AI Briefing

Nvidia is quietly becoming the biggest venture fund on Earth, plowing $40 billion into AI bets this year alone. Meanwhile, the tools developers trust most just got exposed as credential piñatas, Snap's big AI search play with Perplexity fizzled out before it even launched, and Coinbase is replacing middle managers with AI-powered one-person teams. Oh, and Anthropic shipped an entire Wall Street back office in a box.

01IndustryCNBC ↗

Nvidia Tops $40 Billion in AI Equity Bets in 2026, Led by a $30 Billion Stake in OpenAI

Nvidia has committed over $40 billion in equity investments to AI companies in 2026 alone, with the largest single bet being a $30 billion stake in OpenAI from late February. The remaining $10 billion-plus is spread across deals with IREN ($2.1B), Corning ($3.2B), and roughly two dozen private startup rounds. Critics are drawing comparisons to dot-com-era vendor financing, but Nvidia's $5 billion bet on Intel has already ballooned to over $25 billion, suggesting the chipmaker's investment thesis is working. This signals a fundamental shift: Nvidia isn't just selling shovels in the AI gold rush anymore, it's buying the mines.

02ResearchVentureBeat ↗

Six Research Teams Hack AI Coding Agents: Every Attacker Went for the Credential, Not the Model

Between 2025 and 2026, six independent research teams found critical vulnerabilities in the biggest AI coding tools: OpenAI Codex, Anthropic Claude Code, GitHub Copilot, and Google Vertex AI. The universal pattern was striking: every exploit targeted credentials rather than the AI models themselves. BeyondTrust demonstrated that a crafted GitHub branch name could steal Codex's OAuth token in cleartext (classified Critical P1 by OpenAI), and Adversa found Claude Code silently ignored its own deny rules past 50 subcommands. The findings reveal that AI coding agents are essentially supercharged attack surfaces where privileged secrets sit in plain text within agent conversations.

03IndustryTechCrunch ↗

Snap and Perplexity's $400 Million AI Search Deal Collapses Before It Even Launched

Snap revealed in its Q1 earnings that its $400 million partnership with AI search startup Perplexity has been 'amicably ended' before achieving a broad rollout. The deal, announced last November, would have embedded Perplexity's conversational AI search directly into Snapchat's Chat interface. A Perplexity spokesperson said the planned feature was 'not the right fit' for either company. The collapse is a cautionary signal for the flood of AI integration deals being struck across social media: even when both sides have deep pockets and clear user intent, making AI search feel native inside a social app is harder than writing the check.

04IndustryFortune ↗

Coinbase Cuts 14% of Workforce and Replaces Managers with AI-Powered 'Player-Coaches'

Coinbase CEO Brian Armstrong announced the elimination of roughly 700 jobs (14% of headcount) while radically restructuring how the company operates. The reorganization goes far beyond cost cuts: Armstrong is killing the 'pure manager' role entirely, replacing them with 'player-coaches' who both lead and do hands-on work. Most provocatively, Coinbase is creating 'AI-native pods' where a single person directing AI agents can do the work previously handled by full cross-functional teams of engineers, designers, and product managers. The company expects $50-60 million in restructuring charges but sees AI enabling engineers to ship in days what used to take weeks.

05ProductAnthropic ↗

Anthropic Launches 10 Finance Agent Templates and Debuts Claude Opus 4.7 for Wall Street

Anthropic released ten ready-to-run AI agent templates targeting the most labor-intensive financial workflows: building pitchbooks, screening KYC files, reconciling ledgers, and closing the books at month-end. Each template ships as a plugin for Claude Cowork and Claude Code, or as a cookbook for Claude Managed Agents. The launch pairs with Claude Opus 4.7, which leads the Vals AI Finance Agent benchmark at 64.37%, and a full Microsoft 365 integration that lets Claude carry context across Excel, PowerPoint, Word, and Outlook simultaneously. With partners including Citadel, BNY, FIS, and Carlyle already onboard, Anthropic is making its most aggressive play yet to become Wall Street's AI operating system.