Credit carried the sharpest AI numbers of 7 November 2026, by our reading of the news: Meta sold 2031 notes at a 5.850% yield while CoreWeave’s 2031 notes traded at 16.33%, and lenders committed $48.4 billion of GPU-backed debt in 2026 through 10 October, 47% of the AI infrastructure debt American Compute tracks.
As we read it, the rest of the day’s coverage ran on who carries that risk: Anthropic’s IPO plumbing, a split between memory prices and memory shares, research groups folded into product units, and sites and politics. Each item links to the full piece, which names its sources.
Business Pill · CREDIT SPREADS
An evergreen Business Engineer explainer on why two borrowers pay different rates. It teaches the general idea only and does not discuss the companies in this roundup.
AI Daily Briefing · video
Today’s stories in about a minute (7 Nov 2026).
The key insight: As we read it, the day’s coverage priced AI risk through credit: Meta sold 2031 notes at a 5.850% yield on 3 November while CoreWeave’s 2031 notes traded at 16.33% three days later, about 10.5 percentage points apart by our arithmetic, and American Compute’s tracker puts GPU-backed loans at 47% of the AI infrastructure debt it counts for 2026. We do not say one explains the other.
Meta sold $3.25 billion of notes due November 2031 at a 5.850% yield on 3 November, 110 basis points over Treasuries, according to its pricing term sheet filed with the SEC, and CoreWeave’s 9.75% notes due October 2031 traded at a 16.33% yield on FINRA’s TRACE data three days later: a gap of about 10.5 percentage points for the same maturity year, by our arithmetic. One is a new-issue yield, the other a market yield. Read: the two 2031 yields.
Lenders committed $48.4 billion of new debt secured on GPUs in public financings in 2026 through 10 October, against $6.6 billion in all of 2025, according to American Compute’s GPU Financing Tracker, about 7.3 times last year’s total by our arithmetic; the tracker puts GPU-backed debt at 47% of the AI infrastructure debt it counts for 2026, up from 8%. Read: the GPU-backed loans.
Nebius and IREN say they have signed AI cloud contracts priced at roughly $20 million to $40 million of annual revenue per megawatt, while CoreWeave brought in about $6.9 million to $8.2 million of annualised revenue per megawatt of active power in its second quarter, by our arithmetic from its own figures; the companies measure a megawatt differently. CoreWeave reports on 10 November and Nebius on 12 November. Read: revenue per megawatt and the two earnings dates.
Nscale’s IPO registration statement reports about $103.4 billion of active and contracted total contract value, about $2.6 billion of it active, and agreements with Microsoft and Anthropic that could account for up to about 86% of it, by our arithmetic. Read: Nscale’s backlog.
New GPU-backed debt in public financings: $6.6 billion in all of 2025 and $48.4 billion in 2026 through 10 October, according to American Compute’s GPU Financing Tracker, which puts GPU-backed debt at 47% of the AI infrastructure debt it counts for 2026, up from 8%; the 7.3x multiple is our arithmetic.
Where Stocks Ended
The 12 AI stocks we track ended Friday, 6 November, about $469.0 billion above Thursday’s close, about 32.4% of Thursday’s $1,447.4 billion loss, by our arithmetic from Nasdaq’s official closing prices. SpaceX’s 5-year credit default swaps had reached a record 286.56 basis points at 17:33 UTC on 5 November, according to Markit data cited on X, about 3.5 hours before that close. Read: Friday’s close and the credit clock.
Brett Ryan, an economist at Deutsche Bank Securities, told Bloomberg that if moratoriums or regulatory pressures slow the buildout, “it almost prevents you from blowing too big of a bubble”. Read: Deutsche Bank’s view.
Anthropic’s IPO, Set Up
Anthropic will proceed with its IPO roadshow next week, The Information reported; its amended S-1 of 2 November still leaves the price range blank. Anthropic’s own help-centre page lists 33 distinct broker and selling-agent names across 14 countries for retail offers outside the US, by our count, and says it does not currently intend to list on an exchange in those jurisdictions. Read: the roadshow report and the 33 brokers.
