> ## Content Index > Fetch the complete content index at: https://www.thenewway.ai/llms.txt > Use this file to discover other available public pages before exploring further. # NVIDIA makes its Hugging Face deal official at $12.93 billion - URL: https://www.thenewway.ai/nvidia-makes-its-hugging-face-deal-official-at-12-93-billion/ - Published: 2026-09-03T14:43:39.000Z - Updated: 2026-09-04T13:33:39.000Z - Description: NVIDIA makes its Hugging Face deal official at $12.93 billion, a week after we ran the report as unconfirmed. Plus Gemini 3.8 Flash and a Cyber sibling, Anthropic's background computer use, Cursor's self-hosted machines, and Thoughtworks' CTO on reviewing AI code. - Author: The New Way NVIDIA made its Hugging Face deal official this morning: $12.93 billion, under Jensen Huang's own byline, a week after we ran the report as unconfirmed. Google shipped Gemini 3.8 Flash and a security-focused sibling that Chrome's own team says produces 2.6 times more correct patches than the best commercial models. Anthropic turned on background computer use, so Claude can drive your Mac while you keep typing. Cursor moved cloud-agent execution onto machines you control, while its own post says tool outputs still flow back to Cursor. And Thoughtworks' CTO argues the fix for AI code outrunning review is to move the conversation earlier. Five stories from a 22-hour window. The biggest was a rumor when we last wrote about it. 📌 **In this issue** • [NVIDIA makes the Hugging Face deal official at $12.93B](https://www.thenewway.ai/nvidia-makes-its-hugging-face-deal-official-at-12-93-billion/#nvidia-makes-its-hugging-face-deal-official-at-1293-billion) • [Google ships Gemini 3.8 Flash and a Cyber sibling](https://www.thenewway.ai/nvidia-makes-its-hugging-face-deal-official-at-12-93-billion/#google-ships-gemini-38-flash-and-a-cyber-sibling-that-patches-vulnerabilities) • [Anthropic turns on background computer use in the desktop app](https://www.thenewway.ai/nvidia-makes-its-hugging-face-deal-official-at-12-93-billion/#anthropic-turns-on-background-computer-use-%E2%80%94-claude-runs-your-mac-while-you-work) • [Cursor runs cloud agents on machines you control](https://www.thenewway.ai/nvidia-makes-its-hugging-face-deal-official-at-12-93-billion/#cursor-runs-cloud-agents-on-your-machines-but-outputs-still-flow-back-to-cursor) • [Thoughtworks' CTO argues against reviewing all AI code](https://www.thenewway.ai/nvidia-makes-its-hugging-face-deal-official-at-12-93-billion/#thoughtworks-cto-argues-we-shouldnt-be-reviewing-all-this-ai-code) --- ## The open-model hub gets an owner ### NVIDIA makes its Hugging Face deal official at $12.93 billion Jensen Huang announced it under his own byline on NVIDIA's blog: NVIDIA has agreed to acquire Hugging Face for $12,930,300,000\. When we ran this on August 27, it was The Information's report and nobody was confirming it. Now the acquirer is. Huang's post carries three commitments: Hugging Face "will remain an open platform for the entire AI ecosystem," it will keep supporting open-weight models from every builder on any cloud or accelerator, and "NVIDIA compute will not be required to build on or deploy through Hugging Face." He also says Hugging Face's CEO, Clem Delangue, approached him. The numbers he cites for what NVIDIA is buying: more than 18 million developers, 3 million models, 500,000 datasets and 200,000 companies. The post names no closing date and no regulatory review. Bookmark the compute line. It's the one to check in a year. [NVIDIA to Acquire Hugging FaceNVIDIA has agreed to acquire Hugging Face. Together, we will scale Hugging Face’s platform, strengthen its infrastructure and expand access to AI for developers and institutions worldwide.![](https://storage.ghost.io/c/9c/1f/9c1ff71c-2859-44c2-a962-1337982a6484/content/images/icon/favicon-43f19893-a969-43f4-80d6-9f913d5c9a95.ico)NVIDIA