Big week in AI. Anthropic filed to go public at nearly a trillion dollars, Congress dropped a 269-page AI bill that's already making everybody mad, and Microsoft quietly built seven AI models without touching OpenAI. Let's get into it.
Anthropic Is About to Be a Trillion-Dollar Company
Anthropic filed a confidential S-1 with the SEC on June 1st. Revenue run-rate hit $47 billion in May, up from $10 billion a year ago. Their Series H valued the company at $965 billion. Analysts are calling a trillion-dollar debut the base case if markets cooperate.
Let that sink in. A company that didn't exist five years ago is pricing out at nearly a trillion dollars because it makes a coding agent that outperforms most software teams.
Here's the part that should make you stop and think. Anthropic is paying SpaceX $1.25 billion per month for compute. Fifteen billion dollars a year to a single vendor. Their S-1 margins conversation is going to be rough. Investors will love the top-line growth and quietly panic about vendor concentration risk.
The enterprise story is real though. Salesforce ran a migration scoped at 231 days and Claude finished it in 13. That's the ROI driving enterprise adoption. Companies aren't buying into AI because it's cool. They're buying in because they can stop paying the consulting firm.
SpaceX IPOs on Nasdaq under the ticker SPCX on June 12th, targeting a $75B raise at a $1.75 trillion valuation. OpenAI is expected to file shortly after. This summer is going to feel like the dot-com boom but with actual revenue behind it.
Congress Dropped a 269-Page AI Bill and Nobody Is Happy
Representatives Obernolte and Trahan released the Great American AI Act on June 4th. It's a 269-page discussion draft that would freeze all state AI laws for three years, require big AI companies to publish safety frameworks, and impose criminal penalties for using AI to impersonate government officials.
The piece generating heat: it would preempt Colorado's AI Act, which takes effect June 30th. Colorado's law has actual anti-discrimination requirements in employment, healthcare, and financial services. The federal bill's version of those protections is vague enough that lawyers will fight over it until 2035.
Labor unions came out swinging. The AFL-CIO, AFT, and Association of Flight Attendants issued a joint rejection: "Hard no. This bill is a giveaway to the AI industry." Tech trade groups called it a great start. That split tells you everything you need to know about who wrote it.
My read: this is a bill designed to protect companies, not people. A three-year moratorium on state laws without matching federal protections is just telling Colorado and California to sit down while Washington figures out what the industry wants. That's not regulation. That's a stall.
The bill isn't even formally introduced yet. It's a discussion draft open for public comment. Don't panic, but if you're building anything that touches Colorado consumers, you've got 25 days until their law takes effect regardless of what Congress does.
Microsoft Built Seven AI Models and Everyone Was Too Busy Watching Anthropic
At Build 2026, Satya Nadella unveiled seven new AI models built in-house with no OpenAI involvement. They're calling the family MAI: a reasoning flagship, a coding model, image generation, transcription, speech, and two more for Azure enterprise workloads.
Microsoft says MAI-Transcribe-1.5 beats both Gemini and OpenAI on transcription across 43 languages at 5x the speed. MAI-Image-2.5 ranks above Gemini in the Arena leaderboard. These aren't placeholder models. They're going into Teams, GitHub Copilot, and Dynamics 365 right now.
The timing is deliberate. OpenAI's IPO is weeks away. Microsoft needs investors to believe its AI strategy doesn't live or die with OpenAI's equity valuation or model quality. The MAI family is their proof point: we can build this stuff ourselves.
What this means if you're building: the days of picking one AI vendor and going all in are numbered. Copilot is becoming model-agnostic. Azure is going multi-model. The platform strategy is to be the layer above the models, not the model itself. If you're locking your architecture to a single provider right now, that's a risk worth revisiting.