Daily Brief Nov 12, 2026 · ← Back to this issue
How we checked the items below
For readers who want to dig in. For what happened, in brief, go back to this issue.
This issue draws on material from November 11 to early November 12. Last night's collection took in 409 pieces, 61 of them read in full, and this issue uses 48 outside sources with checkable links; the later sections also draw on material from the past week. The full edition of this issue is the archive of record.
1. [Today] NVIDIA's board voiced confidence in Jensen Huang on November 8, then met three days later on whether to keep Huang, and no company document covers the meeting
Not proven yet： the reports don't say whether the outlets' anonymous sources are the same people
How we checked
Everything in this item comes from media reports and reporters' posts; there is no company document. CNBC's report is the outlet's own, and it says CNBC confirmed the meeting itself. The Journal's account is taken from a post on the paper's official X account. The tech outlet The Next Web, relaying Reuters and Bloomberg on November 12, wrote that the evening ended with no decision announced (The Next Web, 2026-11-12). The reports don't say whether the outlets' anonymous sources are the same people, so they can't count as several independent sources. None of these accounts says whether the board voted or adjourned.
2. [Today] Anthropic's IPO roadshow video says Claude was ported to AMD GPUs in three days and gives no performance figures; AMD is also buying Anthropic shares at the IPO
Not proven yet： the three days is Anthropic's own claim in a fundraising video; no one has measured performance
How we checked
The main source is developer Simon Willison's blog. Willison transcribed the roadshow video with a speech-to-text tool and quoted from it; there is no official transcript. That the speaker is cofounder Tom Brown, and that the four families are AWS Trainium, Google TPUs, NVIDIA GPUs and AMD GPUs, are Willison's notes. The three days and the "dozens of engineers months of work" are both Anthropic's own claims in a fundraising video, with no work-hour records and no third-party measurement. The video says only that Claude runs, gives no performance figures and doesn't say the AMD setup is serving customers. Left unexplained are how "basically no human intervention" is counted, how many engineers AMD contributed, which kernels "almost 100% of the kernels" refers to, and which generation of AMD GPU was used. AMD's roughly US$2.5 billion subscription comes from the offering summary of Anthropic's amended IPO registration, shown at the midpoint price.
3. [Today] A reporter at The Information says the US Commerce Department could finalize rules this month restricting Chinese customers' remote rental of advanced chips; the text isn't public
Not proven yet： the posts don't say which chips and countries the rules cover or when they take effect
How we checked
All sources in this item are X posts by reporters introducing their outlets' reports; everything above is what Schwartz, Lessin, Caputo and Swanson said in those posts. The posts don't say which chips and countries the rules cover, when they take effect, or whether US cloud providers' Southeast Asian regions are included; being finalized doesn't guarantee the rules will be issued. Caputo's post says Trump made the decision on "Thursday" without giving the date. Shakeel Hashim, editor of the newsletter Transformer, offered another reading of the Axios report as commentary: someone around Trump may want Lutnick to take the blame (Shakeel Hashim, 2026-11-11).
4. [Today] Venture investor Tomasz Tunguz estimates that 40% of gross profit across the AI value chain lands with applications, and the 28% for compute rental providers is before GPU depreciation
Not proven yet： CoreWeave CEO Intrator's new contract pricing is relayed only by Tunguz
How we checked
The figures below come from Tunguz's blog post and CoreWeave's third-quarter press release. This is one author's model. Tunguz states in a footnote that the conversions rely on approximations, not audited disclosures, and that the inference layer assumes open-weight models, not frontier labs' APIs. Of the 8% for power and data centers, the 3% for power is utilities' revenue, not gross profit. Tunguz says so in a footnote. The post doesn't disclose how Tunguz's portfolio relates to these layers.
[Today] October revenue at four Taiwanese server assemblers was above a year earlier at all four, from 46.02% higher at Inventec to 117.56% at Wistron; Wiwynn and Inventec fell from September.
How we checked
(Taiwan Stock Exchange, 2026-11) (Taiwan Stock Exchange, 2026-11) (Taiwan Stock Exchange, 2026-11) (Taiwan Stock Exchange, 2026-11) (Taiwan Stock Exchange, 2026-11-11)
[Today] Quanta's CFO says US states' limits on data-center construction haven't led any customer to change orders so far, and fourth-quarter rack shipments may dip slightly.
How we checked
(Anue, 2026-11-11) (Commercial Times, 2026-11-12) (Economic Daily News, 2026-11-11)
[Today] TF International Securities analyst Ming-Chi Kuo says Jensen Huang looks unlikely to come through unscathed, and the pragmatic course is for NVIDIA to gradually reduce its reliance on Huang's personal decisions and prepare a succession.
How we checked
This is Kuo's personal judgment with no data attached, naming no candidate or timeline, and Kuo doesn't disclose holdings or client relationships. Read alongside main-line item 1: whether the board has decided is still unknown, and this item is about what the supply chain will watch whatever the outcome.
[Today] Commentator Zvi Mowshowitz, after reading the system card Anthropic released with its new models, argues that Anthropic should determine now that the new model Mythos 5.2 has reached the AI R&D acceleration threshold Anthropic set.
