Microsoft Build Makes Windows an Agent Platform — and Cuts the OpenAI Cord in Copilot
Satya Nadella opened Build in San Francisco this morning with the same one-word thesis NVIDIA used 24 hours earlier: agents. The keynote framed Windows as a runtime where agents are first-class citizens, not just human users — backed by the Windows Agent Framework (MIT-licensed at the show), Azure Agent Mesh for multi-agent orchestration, Copilot Workspace out of beta, and Agent 365 as the enterprise control plane that governs Microsoft, AWS Bedrock and Google Cloud agents from one pane (Windows Forum, Windows News). Azure AI Foundry picked up native multi-modal pipelines, a visual RAG designer, per-project token budgets aimed at runaway agent spend, and an expanded catalogue that now lists Cohere, Mistral and Stability AI alongside OpenAI and Meta (ChatForest).
The most strategically loaded reveal is Project Polaris, Microsoft's own coding model, which replaces the default OpenAI model in GitHub Copilot from August with automatic migration and a three-month fallback (ChatForest, Notebookcheck). Put the catalogue expansion and the in-house model together and the pattern is unmistakable: the firm that owns the developer runtime is deliberately reducing its dependence on any single frontier lab. For a regulated European buyer, that confirms the question worth planning around is no longer "which model" but whose infrastructure runs it, under which jurisdiction, with what audit trail — the same logic that makes multi-model routing on infrastructure you control, as platforms like Stav offer, a procurement question rather than a convenience.
NVIDIA's Vera Rubin Hits Full Production — and the Market Reprices the Supply Chain
A day after the GTC Taipei keynote, the Vera Rubin story moved from slideware to a production ramp, and the market repriced the whole chain around it. NVIDIA said the Vera Rubin platform is ramping into full production with MGX delivering five rack-scale systems built for agentic workloads, positioning it as the foundation for the next generation of "AI factories" (SiliconANGLE). The reaction was broad rather than narrow: NVIDIA rose about 5%, ARM jumped 17% on its central role in the new systems, and IBM gained 9.2% after being named a partner for system building and secure AI storage; in Taipei the weighted index broke a record as TSMC, Foxconn, MediaTek and Quanta all hit all-time highs (Motley Fool, CRN Asia).
The signal underneath the price moves is concentration. A single vendor's product cycle is now strong enough to move an entire national index and lift a half-dozen suppliers in one session (TradingKey). The agentic compute layer that Build and GTC both spent their keynotes describing rests on one accelerator roadmap — which is exactly why the EU's sovereignty debate this week keeps circling back to supply-chain control, not just where data sits.
Brussels' Sovereignty Package Lands Tomorrow — With a Statutory Definition Attached
After three slips, the EU Tech Sovereignty Package is finally due Wednesday, 3 June, and arrives with a legal spine rather than a press release. Its core, the Cloud and AI Development Act (CADA), would establish four statutory levels of cloud sovereignty — graded on control of the service, control of the supply chain, the processing of AI inference data, infrastructure location, and cybersecurity — that public authorities must weigh in procurement against a use case's sensitivity (Implicator.ai, eu-cloud-ai-act.com). Alongside it sit a revised Chips Act with crisis powers to override supplier contracts and fines up to €300,000 for withholding capacity data, and — for the first time — the EU's own legal definition of "digital sovereignty" (Euronews).
Editor's note: today's other two leads are enterprise and hardware; the Sovereignty Package is carried as the one Regulation main, consistent with the standing one-per-day cap, because its publication tomorrow is the defining European event of the week. The number that should focus European buyers is the market structure CADA is responding to: AWS, Azure and Google Cloud hold roughly 70% of European cloud infrastructure while European providers have fallen to around 15%. If the Act ships as drafted, "processing of AI inference data" stops being a marketing line and becomes a scored procurement criterion — which is the moment running a model on infrastructure you control, the way Stav and the wider European inference cohort do, moves from compliance overhead to a buy-side requirement.
Quick Hits
- The open-weight cost counter holds — The current open-weight cohort — Qwen 3.5, DeepSeek V4 (MIT, 1M-token context), GLM-5 and Mistral Large 3 — continues to sit within a point or two of the Western frontier on agentic coding at a fraction of the hosted cost, the structural backdrop to every "which model" decision a Foundry or Copilot buyer now makes (LLM-Stats, Hugging Face).
- Anthropic's enterprise lead, and its strain — Claude holds roughly 34.4% of US business adoption against OpenAI's ~32.3% as overall AI use crosses 50%, even as both labs lean on Wall Street joint ventures (Anthropic's $1.5B vehicle, OpenAI's $10B-valued Development Company) to convert that lead into deals (VentureBeat, TechCrunch).
- Foundry goes multi-model by default — Build's quiet but telling detail: Azure AI Foundry's default catalogue spans Cohere, Mistral and Stability AI alongside OpenAI and Meta, making "route across vendors" a platform feature rather than a customer workaround (ChatForest).
- WWDC is T-6 — Apple's keynote on 8 June is expected to centre on a Gemini-assisted Siri overhaul, the AI story the company has owed since 2024 (9to5Mac).
- SpaceX prices its IPO next week — The roadshow starts 4 June, pricing is set for 11 June and the Nasdaq debut for 12 June under SPCX, targeting roughly $75B raised at a ~$1.75T valuation — with the Anthropic Colossus compute deal a headline AI-revenue line (CoinDesk).