease, and TeraWulf said Google will backstop Anthropic’s rent on about $19 billion of 20-year Kentucky leases. SpaceX’s prospectus prices Anthropic’s compute at $1.25 billion a month for approximately 325,000 Nvidia GPUs, about $5.27 per GPU-hour by our arithmetic. Read: Riot, TeraWulf, SpaceX’s GPU rent, and our AI debt financing tracker.
Oracle cancelled a $7 billion AI chip lease with Tencent under pressure from the Trump administration, the Financial Times reported. Intel chief executive Lip-Bu Tan has bought about $19.7 million of Intel shares since August at an average of $77.21, by our arithmetic from his Form 4 filings, and Google will invest US$15 billion in Malaysia in 2027 and 2028, Reuters reported. Read: Oracle and Tencent, Tan’s purchases and Google in Malaysia.
General Catalyst’s Customer Value Fund committed $500 million to Nooks for a share of returns rather than equity, Reuters reported, and Capital One Ventures led a $150 million investment in Sentra. Read: Nooks and Sentra.
Models and Labs
New OpenAI models could come as soon as Friday, The Information reported, 64 days after GPT-6 Astra, by our count. Moonshot AI released the weights of Kimi K3.1 Pro, which scores above Claude Opus 5.5 on 16 of 37 rows in its own table, by our count. Read: OpenAI’s Friday and Kimi K3.1 Pro.
SpaceX selected Qualcomm’s Dragonwing platforms for its planned Starlink Mobile 5G network, and its FCC filing asks for up to one million orbital data center satellites, about 90 times its Starlink fleet, by our arithmetic. Read: Qualcomm and the orbital filing.
Later on 4 November: Meta’s Muse app was downloaded an estimated 942,000 times on the US App Store in the week to 29 October, against 556,000 for ChatGPT, according to FoxData estimates via ASO World; these are third-party estimates. Applied Compute says its Ari agent took Qwen3.5-9B from 0.118 to 0.629 held-out reward on Lichess chess puzzles in one 16-hour episode, on its own scores. In a Pew Research Center survey, 60% of US adults say they would feel uncomfortable if their own doctor used AI for things like diagnoses and treatments. Read: Muse downloads, Ari Autoresearch and the Pew survey.
Change in stock market value from the 3 November close to 1:09 p.m. ET on 4 November 2026: Amazon, Microsoft, Alphabet and Meta about +$653 billion combined, Microsoft alone about +$248.0 billion, and Nvidia, Micron, Broadcom, Intel and AMD about -$149 billion net. Prices from Nasdaq.com, share counts from 10-Q/10-K cover pages; the dollar changes are our arithmetic. Intraday, not closing, prices.
The Structural Read
As we read it, the vote came with a queue attached: ERCOT’s September board materials show about 194 GW of large loads conditionally kept in its Batch Zero study, 2.1 times the grid’s record peak demand, by our arithmetic, and AP called the Texas governor race for Gina Hinojosa, whose campaign site calls for an immediate moratorium on data centers.
As we read it, the market sorted companies by what they already own; we do not say why prices moved. By our arithmetic from Nasdaq.com prices, the four hyperscalers averaged a 5.1% gain at 11:51 a.m. ET while ten data-center builders and power suppliers were down an average of 11.8%, and six neocloud stocks fell an average of 36.0% from the 2 October close to the 3 November close.
As we read it, cost ran through the rest of the day: Apple cut some engineers’ daily AI budgets to $200 from $300, Bloomberg reported, and in OpenAI’s and Anthropic’s main lineups the cheapest model costs 1% of the top price, by our arithmetic. We do not say one caused the other.
Gina Hinojosa, Texas governor-elect, on her campaign site, as quoted in our piece of 4 November 2026
“We need an immediate moratorium on data centers to last until the legislature can meet to put in place real protections”
Three Implications
BUILDERS AND POWER SUPPLIERS Ten data-center builders and power names were down an average of 11.8% at 11:51 a.m. ET on 4 November, by our arithmetic, and CoreWeave’s 9.625% notes due 2032 last traded at 78.50 cents on the dollar on FINRA’s TRACE data.
LANDLORDS AND LENDERS Riot’s 10-Q names Anthropic as the tenant of its roughly $28 billion Corsicana lease, and TeraWulf said Google will backstop Anthropic’s rent on about $19 billion of 20-year Kentucky leases.
ENGINEERING BUDGETS Apple lowered the daily AI spending cap for some engineers to $200 from $300, Bloomberg reported, while Meta bought Conductor and OpenAI agreed to buy Pierre, two companies that make tools around coding agents.
The analysis describes a stack of nine layers, “running from energy at the bedrock to governance at the ceiling”, and says: “The supercycle has a ceiling now, and the ceiling is political.”
As we read it, the day’s ballots put local votes on the energy and physical layer, and the market priced the companies on either side of them. We do not say one caused the other. The analysis was published in June 2026 and does not discuss these elections.
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 threads, and the reading of the ballot and the market moves as one story, are our grouping, not a finding.
We do not claim that the ballot results moved any stock. The vote counts are unofficial returns, the market figures are intraday prices, Goldman’s 45% rests on a summary posted on X, and the model scores are self-reported by the companies that built the models.
This Week In The AI Supercycle
Anthropic’s Reality Check, The Capital Squeeze, & The New World Order
This week’s issue: for two years the AI build-out was priced as a growthstory. This week it started being priced as a credit story.
On unofficial returns, limits on data centers lead in 40 of 41 local ballot measures, by our recount. Amazon, Microsoft, Alphabet and Meta added about $653 billion of market value by 1:09 p.m. ET on 4 November while five chipmakers lost about $149 billion net, and six neocloud stocks fell an average of 36.0% from the 2 October close to the 3 November close, by our arithmetic.
For the frame behind the day, read The AI Supercycle on The Business Engineer, on the nine layers from energy to governance: read the analysis.
A note on sourcing. This roundup draws on FourWeekMBA’s news coverage of 3 and 4 November 2026, linked above; every story was built from its own primary source, which it cites and links. We re-read each piece on 4 November 2026 and took every figure here from it. The 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.
