Daily Brief Oct 20, 2026
How we checked the items below
For readers who want to dig in. What happened, in brief, is on the Today tab.
This issue draws on SecondSource's own research brief, produced early on October 20, and last night's sweep. All three main-line events date from October 19; the opposing view in item 1 comes from a show posted September 17. The later sections mostly date from October 15 to 19, and From the archive revisits an older topic. Last night's sweep took in 180 pieces, and this issue uses 20 outside sources with checkable links.
1. [Today] (post, October 19) SemiAnalysis forecasts about 50GW of data-center capacity added worldwide in 2027; this year's roughly US$1 trillion of signed leases covers about 20 to 33GW (our calculation)
Not proven yet： the self-build construction detail is in the paid section, which hasn't been reviewed
How we checked
The free section of the SemiAnalysis post was read in full. The paid section hasn't been reviewed, including the self-build detail and the section on whether capacity will be overbuilt. Both inputs to the conversion come from SemiAnalysis alone, and SemiAnalysis sells its own data-center model. One passage of the post says this year's signed leases are "slightly under USD 1T." Another says "over a trillion USD" has been signed and closed. The calculation uses a round US$1 trillion. Gerstner's words come from Happy Scribe, a third-party machine transcript. Gerstner does not name the year in that line. The show's context points to 2027.
2. [Today] (posts, October 19) SB 53 author Scott Wiener and Public First head Brad Carson publicly criticize Anthropic's leaked federal AI safety draft for having too few guardrails
Not proven yet： every reading of the clauses comes from critics; the full draft hasn't been reviewed
How we checked
Carson's and Wiener's original posts and Hashim's two comments were read, and the quotes were checked word for word against archived copies. Both descriptions of the clauses are critics' readings. The full draft and Transformer News's analysis haven't been reviewed. Public First has its own federal legislative proposal.
3. [Today] (announcement, October 19) Tenstorrent, the AI chip company Jim Keller runs as CEO, raises US$1.5 billion at a US$6.2 billion post-money valuation
Not proven yet： the press release gives no revenue, customer count or shipment figures
How we checked
The newsroom announcement was read in the original; it is Tenstorrent's own first-hand announcement. June's unit and customer counts are the CEO's self-report and haven't been independently verified. The announcement doesn't say how the money splits between servers and design licensing.
[Today] (post October 19) OpenAI researcher Jason Wei thinks AI won't replace radiologists soon, because AI's abilities are uneven and demand for reads is elastic; this is Wei's personal judgment, with no data attached.
How we checked
[Today] (release October 19) Chinese AI model company MiniMax releases M3.1 Flash at US$0.15 per million input tokens and US$0.60 per million output tokens through its API, and self-reports a third of the cost and 3 times the speed of MiniMax's M3.
How we checked
The accusation rests on that one post alone (@dummerspast39, 2026-10-19).
[Today] (announcement October 19) AI code-review company Greptile announces a US$50 million Series B at a US$750 million valuation, led by Menlo Ventures.
How we checked
Greptile self-reports reviewing more than 1% of the new pull requests opened on GitHub each day (Greptile, 2026-10-19).
[Looking back] (SecondSource deep dive of October 8, on a September 8 experiment by Harvey and Baseten) In Harvey and Baseten's test, long-document AI scores rose with how much of the data room the system read. Across seven models, a recursive harness averaged 62.4% on a simulated M&A due-diligence task, against 23.3% for a simple tool loop.
How we checked
The numbers are a single self-report from Harvey, graded by a model.
Sources & accounting
The past 24 hours. 180 new pieces came in from October 19 to 20: 118 social-platform posts, 37 company or personal blog posts, 13 academic papers, 8 subscription newsletters, 3 macroeconomic data releases and 1 show transcript. Of those 180, 41 were read in full and 48 were screened out as off-topic; 91 haven't been read yet. Among the 48 screened out, 8 were judged not worth a close read from the title and opening. All three arXiv fetchers hit access limits, so no new papers came in; the next round will fetch them again. The Research watch item comes from preprints submitted in mid-October that had piled up over the previous few days. None of the 37 blog posts has been read yet.