Ondo Finance said on X that it plans to offer tokens tracking Anthropic on IPO day through Ondo Stocks, which its product page says is not available in the United States; the post does not say that Anthropic is involved. Twelve exchange-traded funds built around Anthropic’s stock have prospectuses on file with the SEC, six of them seeking to gain when the shares fall, by our count. Read: Ondo’s plan and the 12 funds.
Kalshi’s contract on Anthropic confirming an IPO before 1 December last traded at 81 cents on 7 November, against 90 cents at 23:03 UTC on 6 November, and Polymarket’s most-backed valuation bracket was still $1.25 trillion to $1.50 trillion. The $80 billion to $100 billion Anthropic is reported to be seeking would sit on either side of SpaceX’s $86.25 billion June offering, by our arithmetic. Read: the odds and the SpaceX yardstick.
Memory: Prices Up, Shares Down
Shares of SK hynix, Samsung Electronics and Micron fell between 18.7% and 26.1% from the 29 October close to the 6 November close, by our arithmetic from exchange closes, while Counterpoint Research said memory prices were still rising and put third-quarter DRAM revenue at nearly $200 billion, up 369% year on year. Read: the memory split.
Samsung Securities estimated that the share prices of Samsung Electronics and SK hynix already reflect memory price declines of about 46% and 36%, eToday reported; Meritz Securities checked the Americas sales channels of the three makers and found “no abnormal signs of a change in demand were detected for now”, Joseilbo reported. SK hynix plans shareholder returns of approximately 100 trillion won for fiscal 2026, about 101% of its September net cash, by our arithmetic. Read: the two brokers and SK hynix’s payout.
Nvidia is providing billions of dollars in financing to expand CoWoS packaging capacity at ASE’s SPIL, Reuters reported, citing an anonymous industry executive; Nvidia did not respond to Reuters. A TSMC spokesperson told Tom’s Hardware that TSMC focuses on its core business, dedicated IC foundry, after relays of The Information said TSMC is considering producing 3D DRAM at Nvidia’s request; the two accounts conflict. Read: the SPIL financing and the TSMC update.
Labs and People
John Langford wrote on hunch.net that Microsoft Research’s AI Frontiers group, about 50 people, is over after being reorganised into Microsoft AI in early September, and that Langford has left. By our count, Google, Meta and Microsoft have each moved a named AI research group under a model-building unit in about 41 months. Read: Langford’s post and the three reorganisations.
New OpenAI offers to senior US software engineers carry a median total pay of $730,000, against $280,000 for comparable new hires at Microsoft, according to Levels.fyi data, 2.6 times as much by our arithmetic. OpenAI chief financial officer Sarah Friar told CNBC that 300 million people use ChatGPT for health and wellness and 200 million for personal finance. Read: the pay gap and Friar’s figures.
Sites, Grid and Politics
Public records appear to show SpaceX building a “Starbase supercomputer” in Texas, The Information reported, according to Techmeme’s summary; the city of Starbase approved a 45-year development agreement with SpaceX on 7 October that prohibits construction moratoriums on the site, RGV Business Journal reported. FundaAI wrote that SpaceX ordered 6,000 space-specific racks for 2027; SpaceX has not confirmed it. Read: the Starbase records and the rack report.
A bipartisan Senate permitting bill would make data center developers pay for new and existing transmission, The Verge reported. In AP-called general-election races, 97.5% of the money the Public First network of AI-safety super PACs reported spending went into races that ended its way, against 52.6% for Leading the Future, by our count from FEC filings and AP calls. Read: the permit bill and the PAC scorecard.
Later on 7 November: CoreWeave sent a letter to Texas Governor-elect Gina Hinojosa seeking to work with her incoming administration on data center projects, Bloomberg reported; the letter, signed by chief executive Michael Intrator, lists six places with active CoreWeave facilities in Texas and puts its contracted power capacity at about 4.2 gigawatts, about 2.7 gigawatts of it not yet active by our arithmetic. Read: the CoreWeave letter.
Agents, Models and Security
Microsoft Threat Intelligence said that, in at least one case, it “observed an AI agent advance from scanning GitHub for secrets to exfiltrating data from Azure Key Vault in just five minutes”; the post does not name the actor. Security firm Apiiro says it found 2,539 real secrets in 1,994 public GitHub repositories in files AI coding agents wrote, most of them a settings file created by Claude Code. Read: the five-minute attack and the leaked secrets.
