Week in AI: June 1-7, 2026 (Week 23)

News of the week: June 1-7, 2026 (Week 23)

Money and infrastructure dominated the week: Anthropic confidentially filed to go public, Alphabet raised a record $45 billion through a share sale, and Google is renting compute capacity from SpaceX for $920 million a month. At Computex, Nvidia entered the Windows PC market with its RTX Spark chip and announced volume production of the Vera Rubin platform. At its Build conference, Microsoft reshaped Windows into a system for AI agents. President Trump signed a softer executive order on AI oversight, and Florida became the first US state to sue OpenAI over violent incidents.

Business and Investment

Anthropic Heads for an IPO

Anthropic, the maker of the Claude models, announced on June 1 that it had confidentially submitted a draft S-1 registration statement to the US Securities and Exchange Commission (SEC). It is the first formal step toward an initial public offering (IPO); the company has not yet set the number of shares or the price (Anthropic, TechCrunch).

The move came just days after a $65 billion Series H funding round that valued the company at $965 billion, higher than OpenAI. According to the company, Anthropic’s annualized revenue has surpassed $47 billion, up from around $9 billion at the end of 2025.

Co-founder Daniela Amodei dismissed doubts about the returns on AI investment at the Bloomberg Tech conference on June 4. She said the IPO is tied to the enormous costs of training and running models (TechCrunch).

Alphabet Sells Shares for a Record Sum

Alphabet, the parent company of Google, announced on June 1 a plan to sell $80 billion in shares to fund AI infrastructure, including a $10 billion sale to Berkshire Hathaway (TechCrunch).

The first tranche was oversubscribed and brought in $45 billion instead of $40 billion; with another planned tranche, the total is expected to reach $85 billion. According to Bloomberg, it is the largest share offering in history, beating the previous record held by Petrobras, which raised $70 billion in 2010 (TechCrunch).

Google Will Rent Compute from SpaceX

A regulatory filing by SpaceX revealed that Google will pay it $920 million a month from October 2026 to June 2029 for roughly 110,000 Nvidia GPUs and other hardware. Google describes it as “bridge capacity” to meet demand for Gemini Enterprise (TechCrunch, SEC).

More Investments and Data Centers

June 3: Music generator Suno raised $400 million at a $5.4 billion valuation, even though it is still being sued by Universal Music, Sony and Germany’s GEMA (TechCrunch).
June 4: Fintech Ramp, which builds on AI-driven automation of corporate spending, raised $750 million at a $44 billion valuation (TechCrunch).
June 5: Data center operator AirTrunk will invest $30 billion in India by 2030 in 5 gigawatts of new capacity (AirTrunk).
June 4: Near New Albany, Ohio, Meta is building data centers in large tents powered by 200 MW gas turbines (TechCrunch).
June 4: Airbnb CEO Brian Chesky is preparing a new AI lab, according to Bloomberg, though he will not run it himself (TechCrunch).
June 2: Uber introduced a monthly cap of $1,500 per employee per AI coding tool (such as Claude Code or Cursor) after burning through its annual AI budget in four months (TechCrunch).

Chips and Hardware

Nvidia Enters the Windows PC Market

At Computex in Taipei on June 1, Nvidia unveiled the RTX Spark chip for thin laptops and small desktop PCs running Windows. It combines a 20-core Arm-based Grace CPU with Blackwell graphics and delivers up to 1 petaflop of AI performance, or a quadrillion operations per second (Nvidia).

According to the company, the chip is designed to run AI agents, programs that carry out tasks on their own, securely. PCs with it are expected this fall from ASUS, Dell, HP, Lenovo, MSI and Microsoft Surface (TechCrunch).

The same day, Nvidia announced that its Vera Rubin data center platform is in production. The Spectrum-X Ethernet Photonics networking technology has also entered full production, with CoreWeave, Lambda and Oracle Cloud among the first to deploy it (Nvidia).

Products and Agents

Microsoft Build: Windows as a System for Agents

At its Build developer conference on June 2, Microsoft unveiled a set of features meant to turn Windows into a secure platform for AI agents. A key role is played by Microsoft Execution Containers (MXC), an isolated environment that limits what an agent can access while it runs (Windows Developer Blog).

