Daily Brief Nov 2, 2026

White House withholds AI framework; a court hasn't ruled

Today's front-page cartoon: one single picture that brings together today's stories.

At a glance

1. In a records lawsuit, the White House withheld all 60 pages on its voluntary AI framework. Most of the 28 letters to AI labs listed in its filing confirm which trusted partners may use a new model before general release.

2. Only one analyst, Sravan Kundojjala, has estimated that the top five chip-equipment makers' third-quarter DRAM equipment revenue rose 80% year on year.

3. A University of Chicago team measured NVIDIA's B300 running a 70B Llama-3 model at 24% to 27% lower throughput than the previous-generation B200; so far only this one team has measured it.

Today's main line

1. The White House withholds all 60 pages on its voluntary AI framework, and most of the 28 letters on its withholding list confirm which trusted partners may use new models before general release(letter dated October 30)

Concept illustration: on a desk at left sits an open envelope with a sheet lettered "INDEX" rising out of it, showing three grey lines each followed by a thick black bar, and a person with a hair bun and a lanyard badge, standing to the right of the desk, looks down and holds the sheet with both hands. At right, against a pink background, stands a dark-framed glass-door cabinet under a sign lettered "WHITE HOUSE ONCD", and on its middle shelf is a stack of red folders lettered "AI FRAMEWORK", tied with cream string in a bow. Printed on the image: 60 pages AI framework records, White House withheld all. — On October 30, in a Freedom of Information Act lawsuit, the White House Office of the National Cyber Director attached a withholding list naming 28 letters sent to AI labs, with most confirming which trusted partners may use a new model before general release. Lab and model names are all redacted, and the list doesn't show whether the government ever rejected a customer. The grounds for withholding are the defendant agencies' own claims, not yet reviewed by the court.

The White House Office of the National Cyber Director (ONCD) withheld all 60 pages of records on its voluntary AI framework on October 30, in a Freedom of Information Act lawsuit. Our inference: AI labs have sent at least some of their early-access customer lists to the government before a new model's launch, so a company joining a lab's preview should assume its name may reach the White House.

The withholding list the government is required to attach names 28 letters sent to AI labs, one by one.

The government's stated reason for withholding says the labs supplied these lists.

The clearance letters began before the framework file's date.

Read the full item →

2. Analyst Sravan Kundojjala estimates the top five chip-equipment makers' third-quarter DRAM equipment revenue rose 80% year on year, and Tokyo Electron raised its forecast for next year's equipment market

Concept illustration: a black-framed landscape screen device sits on a dark stand, with one word in white capital letters on a black band across the top of the screen and two grey bars below it. Inside the screen, along one baseline from left to right, are a red machine lettered "DRAM" with a stack of white discs on a small ledge in its right-hand opening, a grey-brown shallow tray heaped with white discs, and a matching tray with a noticeably larger, taller heap of discs. Printed on the image: up 80% YoY Top five's third-quarter DRAM tool revenue, est.. — Semiconductor analyst Sravan Kundojjala estimated on November 1 the year-on-year growth in the top five semiconductor equipment makers' third-quarter DRAM equipment revenue. The estimate is a single analyst's tally, from Kundojjala alone. The body of Kundojjala's post doesn't say which five companies or how much each contributed.

Semiconductor analyst Sravan Kundojjala estimated on November 1 that the top five semiconductor equipment makers' DRAM equipment revenue in the third quarter was 80% higher than a year earlier. Our inference: memory chips are already short, and tools sold now add chips only after installation and ramp-up, so memory buyers such as PC makers should not expect the shortage to ease right away.

The five makers' combined third-quarter figure is a record and shows no sign of slowing, Kundojjala writes.

Tokyo Electron, Japan's largest semiconductor equipment maker, lifted the industry's spending outlook as well.

NAND flash memory is tight as well.

Read the full item →

Chips & semiconductors

No chips & semiconductors item this issue. The memory-equipment story is main-line item 2.

Named commentary

1. John Burn-Murdoch, chief data reporter at the Financial Times, and investor Gavin Baker read the same set of charts on the US post-pandemic productivity rebound and reach opposite conclusions.

Concept illustration: on a peach background, a large sign lettered PRODUCTIVITY sits at the top above a horizontal line holding a row of white box stacks that grow from a single box beside a post sign lettered USA at the far left to the tallest stack at the far right, and in the middle of the row a red-framed laptop with a speech bubble on its screen rests on a small stool, in front of a post topped with a sign lettered CHATGPT. Below the line, two black head-and-shoulders silhouettes each raise an arm, the left one pointing up-left toward the short stacks and the right one pointing up-right toward the tall stacks, with a large opening quotation mark in the bottom-left corner and a large closing one in the bottom-right corner. Printed on the image: AI payoff or older trend: two opposite readings. — John Burn-Murdoch, chief data reporter at the Financial Times, and investor Gavin Baker read the same charts on the US post-pandemic productivity rebound: Burn-Murdoch says a trend that pre-dates AI and only shows up in the US doesn't look like AI, while Baker says it is hard not to see some evidence of "ROI on AI." Baker's two supporting figures, on industry productivity and the US share of Anthropic's revenue, are relayed without their original sources.