OpenRouter says it routed 790 trillion tokens in October, with agents consuming 76% of them. Hightouch reported that a cheap decision model handing only its unsure cases to Claude Opus 5.5 scored 98.8% accuracy against 99.0% for Opus alone, at $0.88 per 1,000 identities against $3.86, 4.4x cheaper by our arithmetic. Read: OpenRouter’s October and the Hightouch test.
OpenAI lists GPT-6-Cyber at $20 per million input tokens and $100 per million output tokens, twice GPT-6 Astra’s, by our arithmetic, with Daybreak Red access and approval required. Xiaomi’s LLM-Core team says a new attention design cuts the KV-cache memory a 1M-token context needs to 2.69 GB from 12.09 GB, a 4.5x reduction by our arithmetic. Read: the cyber pricing and Xiaomi’s paper.
Deals and Disputes
Groq’s executive chairman has reached out to investors for a new round at an $8 billion valuation, not including the money raised, according to documents seen by Bloomberg News, about 2.5 times its $3.15 billion August mark by our arithmetic. Silver Lake and Saudi Arabia’s Public Investment Fund made a takeover approach for Unity Software, The Information reported, and Reuters reported that Unity was valued at $25 billion in the talks. Read: Groq’s round and the Unity approach.
Eight Mile Style, which owns and administers the rights to nearly 270 Eminem compositions, sued Suno in San Francisco federal court seeking more than $41 million in damages, the Detroit Free Press reported; Suno says it prohibits unauthorized use of an artist’s work. Construction and robotics companies raised about $1.6 billion this year through 29 October, more than 4x all of 2025, according to Crunchbase data. Read: the Suno suit and the construction robots.
Memory makers’ share prices, indexed to the 29 October close = 100, at the 6 November close: Micron 73.9, SK hynix 74.2, Samsung Electronics 81.3, our arithmetic from the falls in our memory piece; Counterpoint Research said on 5 November that memory prices were still rising.
The Structural Read
As we read it, lenders set the price of AI risk borrower by borrower: one is a new-issue yield for a company with other businesses, the other a market yield on notes of a company whose business is renting GPUs.
As we read it, the collateral changed as well: American Compute counts $48.4 billion of new GPU-backed debt in 2026 through 10 October, about 7.3 times all of 2025, by our arithmetic.
As we read it, the rest of the day’s reporting was about who else carries the risk: Anthropic’s retail brokers and the products built around its IPO, and two Korean brokers reading memory demand in opposite ways. We do not say why prices moved.
Brett Ryan, an economist at Deutsche Bank Securities, speaking to Bloomberg, as quoted in our piece of 7 November 2026
“This is like a good excuse to kill some projects here that were already looking shaky”
Three Implications
AI BORROWERS Meta’s 2031 notes priced at 5.850% and CoreWeave’s 2031 notes traded at 16.33%, about 10.5 percentage points apart, by our arithmetic.
LENDERS American Compute counts $48.4 billion of GPU-backed debt in 2026 through 10 October, 47% of the AI infrastructure debt it tracks for the year.
ANTHROPIC’S BUYERS Anthropic’s own page lists 33 distinct broker and selling-agent names across 14 countries, by our count, and Kalshi’s IPO contract traded at 81 cents on 7 November.
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 credit prices and GPU-backed loans sat on the Compute Capacity layer of that stack, and the Starbase agreement and the Senate permit bill on the Energy & Physical and governance layers. We do not say one caused the other. The analysis was published in June 2026 and does not discuss these events.
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 day as a credit story, are our grouping, not a finding.
We do not claim that any one story moved prices. The yields compare a new-issue price with a market price; the revenue-per-megawatt figures use different definitions; several items rest on outlet reports citing people familiar with the matter or on posts on X; prediction-market prices are traders’ bets; and the two Korean brokers’ views are their own.
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.
For anyone pricing an AI borrower this week, two filed or tracked numbers sit side by side: a 5.850% new-issue yield for Meta’s 2031 notes and a 16.33% market yield for CoreWeave’s, about 10.5 percentage points apart by our arithmetic, while GPU-backed loans reached 47% of the AI infrastructure debt American Compute tracks for 2026.
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 6 and 7 November 2026, linked above; every story was built from its own primary source, which it cites and links. We re-read each piece on 7 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.
