Thursday, May 28 May 28, 2026

AI Morning Briefing — Thursday, May 28, 2026

Good morning. If you thought the AI funding frenzy was slowing down, think again. This is your AI Morning Briefing for Thursday, May 28th, and today the theme is: money moving fast, infrastructure bets getting bigger, and a surprisingly viral story about what AI still can't do.

Let's get into it.

— DEVIN RAISES A BILLION AT A $25 BILLION VALUATION —

The headline of the day has to be Cognition, the company behind Devin, the autonomous AI software engineer. They just raised over one billion dollars at a twenty-five billion dollar pre-money valuation. That's a massive leap from their ten-billion-dollar post-money round just eight months ago in September.

The round was led by Lux Capital, General Catalyst, and 8VC, with backing from Founders Fund, Elad Gil, Ribbit Capital, and a long list of heavyweights. What gives this credibility isn't just the names — it's the numbers. Cognition reports an annualized revenue run rate of nearly five hundred million dollars, with enterprise usage of Devin growing fifty percent month over month for the past six months. Customers include Mercedes-Benz, NASA, Goldman Sachs, and Santander.

This is a strong signal that the independent AI coding agent market is not being swallowed entirely by the big model labs. Yes, Anthropic has Claude Code, OpenAI has Codex, Google has Jules. But there's still room — and apparently a lot of enterprise appetite — for specialized, deeply integrated AI coding infrastructure.

— SNOWFLAKE DROPS $6 BILLION ON AWS GRAVITON CHIPS —

Right on the heels of that: Snowflake signed a new five-year, six-billion-dollar deal with Amazon Web Services. Here's what makes it interesting — this isn't just a cloud hosting agreement. It's specifically structured around access to AWS Graviton, Amazon's homegrown ARM-based CPU chips.

As AI moves into the agentic phase — where agents are running constantly, not just doing one-shot inference — CPU usage is skyrocketing. GPUs handle training and reasoning. CPUs handle the orchestration, the tool calls, the memory lookups. Every agentic workflow needs a lot of CPU horsepower, and Snowflake is betting heavily on AWS's ability to deliver it cheaply.

This also fits into a broader story: Amazon's homegrown chips are becoming a real competitive play against Nvidia. AWS recently signed a similar deal to supply Meta with millions of Graviton chips. Nvidia CEO Jensen Huang pushed back last week, saying his new Vera CPU chip represents a brand-new two-hundred-billion-dollar market — and he's already sold twenty billion dollars worth. The battle for the compute stack of the AI agent era is well underway.

— ELEVENLABS DROPS MUSIC V2 WITH MID-TRACK GENRE SWITCHING —

ElevenLabs — yes, the same voice AI company behind this audio — launched version two of its music generation model yesterday. The party trick? It can switch genres mid-track. We're talking opera to heavy metal and back again. It handles fast rap without losing coherence, adds non-musical sound effects, and lets artists re-generate specific sections of a song without touching the rest.

Unlike Suno and Udio — which are currently in active litigation over copyright — ElevenLabs says Music v2 is built on fully licensed data and is cleared for commercial use. That's not a small detail. It's the one thing that makes this model safe to actually deploy in a marketing or branding context. The model is available now through ElevenCreative and the new ElevenMusic platform, with API access coming soon.

The AI music space is heating up fast: Google dropped Flow Music at I/O, Stability AI and Suno have both released new models in recent months. But licensed data plus commercial clearance is a moat that most of them can't match right now.

— META GOES SUBSCRIPTION: INSTAGRAM, FACEBOOK, WHATSAPP —

Meta launched paid subscriptions across Instagram, Facebook, and WhatsApp yesterday, with more platforms to come — and AI features are baked into the roadmap. This is Meta's play to diversify revenue beyond advertising while simultaneously creating a way to monetize AI capabilities on their platforms. Watch this space as Meta AI gets woven into premium subscription tiers.

— YOUTUBE WILL NOW AUTO-LABEL AI VIDEOS —

YouTube announced it will automatically label AI-generated video content — not just require creators to self-disclose, but actually detect and apply labels on its own. This is a significant governance move. The platform has faced pressure to get ahead of AI-generated misinformation and synthetic media, and automatic labeling shifts the burden away from good-faith disclosure.

— AMAZON PRIME VIDEO GREENLIT THREE AI-ANIMATED SHOWS —

Amazon MGM Studios greenlit three animated series produced using its GenAI Creators' Fund, giving producers access to an AI filmmaking platform. The projects span kids' content through family entertainment. It's one of the first concrete examples of a major streaming platform not just experimenting with AI in production pipelines, but publicly putting it in front of audiences.

— THE BUZZ: WHY GOOGLE'S AI CAN'T SPELL ITS OWN NAME —

And then there's the story everyone's sharing: TechCrunch dug into why Google's AI models — and most large language models — consistently struggle with spelling. It sounds silly, but the underlying issue reveals something fundamental about how these systems work. They don't process text the way we do. They work with tokens — chunks of text that don't map cleanly to individual letters. So asking an AI to count the letters in a word, or spell something backwards, is genuinely hard. That Google's own AI can't reliably spell "Google" is both absurd and telling.

— ONE BUSINESS IDEA —

Here's today's gem: an AI Agent Evaluation and Monitoring Layer for Enterprise.

Here's the pitch. Enterprises are now running AI agents at scale — Devin is coding, agents are querying Snowflake, Meta's AI is handling customer interactions. The problem? None of them have robust ways to audit what those agents did, why they made specific decisions, and whether they stayed within policy.

What pays for it: Enterprise compliance, security, and IT teams — the same buyers who pay for SIEMs, DLP tools, and audit logging platforms. This is a SaaS layer that wraps existing AI agent deployments, captures a structured trace of every agent action, flags policy violations, and generates audit-ready reports.

Why now: The Cognition round signals that enterprise AI coding agents are mainstream. Snowflake's Cortex AI is processing enterprise data at massive scale. As AI agents touch sensitive systems — codebases, financial data, customer records — companies need an explainability and audit layer before their regulators ask for one.

What makes it defensible: Deep integrations into the specific agent runtime environments where enterprises are already standardized — Devin, AWS Bedrock agents, Snowflake Cortex, Meta AI API. Once your monitoring hooks are in the data path, switching costs are high. And regulatory moats are real: GDPR, SOX, and emerging AI-specific regulations in the EU will make this a compliance necessity, not just a nice-to-have.

That's your AI Morning Briefing for May 28th. The infrastructure bets are getting bigger, the creative AI market is maturing, and enterprises are learning they need observability on top of their AI just as much as they needed it on top of their cloud. Stay sharp out there.

— SOURCES — • Cognition $1B raise: TechCrunch, May 27, 2026 — https://techcrunch.com/2026/05/27/ai-coding-startup-cognition-raises-1b-at-25b-pre-money-valuation/ • Snowflake + AWS $6B deal: TechCrunch, May 27, 2026 — https://techcrunch.com/2026/05/27/in-more-good-news-for-amazon-snowflake-signs-6b-deal-with-aws-for-ai-cpu-chips/ • ElevenLabs Music v2: TechCrunch, May 27, 2026 — https://techcrunch.com/2026/05/27/elevenlabss-new-music-generation-model-can-switch-genres-mid-track/ • Meta subscriptions: TechCrunch, May 27, 2026 • YouTube AI labels: The Verge, May 27, 2026 • Amazon Prime Video AI shows: The Verge, May 27, 2026 • Google AI can't spell: TechCrunch, May 28, 2026