Skip to content
Source · Daily Brief

AI Daily Brief — 19 May 2025

The densest enterprise-AI Monday of the year. Nadella opened Build with 50+ announcements framed as ‘the open agentic web.’ Jensen opened NVLink to third-party silicon, NVIDIA stood up a global GPU marketplace, and Foxconn agreed to build a 10,000-Blackwell supercomputer in Taiwan. Pro-Palestine protests interrupted the Build keynote.

Top stories

  • Microsoft Build 2025 opens — ‘age of AI agents and the open agentic web.’ 50+ announcements across GitHub, Azure, Windows, Microsoft 365. Sam Altman appears virtually in a fireside with Nadella. via Microsoft
  • GitHub Copilot Coding Agent launches in public preview. Asynchronous coding agent embedded in github.com and VS Code. Assign an issue or invoke from chat; it spins up a GitHub Actions sandbox, makes commits to a draft PR, and handles low-to-medium-complexity tasks. Available to Copilot Enterprise and Copilot Pro+. GitHub also open-sources Copilot Chat in VS Code and adds Grok 3 to GitHub Models. via GitHub
  • Microsoft 365 Copilot Tuning + multi-agent orchestration. Copilot Tuning — low-code Copilot Studio capability for tuning models on company-specific data, workflows and processes. Multi-agent orchestration also lands so agents can delegate. Copilot Studio already used by 230,000+ organizations. via Microsoft
  • Microsoft Discovery — agentic R&D platform. Specialized AI agents collaborate over a graph-based knowledge engine to automate hypothesis generation, candidate design, simulation and analysis. Microsoft says it used the platform internally to discover a new datacenter coolant prototype in 200 hours (vs years), synthesized and validated within four months. via Azure
  • NVIDIA NVLink Fusion at Computex. NVIDIA opens NVLink IP and silicon so partners can build semi-custom AI racks mixing NVIDIA GPUs with third-party CPUs and ASICs. Launch partners: MediaTek, Marvell, Alchip, Astera Labs, Synopsys, Cadence; Fujitsu and Qualcomm CPUs integrable to NVIDIA GPUs. Jensen also confirmed NVIDIA’s new Taiwan HQ in Beitou-Shilin. via NVIDIA
  • NVIDIA DGX Cloud Lepton — global GPU marketplace. Single-platform access to tens of thousands of Blackwell-class GPUs across NVIDIA Cloud Partners — CoreWeave, Crusoe, Firmus, Foxconn, GMI Cloud, Lambda, Nebius, Nscale, SoftBank and Yotta. Pick by region for on-demand or long-term workloads; supports sovereign AI. via NVIDIA
  • NVIDIA RTX PRO Servers + Enterprise AI Factory design. 8x RTX PRO 6000 Blackwell Server Edition GPUs — 30 PFLOPS FP4, 3 PFLOPS RTX graphics, 800 GB GPU memory, 12 TB/s bandwidth. ~1.7x faster than H100 for many workloads. Cisco, Dell, HPE, Lenovo and Supermicro shipping 2U variants. via NVIDIA
  • Foxconn + NVIDIA + Taiwan government build a Blackwell AI supercomputer. Foxconn becomes Taiwan’s first NVIDIA Cloud Partner. AI factory supercomputer with 10,000 Blackwell GPUs and ~100 MW of power. GB300 NVL72 racks with NVLink, Quantum InfiniBand and Spectrum-X. TSMC signaled intent to use the system for R&D. via NVIDIA
  • xAI Grok 3 and Grok 3 Mini land on Azure AI Foundry. Free preview for the first two weeks. Expands Azure’s catalog past 1,900 models. Elon Musk joined the Build session virtually. via TechCrunch

Who shipped

Microsoft and NVIDIA ran the day end-to-end. xAI got a major distribution win on Azure. Foxconn + Taiwan got the sovereign-compute showcase.

Open-source pulse

Microsoft open-sourced Copilot Chat in VS Code. Windows AI Foundry replaced the Copilot Runtime with native MCP support for Windows File System, Windowing and WSL as MCP servers. Entra Agent ID extends identity management to agents in Azure AI Foundry and Copilot Studio with federation for AWS Bedrock and n8n.

Money, infra & hardware

NVIDIA shifted from selling GPUs to operating a global compute marketplace (DGX Cloud Lepton) — a major business-model expansion. The combined Microsoft Discovery + Copilot Tuning + NLWeb stack reframes ‘enterprise AI’ as primarily an agent-orchestration problem rather than a model-selection problem.

Quiet corners

Pro-Palestine protesters demonstrated outside the Seattle Convention Center and disrupted at least three executive talks; Microsoft briefly cut audio. Software engineer Joe Lopez interrupted Nadella’s opening keynote about Microsoft’s services to Israel’s Ministry of Defense and was later fired.

By the numbers

  • 50+ — Microsoft Build announcements
  • 230,000+ — organizations using Copilot Studio
  • 200 hours — Microsoft Discovery’s coolant discovery time
  • 30 PFLOPS FP4 / 800 GB — RTX PRO Server config
  • 10,000 — Blackwell GPUs in the Foxconn Taiwan supercomputer
  • 1,900+ — models now in Azure AI Foundry
  • Most-mentioned company: Microsoft

Compiled by AI Feed’s editor from verified web sources for 19 May 2025.