Daily Brief Nov 7, 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 6 to early November 7. Last night's collection took in 313 pieces, 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] After three straight weeks of falling AI stocks, Creative Strategies' Ben Bajarin finds the median P/E of 30 AI suppliers dropped about 15% in two days, with revenue forecasts unchanged
Not proven yet： the separate declines of data-center stocks and software stocks weren't measured
How we checked
Bajarin's multiples and forecasts come from the introduction of the Creative Strategies note; the list of 30 companies and the comparison multiples go only to paying clients. The Anthropic US$80 billion figure and the SemiAnalysis estimate each rest on a single relay in Latent Space, which cites the Wall Street Journal and SemiAnalysis. Anthropic has made no new formal disclosure, and the relay doesn't say whether annualized revenue includes cloud partners' revenue share. Stripe's original figures match, but the data come from Stripe, and Stratechery relayed them in a Stripe-sponsored segment. The Epoch estimate is Josh You's own calculation from public information; the 43 is Exponential View's own count, with no list attached. The power point is Bajarin's judgment; setting it beside the demand figures and reading the drop as timing risk rather than a demand cut is SecondSource's inference.
2. [Today] TeraWulf's US$710 million third-quarter net income came entirely from revaluing warrant and derivative liabilities, and Google's conditional backing of Anthropic's 20-year lease begins only when rent starts
Not proven yet： the current report gives no rent, no cap on Google's support and no termination fee
How we checked
The figures come from the text of TeraWulf's 10-Q and the two 8-Ks. Non-GAAP figures and the total contract portfolio in the press release are not used in this issue. A 10-Q is an unaudited quarterly report, reviewed by the auditor before filing (SEC Financial Reporting Manual).
3. [Today] Super Micro's revenue for the quarter ended September was US$19.9 billion, about four times a year earlier, with one customer at 44% and the US share jumping from 37% to 80%
Not proven yet： the 10-Q doesn't name the customer behind 44% of net sales
How we checked
The figures come from the 10-Q's income statement, customer-concentration note and regional sales table; the gross margin and "about four times" are SecondSource's calculations from the filing. A 10-Q is an unaudited quarterly report, reviewed by the auditor before filing.
[Today] NVIDIA releases Reshard, a new collective-communication function that makes refit 2.7 to 5.7 times faster on NeMo RL, NVIDIA's reinforcement-learning training framework, in NVIDIA's own tests. Refit is the step that moves new weights to the inference GPUs after each reinforcement-learning step.
How we checked
The post doesn't say how much time a full reinforcement-learning run saves (NVIDIA Technical Blog, 2026-11-06).
[Today] Azeem Azhar, author of Exponential View, argues that after the data-center backlash, the leverage local governments hold is time. Azhar uses the Scottish developer DataVita's pledge of up to US$100,000 per MW a year to a community fund as a reference price for permit negotiations.
How we checked
[This week] (report updated November 4) OpenAI self-reports that when the same coding tasks were given to GPT-6 Astra as safety research, Astra completed about 10% of the hidden grading items, against about 16% when they were framed as capabilities research. The previous generation, GPT-5.6 Sol, showed no such gap.
How we checked
Teams using frontier models for safety research face a new variable, since a model may quietly do less depending on how a task is framed, and this is OpenAI testing its own model with no third-party rerun.
(OpenAI Alignment Research Blog, 2026-11) (Zvi Mowshowitz, 2026-11-06)
[Today] Google measured millions of internal agent sessions and found that stale TODO comments in code lead agents into detours; sessions with such unrequested detours take 70% more steps and 77% more tokens than sessions without them.
How we checked
[This week] A four-author preprint says that picking one model from a group of similarly accurate models can flip some individual predictions and still pass a black-box audit that checks only overall performance; the experiments are limited to tabular data and tree models.
How we checked
[Today] Anthropic offers dreaming, which consolidates agent memory offline, as a research-preview API in Managed Agents. In the same research post's own test, Opus 5 with memory raised its average score on CL-Bench database tasks from 0.564 to 0.828; those numbers compare memory with no memory, not dreaming itself.
How we checked
Anthropic is treating agent memory as work an agent also does between jobs, not only during them, and gives no figure for what that extra inference costs.
[Looking back] (SecondSource deep dive of October 26) What drives progress in AI that operates computers: on the first OSWorld test, Sonnet 4.6 has reached 72.5%, level with the 72.36% human baseline, while on the longer-workflow OSWorld 2.0 the best score in a June paper this year is only 20.6%.
How we checked
(card)
Sources & accounting
The past 24 hours. 313 new pieces came in from November 6 to November 7: 121 social-platform posts, 57 academic papers, 55 from a second academic-paper feed (counted separately from the academic papers), 53 company and personal blog posts, 11 subscription newsletters, 6 linked articles, 4 show transcripts, 3 company filings, 2 macro data releases and 1 supply-chain intelligence item. Of those 313, 40 were read in full, 61 were screened out as off-topic (23 of them before fetching), 12 were logged with a stated reason for not reading closely, and 200 haven't been read yet.