Daily Brief Oct 13, 2026
For readers who want to dig in. If you only want what happened, go back to Today.
This issue draws on the research brief we write ourselves, produced early on October 13, and last night's sweep. Both main-line items happened on October 12, as did most items in the later sections; the exceptions are the power-grid spec, finalized September 17 and made public at a conference on October 12, and From the archive. We swept 336 pieces overnight, and this issue uses 23 outside sources with links you can check.
Not proven yet： neither announcement gives performance figures or mentions cloud customer orders
we read the AMD press release and the NVIDIA blog post in the original, and checked the quoted lines in both word for word. They are two independent primary sources, but both are vendors announcing their own products. We read nothing beyond those two documents. AMD lists eight partners, including Cisco, Dell and Foxconn; that is endorsement, not orders. Huang's remark comes from a March interview in the paid publication Stratechery, not from this announcement (Stratechery, 2026-03). We have no verbatim quote from that interview. What would prove this wrong: no large cloud company publicly buys these CPU-only racks by the first half of 2027. That deadline is one we set ourselves. A nearer thing to watch is whether AMD names a cloud customer when it starts shipping in November.
Not proven yet： we haven't seen Every's task set or how it scores them
what we read were archived copies of three posts. We didn't read the full table on Artificial Analysis's site, Every's page describing the benchmark, or Meta's official blog. Each outside reading comes from a single organization's run. Shipper also doesn't say whether he tested the fast or the full version. Miles Brundage, formerly OpenAI's head of policy research, reported the same day that he had bought a paid plan and couldn't tell which version he was using (Miles Brundage, 2026-10-12). So the outside tests may not have hit the same version either. ⚠️ Disclosure: our analysis is produced with help from Anthropic's models, and Claude Opus 5.5 is one of the models compared in this item.
The 40% doesn't say which GPU system it compares against. There is no outside measurement, and we didn't read the test conditions.
(Microsoft Azure, 2026-10-12) (Satya Nadella, 2026-10-12) (Microsoft official blog, 2026-10-12)
Dan Nystedt, a reporter who has long covered Taiwan's chip industry, relayed Taiwanese media: Nanya's third-quarter gross margin was 85.7%, its average selling price rose 60% from the previous quarter, and revenue was NT$133.65B (Dan Nystedt, 2026-10-12). We didn't check the financial report.
The spec doesn't give a reason for this rule.
This is his personal judgment, with no data attached.
⚠️ Disclosure: our analysis is produced with help from Anthropic's models, and Claude Code and Fable 5.1 are both Anthropic products.
We read only the abstract.
We didn't read how the semantic-search setup was configured, so this isn't necessarily a verdict on the two methods themselves.
The research group Epoch AI estimates that the capability gap between Chinese and US models averages seven months, within a range of 4 to 14 months. We didn't go back and check this round, so there's no result yet.
What you are not getting today. What would most affect judgment is AMD's and NVIDIA's performance figures: neither company has published them, so we haven't read them. Beyond that, we didn't read Every's task set, the full table on Artificial Analysis's site, Meta's official blog, METR's original survey, Nanya's financial report, the full stories from The Information and Bloomberg, or the full TestSuiteBench paper.
The past 24 hours. 336 new pieces came in from October 12 to 13: 168 arXiv academic papers, 127 social-platform posts, 26 company or personal blog posts, 9 papers from other sources and 6 subscription newsletters. Of those 336, we read 98 and screened out 67 as off-topic. The other 171 haven't been read yet and stay on the reading list. Of the 171 unread, 120 are papers, 22 are blog posts and 20 are social-platform posts. No show transcripts came in during the window: two show sources weren't fetched last night, and nine YouTube channels had no new episodes fetched. Five more sources failed to fetch last night: Data Center Dynamics, DeepSeek's official blog, the Thinking Machines blog, and the investor news pages of Broadcom and Qualcomm. If any of them posted something new that night, we didn't receive it. One piece from last night also couldn't be loaded because of a formatting problem. So anything we describe as within what we swept leaves out all of the above. The AMD, NVIDIA and Microsoft posts behind the main line and the chips section, and the OCP spec, are originals we read directly today.
One-time catch-up. No new one-time catch-up of older material today.
A note on source concentration. ⚠️ All three readings in main-line item 2 come from posts on X, and each of the two outside tests comes from a single organization. ⚠️ Main-line item 1 and the Maia and power-grid items in the chips section all come from vendor announcements around the same OCP summit on October 12. ⚠️ Our analysis is produced with help from Anthropic's models: Claude Opus 5.5, a model compared in main-line item 2, and Claude Code and Fable 5.1 in the first Model watch item are all Anthropic products.
The sources we track. Our long-term roster 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. ⚠️ Those are counts of tracked sources, a different population from the 336 new pieces above; they can't be added or compared. Representative names: on social platforms, Alexandr Wang, Ben Horowitz, Teortaxes, Ben Bajarin and Gary Marcus; among newsletters, Import AI, Exponential View and ChinAI; among shows, 20VC, All-In, No Priors and Latent Space. This issue uses 23 outside sources in the body. We count only body links not on our own domain, a different population from the 529 tracked sources above. We added 7 Taiwanese sources in September, bringing the Taiwanese total to 127; Taiwanese media, Taiwanese companies' official channels and local Taiwanese reporters count; foreign outlets relaying Taiwanese media do not.