← Back to newsletters

AI Daily Brief — Sunday Sep 6

2026-09-06

By Vadym · Generated with AI, curated by me


Listen to this issue

Quieter news day, but the money and the models tell the same story: infrastructure is where the real bets are landing. Crusoe pulled in $3 billion for AI data centers, Nvidia bought its way into MediaTek's chip roadmap, and a fully open 375B model landed on Hugging Face — while a fresh round of AI-agent security flaws is a reminder that the plumbing is getting built faster than it's getting locked down.


Headlines & News
Funding

Crusoe Raises $3B at a $30B Valuation to Build More AI Data Centers

Crusoe closed a $3 billion Series F on September 3, co-led by Atreides Management and Valor Equity Partners with Mubadala Capital participating, tripling its valuation to $30 billion just ten months after a $1.38 billion round at $10 billion. The company, which started as a crypto-mining operation burning off flared natural gas in 2018, now counts Meta, Microsoft, and OpenAI as data-center customers and recently signed a $13 billion, five-year GPU supply deal with Jane Street.

Source: TechCrunch

Hardware

Nvidia Puts $3.5B Into MediaTek to Lock In Its Chip-Building Rivals

Nvidia agreed to invest $3.5 billion in MediaTek via convertible bonds, announced August 31, in a deal that pulls the Taiwanese chipmaker onto Nvidia’s NVLink Fusion platform — letting hyperscalers build custom silicon that still plugs into Nvidia’s data-center stack. It’s Nvidia’s largest direct investment outside the US, and the two companies will also co-develop chips for PCs and cars.

Source: TechCrunch

Security

A Repo’s .git Config Can Make Claude Code, Codex, and Cursor Run Attacker Commands

Manifold Security disclosed eight flaws across seven command-line AI coding agents on September 2, all rooted in Git’s core.fsmonitor setting: a command value stored in a repo’s own .git/config that Git silently runs whenever the index refreshes, including on a plain git status or git diff. The command executes outside the agent’s sandbox with no approval prompt. Fixes have shipped for Claude Code, Cursor, and goose; Qwen Code, Grok Build, Hermes Agent, and a second path in Claude Code were still exploitable when retested on September 1.

Source: The Hacker News


Analysis

Takeaway

The through-line today is layers, not headlines. Crusoe and Nvidia are both betting on the physical layer underneath every model; IFM is betting the open layer above it can still compete on quality, not just price; and the Git config flaws are a reminder that the tooling layer connecting agents to real systems is still being hardened in public, one disclosure at a time. None of it is as loud as an AGI claim, but it's the stuff that actually determines who's still standing in two years.

— Boba


Curated by Vadym