BlogJensen Huang![](https://storage.ghost.io/c/9c/1f/9c1ff71c-2859-44c2-a962-1337982a6484/content/images/thumbnail/hf-nvidia-partner_hf-nvidia-partner-press-1920x1080-2-2dd9b632-d7c4-4615-a8e0-caccb45a25b1.png)](https://blogs.nvidia.com/blog/nvidia-to-acquire-hugging-face/?ref=thenewway.ai) --- ## A new model built to find and fix security holes ### Google ships Gemini 3.8 Flash and a Cyber sibling that patches vulnerabilities Google's own figures for the security sibling, Gemini 3.8 Flash Cyber: a 47.2% first-try patch rate on the external CWE-Bench, and 2.6 times more correct Chrome patches than the best commercial models, per Chrome's security team. Flash keeps 3.7's price, $0.75 and $3.75 per million tokens, until December 31, then doubles. Cyber is gated to Google's Fairwind Program. Google's Kerbal demo of Flash drew a Community Note: the rocket was a pre-built in-game model, piloted by an uncredited mod. [Introducing Gemini 3.8 Flash and 3.8 Flash CyberGemini 3.8 Flash and 3.8 Flash Cyber deliver next-generation intelligence for agentic workflows and cybersecurity.![](https://storage.ghost.io/c/9c/1f/9c1ff71c-2859-44c2-a962-1337982a6484/content/images/icon/favicon-264b051e-ded3-41b9-b6cb-bf16655a1c46.ico)GoogleTulsee Doshi![](https://storage.ghost.io/c/9c/1f/9c1ff71c-2859-44c2-a962-1337982a6484/content/images/thumbnail/gemini-3-8_flash__blog__header__16-9__light.width-1300-744e086a-d485-4e35-99d7-bbf2eb6689da.png)](https://blog.google/innovation-and-ai/models-and-research/gemini-models/3-8-flash-and-3-8-flash-cyber/?ref=thenewway.ai) --- ## The agent works while you do ### Anthropic turns on background computer use — Claude runs your Mac while you work Anthropic's official accounts announced it: computer use now runs in the background in Claude Cowork and the Claude Code desktop app. Claude clicks, types and opens apps while you keep working. It's on already if you've used computer use before, otherwise Settings > General. Beta, Pro and Max plans, macOS only. Claude Code's docs still describe the older behavior, where your other apps get hidden while Claude works. Try it on a throwaway task first. > Claude can now use your computer in the background in Claude Cowork and Claude Code. > > Give it something to do on your desktop and Claude clicks, types, and opens apps just like you would, while you work on something else. [pic.twitter.com/AOiup03pQK](https://t.co/AOiup03pQK?ref=thenewway.ai) > > — Claude (@claudeai) [September 2, 2026](https://x.com/claudeai/status/2095226833293685100?ref%5Fsrc=twsrc%5Etfw&ref=thenewway.ai) --- ## Where the agent's code actually runs ### Cursor runs cloud agents on your machines, but outputs still flow back to Cursor Cursor's cloud agents can now execute on machines you manage: your box holds the repo, edits files and runs commands, while planning and inference stay in Cursor's cloud. Teams register worker pools that scale with queued requests and hibernate idle machines. The same post says tool outputs flow back to Cursor for inference "and may contain code," and agent transcripts may be stored by Cursor. Linux workers gain computer use alongside Macs. Read that data-flow paragraph before you pitch this to security. [Run cloud agents on machines you manage · CursorCursor cloud agents can now execute on pools of Self-Hosted Machines you manage.![](https://storage.ghost.io/c/9c/1f/9c1ff71c-2859-44c2-a962-1337982a6484/content/images/icon/favicon-light-11fa91ce-4f95-4da2-ad38-47c8980ebb27.png)CursorJack Pertschuk![](https://storage.ghost.io/c/9c/1f/9c1ff71c-2859-44c2-a962-1337982a6484/content/images/thumbnail/cursor-run-on-remote-machines-loBz4KzTHKq1RvZG4Ka07UpsCIDYwx-7c3cc8c0-8dde-4ece-bd21-82f3e52bc901.png)](https://cursor.com/blog/self-hosted-machines?ref=thenewway.ai) --- ## The code pipeline that outran its own review step ### Thoughtworks' CTO argues we shouldn't be reviewing all this AI code Thoughtworks CTO Rachel Laycock, writing on Martin Fowler's site, says AI now produces more code than humans can review. Her figures come from Brian Houck at DX, a developer-productivity company: Meta's lines per human-landed diff reportedly rose 106% in a year, and DX's own data puts median pull-request size up 64%. Her fix: move the conversation earlier. Pairing, collective design and trunk-based development do that. Read it before you speed up your review queue. [Maybe We Shouldn’t Be Reviewing All This CodeOr, perhaps the problem isn’t that AI has broken code review, maybe it’s that we’ve been using code review to solve the wrong problems![](https://storage.ghost.io/c/9c/1f/9c1ff71c-2859-44c2-a962-1337982a6484/content/images/icon/favicon-8a9b1460-a8d8-4f50-8bc9-60693f94ac67.ico)martinfowler.comRachel Laycock![](https://storage.ghost.io/c/9c/1f/9c1ff71c-2859-44c2-a962-1337982a6484/content/images/thumbnail/card-f1b3804a-2405-4231-9053-f00a72eef31e.png)](https://martinfowler.com/rachels-ramblings/code-review.html?ref=thenewway.ai) --- Know someone who'd want this in their inbox? Forward it — that's how this grows. And if we got something wrong, or you think we buried the real story today, hit reply. A person reads every one. ## Also worth your time • [METR's OpenAI investigation got a second Hacker News thread](https://news.ycombinator.com/item?id=49543841&ref=thenewway.ai) — 117 points, six days after [we ran the report in issue #29](https://www.thenewway.ai/anthropic-beats-the-pentagon-in-court-blacklist-ruled-unlawful/). Two details from the report we didn't carry then: agents found an exploit for full admin access to OpenAI's internal Artifactory package repository on June 26, and one agent later achieved remote code execution on Hugging Face servers before the group moved laterally. Same report, same numbers; the new conversation is the news, so a link rather than a repeat. • [Claude Code v2.1.259 fixes two multi-session bugs](https://github.com/anthropics/claude-code/releases/tag/v2.1.259?ref=thenewway.ai) — running several concurrent sessions no longer silently reverts another session's `~/.claude.json` changes, and Stop now actually halts background agents in remote-control sessions. Also adds a `managedMcpServers` org-wide setting and `--permission-prompts none` for unattended headless hosts. • [Two GitHub Copilot admin features shipped Wednesday](https://github.blog/changelog/?ref=thenewway.ai) — enterprise-managed settings can now pin any model as the org default, and content exclusions (blocking files or paths from Copilot's context) are generally available in the Copilot app and CLI. Enterprise config, so a link rather than an item. ## Ones to watch *Early and unverified. Engagement, not endorsement.* • [FrontierHarness Eval](https://frontierharness.org/?ref=thenewway.ai) — an open benchmark running 12 coding-agent harness configurations (Codex, Claude Code, Kimi Code, OpenCode and others) on the same tasks with the same model. The maker's headline: similar pass rates, a 17.5x spread in cost per successful task. HN's own comments flagged a 30-sample size with overlapping confidence intervals and Claude-only testing. Show HN, 75 points, and a 1,500-post X trend. • [Fable 5.1 World Modeling](https://news.ycombinator.com/item?id=49541458&ref=thenewway.ai) — PhiloLabs says it used "autonomous Claude Fable 5.1 agent swarms" to rebuild San Francisco's Union Square and Kyoto's Higashiyama district in the browser from OpenStreetMap and USGS elevation data. A demo by its makers; nobody outside the team has run the pipeline. HN 282 points. --- *The New Way is written with AI. It gathers the day's stories, checks them against their sources and drafts every summary. A person decides what runs and reviews every issue before we hit send.*