Microsoft held its biggest Build in years, Anthropic confidentially filed for an IPO, and the White House signed its first AI executive order. Capital is still flowing, the model release pace is relentless, and the questions are getting harder. Here’s the week.
—
## Microsoft Build 2026: MAI Models, Agentic Windows, and Majorana 2
Microsoft’s annual developer conference in San Francisco was its most AI-dense Build to date. The flagship announcement was MAI-Thinking-1 — Microsoft’s first in-house reasoning model, trained from scratch on enterprise-licensed data, with 35 billion active parameters, a 256K context window, and benchmarks that match Anthropic’s Opus 4.6 on coding tasks. The company also launched MAI-Code-1, MAI-Image-2.5, MAI-Transcribe-1.5 (supporting 43 languages), and MAI-Voice-2, while formally repositioning Windows as the OS for autonomous AI agents with the Windows Agent Framework now in production. On the hardware side, Microsoft unveiled Majorana 2, a quantum chip claiming 1,000x higher qubit reliability than its predecessor and a stated roadmap to one million qubits on a palm-sized chip.
—
## Anthropic Files Confidential S-1 at a $965B Valuation
Anthropic submitted a Form S-1 to the SEC on June 1, setting up a potential public offering later in 2026. The filing values the company at $965 billion post-money — above OpenAI’s most recent $852 billion private valuation — following its $65 billion Series H round. No ticker or price range has been set, and the move puts Anthropic in a three-way IPO race alongside OpenAI (targeting Q4 2026) and the SpaceX/xAI roadshow scheduled for June 8.
—
## Project Glasswing Expands to 150+ Organizations Across 15 Countries
Anthropic’s AI-powered vulnerability discovery program, Project Glasswing, expanded significantly this week — growing to roughly 150 new partner organizations across 15+ countries, including India and EU member states. The program, which uses the unreleased Claude Mythos Preview model, has now helped partners identify more than 10,000 high- or critical-severity vulnerabilities. New partners include NYSE and Intercontinental Exchange. Anthropic also issued a notable warning: within 6 to 12 months, other labs may deploy Mythos-class models without equivalent safety measures in place.
—
## Trump Signs AI Executive Order: Voluntary 30-Day Pre-Release Review
President Trump signed “Promoting Advanced Artificial Intelligence Innovation and Security” on June 2, directing AI companies to voluntarily submit frontier models for government testing up to 30 days before public release. The order also establishes an AI cybersecurity clearinghouse. An earlier draft included mandatory 90-day review periods, which were dropped following industry pushback — a point critics were quick to note, along with the absence of binding standards, mandatory testing requirements, or independent oversight mechanisms.
—
## GitHub Copilot Drops Flat-Rate Billing — Developers Push Back
GitHub switched Copilot from flat-rate subscription pricing to token-based (usage-based) billing on June 1, and the backlash was swift. One widely cited example showed a $29/month plan potentially reaching ~$750/month; another projected a jump from $50 to $3,000/month under heavy use. Developers across social platforms announced switching to OpenAI Codex, Claude, or local open-source alternatives. As of this writing, Microsoft has not announced a rollback.
—
## Stargate Michigan: OpenAI and Oracle Break Ground on a 1GW Data Center Campus
OpenAI and Oracle broke ground on the Stargate Michigan campus in Saline Township — a 1-gigawatt AI data center development carrying a $45B+ construction cost plus an estimated $30 to $40 billion in equipment. Sam Altman attended the groundbreaking. The first of three 550,000-square-foot buildings is reportedly near completion, and the project is expected to generate more than 2,500 union construction jobs and 450+ permanent roles.
—
## A Busy Week for Model Releases
Anthropic released Claude Opus 4.8 on May 28, joined by Alibaba’s Qwen3.7 Max. GPT-5.5 Instant received an update delivering cleaner natural responses and new writing and coding output blocks. The current pace: roughly one new frontier model every three days. OpenAI’s GPT-5.6 is targeting a June release with reported advances in reasoning and agentic workflows.
—
## AI Labs Are Formally Studying Machine Consciousness
The Financial Times reported that Google DeepMind, Anthropic, and Meta have each hired psychologists, philosophers, and ethics researchers to formally investigate machine consciousness and AI welfare. Anthropic disclosed that it is actively testing models for behaviors resembling “panic” and “anxiety” as part of what it calls “model welfare research,” stating: “We remain deeply uncertain about this, but we think the question is serious enough to study carefully.” DeepMind brought on Cambridge philosopher Henry Shevlin. No current AI system has been confirmed as sentient.
—
## Alphabet Plans $80B Stock Sale for AI Infrastructure
Alphabet announced plans to raise $80 billion through stock sales — including a $10 billion investment from Berkshire Hathaway — to fund AI compute infrastructure. The company simultaneously revised its 2026 capital expenditure forecast upward to $180 to $190 billion. The raise signals Alphabet’s intent to match Microsoft and Amazon in the hyperscaler infrastructure buildout.
—
## DoD Awards Microsoft a $9.69B Contract for Azure and AI Copilot
The U.S. Department of Defense awarded a $9.69 billion enterprise contract to Microsoft covering Microsoft 365, Azure, and AI Copilot services across DoD environments. The contract is projected to save the government $422 million. The deal marks one of the largest federal AI adoptions on record and a clear signal of where enterprise AI procurement is heading.
—
The through-line this week: consolidation at the top, pain at the developer level, and a government still trying to catch up. More next week.
