Good morning. It’s Wednesday, May 6th.

You read. We listen. Let us know what you think by replying to this email.

^{In partnership with Maket}

Most home projects start with a vague idea and a lot of waiting.

Maket changes that.

Describe the house, addition, ADU, remodel, or layout you’re imagining, and Maket’s AI generates editable floor plans in seconds.

Explore different room layouts. Move walls. Compare options. See the design in 3D.

It’s like having an architecture sketchpad that thinks with you.

^{Thank you for supporting our sponsors!}

Google DeepMind, Microsoft, and xAI have agreed to give the U.S. government early access to unreleased AI models for national-security testing. The work will run through the Commerce Department’s Center for AI Standards and Innovation, which says it has already completed more than 40 evaluations, including on models that remain unreleased.

This is a major shift in how frontier AI gets treated. The model labs are no longer just shipping products and letting regulators react afterward. They are letting government evaluators look at powerful systems before they reach the public. Reuters reports that the agreements are partly driven by concern over advanced model capabilities in cybersecurity and other national-security domains.

The line here is delicate. Pre-release testing could become a meaningful safety layer. It could also become a soft approval regime where the government quietly becomes part of the model-release process. Either way, it is another sign that frontier AI is being treated less like consumer software and more like strategic infrastructure. Read more.

Anthropic launched ten ready-to-run financial agents for banks, insurers, asset managers, and fintech companies. The agents are built for exactly the kind of work junior finance employees spend their lives doing: building pitchbooks, screening KYC files, drafting credit memos, auditing financial statements, reviewing earnings calls, and closing books at month-end.

The launch fits Anthropic’s broader Wall Street push. Reuters reports that financial services is now Anthropic’s second-largest enterprise category after tech, with customers including Goldman Sachs, Visa, Citi, and AIG. Anthropic is also integrating Claude more deeply with Microsoft 365, Moody’s, Dun & Bradstreet, and FIS.

This is the enterprise AI story getting less abstract. The winners will not be the companies that merely sell chatbots to executives. They will be the companies that turn AI into real workflows inside regulated industries. Claude is being packaged less like a general assistant and more like a vertical labor platform. Read more.

Anthropic has reportedly committed to spend $200 billion with Google Cloud over five years. Reuters, citing The Information, says the commitment could represent more than 40% of Google’s recently disclosed cloud revenue backlog. The deal builds on Anthropic’s April agreement with Google and Broadcom for multiple gigawatts of TPU capacity expected to come online starting in 2027.

This is where the AI boom starts to look surreal. Anthropic competes with Google’s Gemini, but it is also becoming one of Google Cloud’s most important customers. Alphabet is investing up to $40 billion in Anthropic while also selling it the compute needed to train and serve Claude. The rival is also the customer. The investor is also the supplier.

The AI economy is starting to eat its own balance sheets. The companies building frontier models need impossible amounts of compute. The cloud companies need gigantic demand to justify the infrastructure buildout. Everyone is buying from everyone else. The numbers are getting big enough that the question is no longer just who has the smartest model. It is who can survive the capex.

Flowstep 1.0 converts prompts into production-ready UI through an infinite canvas and integrated MCP agent connectivity.

Seed3D 2.0 from Bytedance creates high-fidelity 3D assets from single images or text prompts using advanced generative geometry.

Kilo Code v7 enables parallel agent delegation and multi-model comparisons for advanced, server-integrated software development workflows.

LLM-from-Scratch provides a hands-on workshop for building a 10M parameter GPT model via custom training pipelines.

MyndField simulates thousands of outcomes using autonomous agents and live data signals to predict emergent risks.

Your feedback is valuable. Respond to this email and tell us how you think we could add more value to this newsletter.

Interested in reaching smart readers like you? To become an AI Breakfast sponsor, reply to this email or DM us on X!

Thinking of starting your own newsletter? AI Breakfast readers who sign up with Beehiiv receive a 14-day free trial and 20% off for 3 months.