Windows 365 for Agents, a service that assigns managed cloud PCs to agents, is generally available as part of Agent 365.
The local models Aion 1.0 Instruct and Aion 1.0 Plan will run directly on the PC without the cloud; the planning model has 14 billion parameters and will ship with Windows on high-performance devices.
Scout is an always-on personal agent built on the open source OpenClaw framework and connected to Microsoft 365; it checks every action against policies and records it in an audit trail (TechCrunch).
Mustafa Suleyman introduced seven in-house MAI models; the coding model MAI-Code-1-Flash has been rolling out in GitHub Copilot since June 2 (Microsoft for Developers).

OpenAI Expands Codex Beyond Programmers

On June 2, OpenAI released six plugins for Codex focused on data analysis, creative work, sales, product design and investment banking. According to the company, Codex has more than 5 million weekly active users, and about a fifth of them are not programmers (OpenAI, TechCrunch).

Another new feature is Sites, for building hosted interactive websites in partnership with companies such as Wix, Figma and Replit.

More Product News

June 3: After two years of testing, the customer service AI agent Meta Business Agent is available in WhatsApp worldwide; it can answer questions, recommend products and book appointments (TechCrunch).
June 2: Anthropic now lets Enterprise plan customers assign admin permissions through custom roles, for example billing only (Claude Help Center).
June 5: Anthropic announced the retirement of Claude Opus 4.1 on August 5, 2026 and advised developers to move to Opus 4.8 (Claude Platform Docs).

New Models and Open Source

Google Releases Gemma 4 12B

On June 3, Google released the open model Gemma 4 12B under the Apache 2.0 license. It processes images and audio directly in its language core without separate encoders and needs only 16 GB of graphics or shared memory, so it can run on an ordinary laptop (Google).

According to Google, its performance approaches that of the larger Gemma 4 26B while using less than half the memory. The company says the Gemma 4 family has passed 150 million downloads.

June 1: Google shut down the older Gemini 2.0 Flash and Flash-Lite models in the Gemini API; they are replaced by Gemini 3.5 Flash and 3.1 Flash-Lite (Google AI Studio).

Regulation and Law

Trump Signs a Softer Executive Order on AI Oversight

On June 2, President Donald Trump signed an executive order asking selected AI companies to voluntarily submit new models to the government for testing 30 days before release. An earlier draft had called for up to 90 days, while the industry pushed for about two weeks (White House, TechCrunch).

The order explicitly does not introduce mandatory licensing or model approval. It does, however, direct the Department of Justice to prioritize prosecuting AI-assisted hacking.

Florida Sues OpenAI

On June 1, Florida Attorney General James Uthmeier sued OpenAI and its CEO Sam Altman. According to the 83-page complaint, ChatGPT is linked to several violent incidents, including the 2025 shooting at Florida State University; OpenAI has previously denied responsibility (TechCrunch).

More Policy News

June 3: Following a decision by the UK Competition and Markets Authority (CMA), Google will let publishers opt their sites out of AI search features such as AI Overviews; it will test the option in the UK first (Google, CMA).
June 5: Trump said he is negotiating deals with AI companies that would benefit Americans; according to CNBC, the government is considering a stake in OpenAI (CNBC).
June 6: White House AI adviser Sriram Krishnan announced that he will leave the administration at the end of June (TechCrunch).

Security and Incidents

Claude Mythos Turns to Protecting Infrastructure

On June 2, Anthropic expanded Project Glasswing to roughly 150 organizations from more than 15 countries. They will get access to Claude Mythos, a model that finds security flaws in software; since April, the first group of about 50 partners has used it to find more than 10,000 serious vulnerabilities (Anthropic).

The new partners come from the energy, water, healthcare and telecommunications sectors. According to the Financial Times, as reported by TechCrunch, Anthropic sent about six engineers to the US NSA on June 5 to help the agency with the model; the NSA did not confirm this (TechCrunch).

Attackers Outsmart Meta’s Chatbot

Hackers took over Instagram accounts by asking Meta‘s AI support assistant to add a new email address to someone else’s account. The victims included the former Obama-era White House Instagram account; spokesperson Andy Stone said on June 1 that the flaw has been fixed (TechCrunch).

June 6: OpenAI introduced Lockdown Mode in ChatGPT, which disables web browsing, deep research and agent mode; it is meant to reduce the risk of data leaks through prompt injection, malicious instructions hidden in content (OpenAI, TechCrunch).
June 7: Due to degraded performance, Notion temporarily disabled all Anthropic models in Notion AI; it restored access after about twelve hours, and Anthropic confirmed a brief infrastructure outage (TechCrunch).