Today in AI — 12 May 2026
Today's top AI news — curated links and commentary on the stories that matter for product builders.
Today's AI news splits into two registers: billions flowing into infrastructure and defence, and a quieter race to make AI agents feel less like tools and more like colleagues. The money dominates headlines, but the product shifts matter more for anyone building.
Capital keeps accelerating
Billion-dollar commitments stacked up today across chips, power, and defence. Cerebras has upsized its IPO to $4.8 billion as demand hit 20x oversubscription. ByteDance plans over $30 billion for AI expansion in 2026. Blackstone and Halliburton are investing $1 billion in VoltaGrid for AI data centre power, while Helsing is raising $1.2 billion at an $18 billion valuation for autonomous defence drones.
- Cerebras upsizes IPO to $4.8 billion as demand hits 20x oversubscription — CNBC
- ByteDance boosts 2026 AI infrastructure spending to $30 billion — Winbuzzer
- Blackstone and Halliburton invest $1 billion in VoltaGrid for AI data center power — GlobeNewsWire
- Helsing raises $1.2 billion at $18 billion valuation for autonomous defense drones — TechCrunch
AI goes operational
OpenAI's $4 billion Deployment Company points to a more hands-on enterprise model, where labs place AI engineers inside customer organisations. Microsoft is adding AI agent monitoring to the Windows taskbar, and Hermes Agent has overtaken OpenClaw on OpenRouter rankings. The direction is clear: AI is moving from "call an API" to "embed in the organisation."
- OpenAI launches $4 billion Deployment Company to embed AI engineers inside enterprises — OpenAI
- Windows 11 May update adds AI agent monitoring to the taskbar — Windows Central
- Hermes Agent overtakes OpenClaw as the most-used AI agent on OpenRouter — MarkTechPost
Cyber offence meets cyber defence
Google detected the first confirmed AI-generated zero-day exploit in the wild, targeting a two-factor authentication flaw. The labs are responding differently: OpenAI gave the EU access to GPT-5.5-Cyber for defensive use, while Anthropic keeps its Mythos model restricted. Two defensible strategies; we will find out which ages better.
- Google thwarts first AI-generated zero-day exploit aimed at mass exploitation — CNBC
- OpenAI shares cyber model with EU while Anthropic keeps Mythos restricted — CNBC
Interaction and discovery
Thinking Machines says it is building an AI that listens while it talks, a direct push on conversational quality. Digg is back as an AI news aggregator powered by X engagement data. One story is about interaction design; the other is about discovery distribution.
- Mira Murati's Thinking Machines unveils AI that listens while it talks — TechCrunch
- Digg relaunches as an AI news aggregator powered by X engagement data — TechCrunch
The full-stack premium
Alphabet is closing in on Nvidia as AI stack dominance drives a 160% rally. For builders, the market is pricing integrated capability over any single breakthrough.
Capital is solving the infrastructure bottleneck, agents are becoming something users monitor rather than invoke, and the cybersecurity question just got real. Build accordingly.