Read the full item →

Research & models

1. The University of Chicago's Research Computing Center measured NVIDIA's B300 running a 70B Llama-3 model at 24% to 27% lower throughput than the previous-generation B200.(preprint, arXiv 2610.41919)

Concept illustration: a hand at the right holds a black, round-cornered screen sideways, with a black strip along the screen's bottom lettered "UNIVERSITY OF CHICAGO" and, on the surface above it, a beige plaque at the centre lettered "NVIDIA" beside a green emblem. Left of the plaque stands a dark grey server rack lettered "B200" with four speech bubbles containing lines above it, right of the plaque stands a brick-red rack lettered "B300" with three such bubbles above it, and a stack of lined paper sits on the outer side of each rack. Printed on the image: 24% to 27% B300 throughput below B200 on 70B, Chicago test. — In a preprint, the University of Chicago's Research Computing Center measured NVIDIA's B300 trailing the previous-generation B200 on several workloads, a result only this one team has measured. The authors also found that two B300 servers of the same model, each on a different development build of the inference software vLLM, differed by 2.1 times in time to first token, so software versions may sway the result.

Read the full item →

From the archive

1. [Looking back] (SecondSource call, July 24) Alphabet's third-quarter free cash flow was negative US$9.07 billion, and in the same quarter it raised US$28.7 billion net from bond issuance.

Concept illustration: a thick book, spine at left and page edges at right, has a cover label reading QUARTERLY REPORT and a cutaway peaked-roof building drawn on its cover, with a leader-line label reading ALPHABET. A stream of black coin-like discs enters from the left through the building's left wall and slides down and to the right along a chute into a shallow tray at the center, while a second stream starts just right of the tray and runs down a ramp to the right, piling at the foot of a tall black server rack; at lower left, beneath the chute, a tied red sack labeled DEBT sits on a flat wheeled cart. Printed on the image: -US$9.07B Third-quarter free cash flow, Alphabet-defined. — Alphabet's quarterly report, filed October 28, says Alphabet raised US$28.7 billion net from senior unsecured notes in the third quarter, an amount that covers one quarter only. Free cash flow is as Alphabet defines the measure, with third-quarter capital spending of US$51.36 billion against operating cash flow of US$42.29 billion.

Our inference: Alphabet borrowed in a quarter when its capital spending exceeded its operating cash flow, so investors tracking the large cloud companies' spending now have one quarter in which part of it looks debt-funded.

Read the full item →

Also happened — not verified by us yet

No new items in this section this issue.

Academic papers, 203: 57 read, 29 screened out (1 of them before fetching), 117 not yet read. Linked articles, 93: 52 read, 3 screened out, 38 not yet read. Social-platform posts, 89: 20 read, 38 screened out, 5 logged only, 26 not yet read. Second paper feed, 5: none read yet. Blogs, 3: 2 screened out, 1 logged only. Show transcripts (6) and newsletters (3): all read.

The government response letter and withholding list in main-line item 1 came in through the link attached to Leo Schwartz's post, and are counted among the 93 linked articles. Last night's collection pulled 266 original X posts from 374 accounts. These 266 are a separate count from the 89 above and can't be converted into each other. @teortaxesTex alone accounted for 37 of the 266, followed by @elonmusk and @mitsuhiko.

One-time catch-up. The show-transcript catch-up totals 770 pieces and the linked-article catch-up 61 pieces, dated through October 30. None of it was used in this issue.

Still missing today. The gap that most affects judgment is the 60 withheld pages themselves, none of which the government has released; the body of Kundojjala's post doesn't name the five equipment makers. The full text from the shows' official pages wasn't obtained last night. This issue has no macro section.

Tracked sources. The long-term roster currently has 529 named sources: 302 on social platforms, 90 shows, 51 news outlets, 48 blogs, 48 paper authors and 46 newsletters, with the rest spread across earnings reports, keynotes and other channels. The same source may be tracked on several channels, so the channel figures add up to more than 529. These are counts of tracked sources and can't be added to the piece counts above; the 374 accounts actually pulled last night are also not the same population as the 302 social-platform sources on the roster. Representative names include Daniel Kokotajlo, Sergey Levine, Arvind Narayanan and Chris Lattner on social platforms; Terence Tao and Lilian Weng among blogs; and Sam Altman, Dario Amodei and Demis Hassabis among show interviewees. This issue uses 15 outside sources in the body. Only outside links actually cited in the body are counted, and links to SecondSource's own site are left out; these 15 are a separate count from the 529 tracked sources. Taiwanese-side sources total 148, with 43 added in October and 7 so far in November; Taiwanese media, Taiwanese companies' official channels and local reporters in Taiwan count, while foreign outlets relaying Taiwanese media do not.

SecondSource isn't a news digest: AI reads primary sources across fields, every judgment comes with its original source, and mistakes are corrected in public. The focus is what has changed and which claims still lack evidence, not what happened today.

Written from the same research and judgments as the Traditional Chinese edition. Sources are linked; we distinguish original documents from reporting and mark what we could not verify.