The AI Daily

Anthropic dominated the model layer today, shipping Claude Sonnet 5 as a faster, safer default alongside the Claude Science research workbench, while Google quietly expanded its Gemini media-generation lineup with two new efficient models.

The AI chip rally broadened dramatically in Q2 — Micron, Intel, and AMD collectively added $2 trillion in market value — while inference chip upstart Etched formally launched with $800M raised and $1B in contracted sales, signaling that purpose-built inference silicon is becoming a real market.

Chip M&A and AI infrastructure financing headlined the deals layer: Onsemi's $7B Synaptics acquisition targets physical AI silicon, Schneider Electric paid $3.1B for industrial AI firm Cognite, and inference chip startup Etched formally surfaced with $800M raised and a $5B valuation.

Google's Gemini Enterprise Agent Platform gained two new media-generation models today, while Couchbase entered the agent-memory race with a dedicated AI Data Plane — reflecting growing investment in the context and orchestration layer that sits between models and applications.

Enterprise AI moved from pilots to production accountability today: Morgan Stanley cut P&L reconciliation time in half using deliberately constrained agents, Anthropic launched a drug discovery program, and Google's NotebookLM added short-form video summaries — all pointing to AI deepening in high-stakes workflows.

The Trump administration's decision to lift export controls on Anthropic's Fable and Mythos models was the sharpest policy signal of the day, while a new US KIDS Act and rising China-linked AI espionage add to a thickening regulatory and threat environment around AI.

India's government made a major infrastructure bet today, with the Finance Ministry approving a ₹1.25 lakh crore outlay for Semiconductor Mission 2.0, while Aarogya Setu 2.0's deployment of Google's Gemma model to digitise health records marks one of the country's most concrete government AI deployments to date.

Morgan Stanley cut its riskiest reconciliation job in half — by making its agents less autonomous — While most AI coverage chases new model benchmarks, this piece documents a live production deployment in one of banking's most unforgiving workflows — P&L reconciliation — and the counterintuitive finding that deliberately reducing agent autonomy was the key to unlocking a 50% efficiency gain. For any strategist designing enterprise AI rollouts, this is the template for how to build trust and measurable ROI before expanding scope. Read →