Daily Brief Nov 18, 2026 · ← Back to this issue
How we checked every item today except "Also happened"
For readers who want to dig in. For what happened, in brief, go back to this issue.
This issue draws on material from November 17 to early November 18. Last night's collection took in 649 new pieces, 108 of them read in full, and this issue uses 27 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 guarantees for customers total up to US$118 billion, US$105 billion of it on one campus
Not proven yet: the guarantee cap and equity commitments have no version from after Huang's resignation
How we checked
The guarantees, the absence of payments and the impairment are from the SEC-filed version of NVIDIA's CFO commentary; the US$118 billion and US$105 billion figures match CNBC's report, and the rest appear only in that one document. The other two guarantee items are US$9.5 billion of credit support for power and systems, and US$3.5 billion backing land, power and buildings for new cloud providers that rent out AI compute. The Ohio campus is about 4.25 GW across nine buildings and houses only NVIDIA equipment. This quarter's CFO commentary doesn't list last quarter's impairment, so the available data can't tell whether the US$1.8 billion is a new signal.
2. [Today] NVIDIA still issues guidance after Jensen Huang's resignation: US$130 billion next quarter, above analysts' US$125.1 billion
Not proven yet: the 75% growth figure is Kress's spoken claim; the press release doesn't give it
How we checked
Revenue, guidance and gross margin are from the SEC-filed version of NVIDIA's press release. The press release gives guidance only for next quarter, with no figure for the next fiscal year (NVIDIA press release, 2026-11-17); the 75% growth figure is Kress's spoken claim, according to CNBC's live coverage. CNBC is also the source for Jones's words and the analyst estimates. The 13-week adjustment is More Than Moore's; that newsletter's financial tie to NVIDIA is noted in main-line item 1. This quarter ended before Huang's resignation, so the figures reflect operations under Huang. Huang's statement is from his post on X. According to CNBC, the stock rose nearly 4% after hours.
3. [Today] OpenAI releases GPT-6.2, cutting the flagship model's list price 20% from GPT-6 Astra
Not proven yet: every score that beats Fable 5.2 is OpenAI's own
How we checked
List prices and benchmark scores are from the original text of OpenAI's release page. GPT-6 Astra was priced at US$10 for input and US$50 for output. The everyday GPT-6.2 Sol keeps GPT-6.1 Sol's US$2 and US$10, and both models are available in ChatGPT, Codex and the API. The previous prices and the match with Fable 5.2 are from the summary by Simon Willison, who has long tracked model pricing across vendors.
4. [Today] A US Commerce Department rule extends export controls to remotely rented AI compute
Not proven yet: the thresholds and countries come from McGuire and Low; BIS's press release lists neither
How we checked
That the rule has been issued, its purpose and the agencies that wrote it are from the original text of BIS's press release; the press release lists no thresholds, countries or effective date (BIS press release, 2026-11-16). The thresholds and countries come only from analysts: McGuire's account is from the ChinaTalk transcript and was spoken; Low's summary is as cited on X. McGuire says the full rule runs 132 pages. McGuire and his fellow guest on the episode, Julian Gewirtz, are both former National Security Council officials in the Biden administration who favor strict controls. On the length of the transition period, the show's host said about 90 days, consistent with Low's summary.
[This week] NVIDIA announced multi-year supply deals with substrate maker Kinsus and chip packaging-and-test company Amkor, and Kinsus aims to expand its ABF substrate capacity to more than 3.5 times today's level by the third quarter of 2028.
How we checked
[Today] According to More Than Moore's account of the earnings call, NVIDIA said the Vera Rubin platform brought in nearly US$20 billion of revenue in its first quarter of volume shipments, NVIDIA's fastest product ramp ever.
How we checked
[Today] Beth Barnes, CEO of METR, a group that evaluates the risks of frontier models, says that while embedded at OpenAI, METR could talk only to employees OpenAI approved, and could not quote or summarize what they said.
How we checked
[Today] Epoch AI, an independent research group that runs its own benchmarks on AI models, found that GPT-6.2 Astra can chain 12.3 steps internally without writing out its reasoning, against 10.5 steps for the previous GPT-6 Astra.
How we checked
[This week] According to the paper's abstract, author Abhishek Rai reports that open-weight agents that write scripts to call tools use 70% to 90% fewer tokens than agents calling MCP tools one at a time.
How we checked
Beyond a thousand tools, the savings come from the script-writing approach as a whole, while an added routing layer that pre-filters tools gives no significant help, and only open-weight models were tested.
[This week] According to the paper's abstract, Zhen Xu and four co-authors find that after a tool interface changes, agents carrying old experience keep calling it the old way, report success, and do worse than agents with no experience; the abstract gives no figures.
