AI News Roundup: Anthropic IPO, G7 AI Summit, and Microsoft's Independence Play
Anthropic files for a record IPO at $965B, G7 leaders meet with AI CEOs in France, and Microsoft launches 7 new models to break free from OpenAI. Here is everything you need to know.
TL;DR
Anthropic filed for what could be the biggest tech IPO in history. The G7 summit in France just wrapped with Sam Altman, Dario Amodei, and Demis Hassabis all in the same room. Microsoft dropped 7 new in-house AI models to reduce its OpenAI dependency. And Nvidia launched a new superchip that puts AI agents directly on your laptop. Here is my take on what actually matters this week.
Anthropic files for IPO at $965 billion
This is the big one. Anthropic confidentially filed its S-1 with the SEC, targeting a valuation around $965 billion. For context, that is higher than the GDP of most countries.
The numbers behind it are staggering. Anthropic closed a $65 billion Series H round right before filing. Its annualized revenue run rate hit $47 billion, driven almost entirely by Claude enterprise adoption for coding and agent workflows. Morgan Stanley and Goldman Sachs are leading the underwriting.
One detail that did not get enough attention: Anthropic disclosed that Claude now writes over 80 percent of its own codebase. Not boilerplate. Not tests. The actual product code. Before Claude Code launched in early 2025, that number was in the low single digits. Whether you find that impressive or terrifying probably depends on your job title.
I have been covering Anthropic since the Claude 3 days, and I wrote about their Claude Code pricing war back in May. The speed of their growth is unlike anything I have seen in tech, not even during the dot-com era.
OpenAI is not far behind. They also filed confidentially, with reports pointing to a valuation above $1 trillion. Both companies are racing to be the first pure-play AI lab to go public. SpaceX already beat them to the punch with their $75 billion IPO on June 12, but that is a different category entirely.
G7 Summit puts AI governance front and center
The G7 summit in Evian-les-Bains, France wrapped up on June 17 with AI as the dominant topic. Sam Altman, Dario Amodei, and Demis Hassabis all attended, which is a first. You do not usually see the CEOs of the three most important AI companies in the same room with world leaders.
The key outcome was a joint statement on AI export controls and a framework for "coordinated pauses" on frontier AI development. Anthropic had published a report right before the summit arguing that the length of autonomous tasks models can complete is doubling roughly every 4 months. That pace, if it holds, means models could autonomously complete month-long projects by sometime next year.
The report also noted that typical engineers now merge roughly 8 times more code per day than in 2024. I do not think the G7 leaders fully grasp how fast this is moving. Most of them are still asking basic questions about what a language model is. Meanwhile, the labs they are trying to regulate are shipping capabilities that would have been science fiction two years ago.
Microsoft Build 2026: 7 new MAI models
Microsoft made its clearest move yet to reduce reliance on OpenAI. At Build 2026, they launched 7 proprietary models under the "MAI" brand, all trained from scratch with no distillation from OpenAI.
The lineup includes MAI-Thinking-1, a reasoning model with roughly 35 billion active parameters and a 256K context window. Early benchmarks put it close to leading frontier models on multi-step reasoning tasks. MAI-Code-1 and MAI-Code-1-Flash are already rolling into GitHub Copilot, which is switching to usage-based token billing. Developers are not happy about that part.
The full stack now covers reasoning, code generation, image generation, voice synthesis, and transcription. Microsoft controls the entire pipeline on Azure and Foundry. Satya Nadella framed it as "optionality," not a divorce. But the writing is on the wall. When your biggest partner starts building competing models from scratch on clean data, the relationship has changed.
I compared the major chatbot options in my ChatGPT vs Claude vs Gemini guide. With Microsoft entering the chat, I might need to update that comparison soon.
Nvidia RTX Spark brings AI agents to your laptop
At Computex 2026, Nvidia unveiled the RTX Spark, a new superchip that combines an Arm CPU, Blackwell GPU, AI accelerators, and unified memory into a single package. The pitch is simple: run advanced AI agents locally on your laptop without touching the cloud.
Dell, HP, Lenovo, Asus, and Microsoft are all expected to ship RTX Spark devices by autumn 2026. Jensen Huang called this the start of "Physical AI," which he estimates is a multi-trillion dollar market. Nvidia also partnered with Unitree Robotics on the Isaac GR00T humanoid robot platform.
This matters because it changes the economics of AI. Running models on-device means lower latency, better privacy, and no per-token API costs. For developers building AI agents that need to run all day without a network connection, this changes things.
Other stories worth your attention
China's BAAI unveiled Physis-v0.1, what they call the world's first general world foundation model. Unlike language models, it understands physics and spatial relationships. The obvious application is robotics, but digital twins and scientific research are close behind.
SpaceX went public on June 12 at a $1.77 trillion valuation, raising $75 billion. The AI angle: SpaceX has massive compute contracts with Anthropic ($1.25 billion per month) and Google ($920 million per month) through xAI data center infrastructure.
Alphabet raised $84.75 billion in the largest equity capital markets transaction in history. $10 billion of that came from Berkshire Hathaway. The money is going straight into AI data center infrastructure, with 2026 CapEx guidance pushed to $180-190 billion.
The Great American AI Act discussion draft landed in Congress, proposing a 3-year freeze on state AI laws and mandatory safety frameworks for companies with over $500 million in revenue. Criminal penalties for AI impersonation of government officials are included.
FAQ
Q: Should I care about the Anthropic IPO as a regular user?
Not directly. But IPOs create pressure to monetize, and that usually means price increases or feature gating. If you rely on Claude for work, keep an eye on it.
Q: Will Microsoft really stop using OpenAI models?
No. They will use both. But having in-house models gives Microsoft stronger negotiating power and a backup if the OpenAI relationship goes further south. For users, it means more choice and potentially lower costs.
Q: When can I actually buy an RTX Spark laptop?
Autumn 2026 is the target. I would not pre-order the first generation. Wait for real-world battery life and thermal benchmarks before dropping several thousand dollars on an AI laptop.
Q: Is the G7 actually going to regulate AI?
The joint statement is non-binding. Real regulation will come from individual countries. The US Great American AI Act is the one to watch. If it passes, it would preempt state laws and create a federal framework. If it stalls, California and Colorado regulations will fill the gap.