Skip to content
Source · Daily Brief

AI Daily Brief — 9 April 2025

The biggest single AI day of Q2 so far. Google Cloud Next opened with Ironwood — its first TPU built for inference — plus the Agent2Agent protocol, an Agent Development Kit, Firebase Studio, Gemini 2.5 Flash, Lyria music in Vertex AI, and a massive Workspace push. Anthropic launched Claude Max. xAI shipped Grok 3 API. And Trump paused his reciprocal tariffs for 90 days for everyone except China — triggering one of the largest single-day tech rallies in history.

Top stories

  • Google unveils Ironwood — 7th-gen TPU, first designed for inference. 4,614 FP8 TFLOPs per chip, 192 GB HBM3E (~7.37 TB/s bandwidth), 9.6 Tb/s ICI. Pods scale to 9,216 chips for 42.5 ExaFLOPs. Up to 10x peak vs TPU v5p; >4x per-chip efficiency vs Trillium. via Google
  • Google launches Agent2Agent (A2A) — open protocol for agent interoperability. HTTP + Server-Sent Events + JSON-RPC 2.0 plus ‘Agent Cards’ for capability advertisement. Launched with 50+ partners including Atlassian, Box, Cohere, Intuit, LangChain, MongoDB, PayPal, Salesforce, SAP, ServiceNow, Workday. Positioned as the complement to Anthropic’s MCP (agent-to-tool). via Google Developers
  • Anthropic launches Claude Max — $100 and $200/mo tiers. $100/mo for 5x Pro usage, $200/mo for 20x Pro usage. Priority access to new models/features (advanced reasoning, Claude Code, voice mode). Direct response to OpenAI’s $200 ChatGPT Pro, with a softer middle tier. via TechCrunch
  • Trump pauses reciprocal tariffs for 90 days — except China, hiked to 125%. Reciprocal-tariff rate lowered to 10% for ~75 countries seeking deals; China tariff raised from 104% to 125% after Beijing retaliated. Dow +7.87% (~2,963 pts), S&P 500 +9.5%, Nasdaq +12.2% — largest single-day tech rallies since 2008/2001. via NPR
  • xAI ships Grok 3 API. Two flavors: Grok 3 ($3/M input, $15/M output) and Grok 3 Mini with reasoning capabilities. First commercial API access to xAI’s flagship. via TechCrunch
  • Gemini 2.5 Flash announced. Low-latency ‘thinking’ model with adjustable reasoning — adjusts depth of reasoning to prompt complexity with developer-controllable cost/latency budget. Coming to Vertex AI, AI Studio and the Gemini app. via Google Cloud
  • Agent Development Kit (ADK) open-sourced. Powers Google’s Agentspace and Customer Engagement Suite. Supports Python, Java, Go, TypeScript; multi-agent composition; built-in tools (Search, Code Exec) and MCP tool integration; bidirectional audio/video streaming. Optimized for Gemini 2.5 Pro and Vertex AI. via Google Developers
  • Firebase Studio launches — cloud-based agentic dev environment. Powered by Gemini, 60+ pre-built templates, App Prototyping agent that designs UI/API/AI flows from natural language, images, sketches or screenshots. Three free workspaces during preview (30 for Google Developer Program members). via Firebase

Who shipped

Google owned the day with the largest single-vendor AI announcement bundle since GTC. Anthropic moved into the premium-subscription tier. xAI opened Grok 3 to developers. Lyria music model arrived in private preview on Vertex AI, joining Veo 2 (now widely available) and Chirp 3 — making Vertex ‘the only platform with generative-media models across video, image, speech and music.’

Open-source pulse

ADK and A2A both shipped open-source. A2A’s 50+ partner roster signals Google trying to make agent interoperability a standard layer rather than a closed moat.

Money, infra & hardware

Google Cloud broadened the Blackwell lineup with A4 (8x B200) and A4X (GB200 NVL72) VMs as the GPU complement to Ironwood. A4X delivers >1 ExaFLOP per GB200 NVL72 system; Google positions itself as the only cloud offering both A4 and A4X. Cloud WAN opens Google’s private global network (>2M miles of fiber, 33 subsea cables, 202 PoPs) to enterprises with 99.99% SLA.

Quiet corners

The AI Futures Project published its much-discussed ‘AI 2027’ scenario forecast (Kokotajlo, Lifland, Larsen, Dean; edited by Scott Alexander) — a 71-page projection of AI agents in workforces by 2026, near-full coding automation by early 2027, and an intelligence explosion in late 2027. via AI Futures

By the numbers

  • 192 GB HBM3E / 4,614 TFLOPs / 9,216 chips / 42.5 ExaFLOPs — Ironwood TPU specs
  • 50+ — A2A protocol launch partners
  • $100 / $200 per month — Claude Max tiers
  • $3 / $15 per million tokens — Grok 3 API pricing
  • +7.9% / +9.5% / +12.2% — Dow / S&P / Nasdaq single-day rally
  • 104% → 125% — China tariff hike
  • >2M miles fiber / 33 subsea cables / 202 PoPs — Cloud WAN substrate
  • Most-mentioned company: Google

Compiled by AI Feed’s editor from verified web sources for 9 April 2025.