How we checked
[This week] (event date November 12) OpenAI named Aleph the exclusive reseller of ChatGPT ads, so brands and agencies headquartered in 44 markets can sign contracts locally and pay in local currency.
How we checked
What's new is a buying channel where advertisers are based, not new countries where ads can run, and OpenAI's own count of more than 100,000 marketers who have bought its ads has no second source. The post doesn't say what kind of company Aleph is, how big it is, how long the exclusivity lasts or how revenue is split.
[Today] Perplexity, the AI search company, launched Orchestrator API, one API for more than 85 models; in its own test, passing Claude Fable 5.2's refusals to an OpenAI model solved 41 of 63 tasks, against 38 for Fable 5.2 alone.
How we checked
[Follow-up] (deep dive, August 29, 2026) OpenAI raised its flagship API list price in September, and GPT-6.2 Astra, released November 17, cuts it by 20%.
How we checked
Also happened (2 items): not verified by us yet
Still missing today. The gap that most affects judgment is last quarter's version of the same guarantee table and impairment figure, which this issue has no comparison against; the call's Q&A exists only in two relays, More Than Moore's and CNBC's. The thresholds, countries and effective date in the BIS rule text haven't been checked against the two analysts' accounts. The benchmark operators' leaderboards for GPT-6.2 are outside this issue's sources. The Megaspeed shipping records and the TensorWave contract appear only in posts on X, which is why both sit in Also happened. Both research papers rest on their abstracts; model versions and per-task figures are in the full papers. Of the 404 pieces not yet read, 267 are academic papers and 51 are social-platform posts.
Sources & accounting
The past 24 hours. 649 new pieces came in from November 17 to November 18: 339 academic papers, 151 social-platform posts, 55 company or personal blog posts, 55 from a paper-summary site (counted separately from the academic papers), 21 originals linked from posts and articles, 14 company filings, 12 subscription newsletters, 1 podcast transcript and 1 company filing change. Of those 649, 108 were read in full, 128 were screened out as off-topic (15 of them before fetching), 8 were only indexed without a full read, 1 is waiting for its full text to be refetched, and 404 haven't been read yet. Academic papers, 339: 21 read, 51 screened out, 267 not yet read. Social-platform posts, 151: 44 read, 52 screened out (1 of them before fetching), 3 only indexed, 1 waiting for a refetch, 51 not yet read. Blogs, 55: 12 read, 6 screened out, 2 only indexed, 35 not yet read. Paper-summary site, 55: 13 screened out (all 13 before fetching), 42 not yet read. Linked originals, 21: 18 read, 3 screened out (1 of them before fetching). Company filings, 14: 1 read, 1 screened out, 3 only indexed, 9 not yet read. Newsletters, 12: 10 read, 2 screened out. The podcast transcript and the company filing change were both read. Named sources: Gene Munster, Beth Barnes, Ana Swanson, Beth Kindig, Aravind Srinivas and Epoch AI on social platforms; More Than Moore among newsletters; and the ChinaTalk episode on the BIS rule among shows. NVIDIA's quarterly press release and CFO commentary are not among these 649 pieces; they were read separately from the SEC filings. Last night's scan covered arXiv, X, blogs and newsletters; one blog source failed to fetch, and no podcast transcripts other than ChinaTalk's came in last night.
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. Among the company filings collected last night, several from ASML, Intel, Meta and Marvell are older filings from November 2025 to early November 2026, and this issue doesn't treat them as news. The August deep dive and OpenAI's September prices cited in From the archive are not among the 649 pieces above either.
Source concentration. Main-line items 1 and 2 both come from NVIDIA's documents of the same evening, with everything said in the call's Q&A relayed by More Than Moore and CNBC; both Chips & semiconductors items also come from the same earnings release. Main-line item 3's prices, benchmarks and safety document are all OpenAI's own documents, with outside comparison only from Epoch AI, Simon Willison and Every. Main-line item 4's rule details come from one guest on a single podcast episode and one analyst's summary. Both research papers rest on arXiv preprint abstracts, with results reported by their authors.
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. Representative names include Gene Munster, Ethan Mollick, Satya Nadella and Gary Marcus on social platforms; Ian Cutress and Zvi Mowshowitz among newsletters; and Jordan Schneider among shows. This issue uses 27 outside sources in the body, counting only outside links cited in the body. Sources from Taiwan total 281, with 44 added in October; Taiwanese media, Taiwanese companies' official channels and local reporters in Taiwan count, while foreign outlets relaying Taiwanese media do not.