The Week AI Stopped Being a Lab Experiment

A near-trillion-dollar valuation, a tech giant declaring independence, Congress writing real law, and developers learning what AI tools actually cost.

Editorial illustration: a silhouette transitioning from laboratory to industry — Figure 01 — The lab-to-industry silhouette. Five years from research curiosity to near-trillion-dollar business.

Last Tuesday, I watched Anthropic close a $65 billion funding round. Not million. Billion. At a $965 billion valuation. Then I opened my laptop on Wednesday morning to see Microsoft unveil seven AI models it built entirely without OpenAI’s help. By Thursday, Congress had dropped a 269-page bill to regulate the whole thing. And by the weekend, developers were posting screenshots of projected GitHub Copilot bills jumping from $29 to $750 a month.

I have been building with AI tools every day for the past two years. I have watched this industry go from “interesting research” to “serious product” to whatever this is now. This week made the answer obvious. This is an industrial-scale business with industrial-scale consequences. The lab era is officially over.

Anthropic Becomes the Most Valuable Private Company on Earth

Venture capital closing meeting in a glass high-rise at golden hour — Figure 02 — The kind of room where $65 billion changes hands.

Anthropic closed its Series H on May 28, raising $65 billion at a post-money valuation of $965 billion. Read that again. A company founded in 2021 is now worth more than every private company in history, including OpenAI at $852 billion. The round was led by Altimeter Capital, Dragoneer, Greenoaks, and Sequoia Capital, with Amazon contributing $5 billion as part of a $15 billion hyperscaler tranche.

The numbers behind the valuation are real. Anthropic’s annualized revenue run rate crossed $47 billion at the time of announcement, nearly tripling from its $380 billion February valuation. Claude Code alone generates over $2.5 billion in annualized revenue, giving Anthropic roughly 42% of the AI coding tools market versus OpenAI’s 21%.

Three days later, on June 1, Anthropic confidentially filed its draft S-1 with the SEC. An IPO is coming.

A five-year-old company valued at nearly a trillion dollars means the market genuinely believes AI is not a bubble. But it also means the bar for what counts as a "real" AI company just moved to a place most startups will never reach.

Microsoft Cuts the Cord

Technology keynote stage with a lone presenter and packed audience — Figure 03 — Build 2026. Seven models, zero borrowed weights.

At Microsoft Build on June 2, CEO Mustafa Suleiman announced seven in-house AI models under the MAI family. All built from scratch. No OpenAI distillation. No borrowed weights. This was not a subtle move.

The flagship, MAI-Thinking-1, runs 35 billion active parameters with a 256K context window and scored 97% on AIME 25. Microsoft claims it matches Claude Opus 4.6 on coding benchmarks and beats GPT-5.5 on cost efficiency by 10x. Then there is MAI-Code-1-Flash, a tiny 5-billion-parameter model scoring 51% on SWE-Bench Pro, which is now the default in VS Code. The remaining five cover image generation, voice, transcription, and specialized reasoning.

The message to the industry was clear. Microsoft spent years and billions on its OpenAI partnership. Now it is building its own path. For enterprises choosing an AI stack, this matters enormously. You can now bet on Microsoft’s models, OpenAI’s models, or both, all within the same Azure ecosystem. The power dynamic has shifted.

For those of us building products on top of these platforms, the takeaway is practical: more competition means better pricing, better models, and less vendor lock-in. That is unambiguously good.

Washington Writes Its First Real AI Law

US Capitol Building at twilight with legislative documents in the foreground — Figure 04 — 269 pages of proposed federal AI law, and the building where it will be debated.

On June 4, Representatives Jay Obernolte (R-CA) and Lori Trahan (D-MA) released the Great American Artificial Intelligence Act, a 269-page discussion draft. It is the most comprehensive proposed federal AI framework in U.S. history.

The headline provision: a three-year preemption of state AI development laws. That means California, New York, Illinois, and others would temporarily lose the ability to regulate how AI models are built, though state laws governing how AI is used and deployed remain untouched.

One federal standard, even a strict one, is easier to build for than fifty different rules.

The bill also imposes real obligations. Frontier developers with more than $500 million in gross revenue must hire licensed independent auditors every six months. Civil penalties run up to $1 million per violation per day. And $100 million per year is authorized for fiscal years 2027 through 2029 to fund a new Center for AI Standards and Innovation.

From a founder’s perspective, federal preemption is actually good news. The patchwork of state laws was becoming a compliance nightmare for anyone shipping AI products nationally. Whether this draft survives the legislative process is another question, but the fact that it is bipartisan and this detailed suggests it will not disappear quietly.

The $750 Copilot Bill

Developer staring at a billing dashboard showing a cost spike on their monitor at night — Figure 05 — The moment a developer realizes what “usage-based billing” actually means.

On June 1, Microsoft switched GitHub Copilot from flat-rate subscriptions to usage-based billing. The new plans look reasonable on paper: Copilot Pro+ at $39 per month, Business at $19 per user, Enterprise at $39 per user. Each includes a matching dollar value in AI Credits.

Then reality hit. Heavy agentic users started projecting their monthly costs. $29 became $750. $50 became $3,000. Code completions stayed unlimited, but chat, agentic workflows, and code review now consume tokens at market rates.

The developer community’s response was immediate. GitHub’s community forum post drew over 900 downvotes. Microsoft rushed out promotional credits ($30 per month for Business, $70 for Enterprise) through August 2026. But the damage was done.

This is the first major test of whether developers will pay for AI tools by the token. And right now, the answer looks like no. Not because the tools are not valuable, but because unpredictable costs terrify individual developers and engineering managers alike. The backlash is already accelerating migration to Claude Code, Cursor, and open-source alternatives.

I think Microsoft made the right long-term call. Flat-rate pricing for compute-intensive AI tools is not sustainable. But the rollout was brutal. If you are building AI products with usage-based pricing (and I am), this is a case study in what not to do: surprise your users, give them no cost controls, and let them discover the bill themselves.

That is the week. A near-trillion-dollar private company filing to go public. A tech giant declaring independence from its AI partner. Congress drafting a real law. And developers learning what AI tools actually cost when someone stops subsidizing them.

The common thread is maturity. Every one of these stories would have been unthinkable two years ago. Not because the technology was not there, but because the industry was not ready for the business, regulatory, and pricing structures that come with operating at scale.

It is ready now. Whether the rest of us are ready for what that means is a different question.