How we checked
According to Mowshowitz, Anthropic estimates internally that AI speeds up its R&D 1.88 times, against a threshold of 2 times that Anthropic set.
[Today] Epoch AI, an independent research group that tracks AI compute and capability, estimates that the AI compute six major chipmakers add each quarter is growing 2.5 times a year, down from about 6 times in 2023 by Epoch's estimate.
How we checked
The sign of a slowdown isn't statistically significant, and these are Epoch's estimates, counted when chips are sold. Epoch also estimates that the compute in Google TPUs supplied by Broadcom rose from about 960,000 to about 1.97 million H100 equivalents per quarter, and infers that about 40% of this quarter's TPU revenue came from Anthropic; that share assumes all earlier TPU revenue came from Google (Epoch AI, 2026-11-11).
[This week] A preprint has a strong model write a skill library for a weaker one, and when the teacher profiles the student model first, task success reaches 86.46%, against 81.04% without profiling.
How we checked
The abstract doesn't give the task set or the number of tasks; the results are the authors' own and no one else has repeated them (arXiv 2611.13144, 2026-11-09).
[Today] Databricks has made Genie Ontology, which organizes enterprise context for data agents, generally available, and on a benchmark Databricks ran, Genie scores 91%, against 58% and 45% for two unnamed rivals from frontier AI labs.
How we checked
Databricks and partner Turing built and scored the test, so it isn't a third-party benchmark.
[Today] CoreWeave, a cloud provider that rents out GPUs, has made CoreWeave Database, a managed PostgreSQL service that runs inside customers' clusters, generally available, and the announcement gives no performance or pricing figures.
How we checked
[Looking back] (SecondSource's public reading of November 6) On Alibaba's open-source Qwen3 models, penalizing only the final output during training, without touching the chain of thought, doesn't keep the chain of thought monitorable.
How we checked
On this reading, teams post-training on Qwen3 can't treat "training didn't touch the chain of thought" as evidence that the chain of thought stays monitorable, and both papers are self-reported experiments that can't be extrapolated to frontier models.
Still missing today. The gap that most affects judgment is the NVIDIA board's conclusion: as of this issue's cutoff, NVIDIA has made no announcement and no filing. The Wall Street Journal's, Reuters' and Bloomberg's reporting on the meeting appears here only through the outlets' official posts and other outlets' relays. The text of the Commerce Department's remote-use rules hasn't been published. Anthropic's roadshow video has no official transcript, and there is no public measurement of performance after the port. Intrator's remark rests only on Tunguz's relay; it hasn't been checked against CoreWeave's earnings call transcript. Quanta's earnings call content comes from three Taiwanese outlets' accounts. The system card figures Mowshowitz cites haven't been checked against the system card itself. Of the 239 pieces not yet read, 79 are academic papers and 68 are blog posts.
Sources & accounting
The past 24 hours. 409 new pieces came in from November 11 to November 12: 150 social-platform posts, 108 academic papers, 83 company and personal blog posts (including Taiwan Stock Exchange material-information filings and monthly revenue), 55 from a second academic-paper feed (counted separately from the academic papers), 6 subscription newsletters, 5 linked articles and 2 show transcripts. Of those 409, 61 were read in full, 102 were screened out as off-topic (23 of them before fetching), 7 were only indexed without a full read, and 239 haven't been read yet. Social-platform posts, 150: 40 read, 50 screened out, 2 only indexed, 58 not yet read. Academic papers, 108: 5 read, 24 screened out, 79 not yet read. Blogs, 83: 8 read, 2 screened out, 5 only indexed, 68 not yet read. Second paper feed, 55: 21 screened out (all 21 before fetching), 34 not yet read. Newsletters, 6: 4 read, 2 screened out. Linked articles, 5: 2 read, 3 screened out (2 of them before fetching). Both show transcripts were read. No source failed to fetch last night; this window had no new SEC filings or macro data.
One-time catch-up. No new one-time catch-up material came in this window. The catch-up material already collected is dated between July 1 and November 9 this year, counted separately from the past 24 hours above, and none of it was used in this issue.
Source concentration. Of the four main-line items, items 1 and 3 rest mainly on X posts by reporters and outlets, and whether the anonymous sources are the same people is unknown. Main-line item 3 and Also happened item 1 both come from posts by reporters at The Information. Main-line item 2, Named commentary item 2 and Research & models item 1 all have Anthropic as one of the parties. Both Product moves items are the companies' own announcements, and the three outlets in Chips & semiconductors item 2 relay the same earnings call.
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, a different list from last night's piece counts above, so the two can't be added or compared. Representative names include Joe Weisenthal, Dan Nystedt, Ming-Chi Kuo and Paul Triolo on social platforms; Zvi Mowshowitz and Ben Thompson among newsletters; and Lisa Su and Satya Nadella on the earnings roster. This issue uses 48 outside sources in the body. Only outside links actually cited in the body are counted; these 48 are separate from the 529 tracked sources. Sources from Taiwan total 222, with 43 added in October; Taiwanese media, Taiwanese companies' official channels and local reporters in Taiwan count, while foreign outlets relaying Taiwanese media do not.