GM AI lover,

Mira Murati, the former CTO of OpenAI, has launched a new AI lab with an astonishing $12 billion valuation. The massive valuation comes from a $2 billion seed round, all secured before the company has even released a single product.

The stealth startup, Thinking Machines Lab AI, is already sparking massive industry hype with claims of breakthroughs on difficult problems. With such a high valuation and a proven leader at the helm, the key question is whether this new lab can live up to the immense expectations and truly challenge established players.

In today’s AI recap:

• Mira Murati’s new $12B AI lab

• Anthropic’s Claude for Wall Street

• Nvidia resumes China H20 chip sales

• Mistral’s open-source voice models

Mira Murati's $12B Mystery Box

The Recap: Former OpenAI CTO Mira Murati just announced a massive $2 billion seed round for her stealth startup, Thinking Machines Lab AI, rocketing its valuation to $12 billion with no product yet released.

Unpacked:

• The lab is developing multimodal AI that interacts with users through conversation and sight, with its first product set to feature a major open-source component.

• According to a report from The Information, the startup plans to build custom AI models to help businesses increase profits.

• The company’s research head, Alexander Kirillov, claims the team has already made breakthroughs on problems that “previously seemed intractable,” fueling massive investor and industry hype.

Bottom line: This massive pre-product valuation highlights the intense investor belief in proven AI talent to deliver breakthrough technology. Murati’s entry officially puts another heavyweight contender in the ring, promising to accelerate competition among the top AI labs.

Claude Takes on Wall Street

The Recap: Anthropic just launched Claude for Financial Services, a new platform that aims to give financial professionals an AI-powered co-pilot for complex market analysis and research.

Unpacked:

• The platform connects external market data from sources like S&P Global and FactSet with a firm’s internal data, allowing analysts to ask complex questions in a single interface.

• Early results show impressive gains, with AIG reporting it can compress underwriting review timelines by 5x, and on the latest finance benchmark, Claude is nearly neck-and-neck with OpenAI’s top model.

• Anthropic is building an entire ecosystem around the tool, with pre-built connectors to data providers like PitchBook and enterprise platforms like Snowflake and Databricks.

Bottom line: This launch isn't about replacing analysts, but about equipping them with specialized AI that can handle the tedious data-wrangling. It marks a significant push for AI to solve specific, high-value problems within major industries.

Nvidia's China Chip Re-Entry

The Recap: In a significant policy shift, Nvidia has received the go-ahead from the U.S. government to resume selling its H20 AI chips to China. The decision reopens a critical market for the chipmaker and signals a new phase in the tech competition between the two nations.

Unpacked:

• The reversal is part of a larger U.S.-China trade agreement, where the U.S. eased chip export restrictions in exchange for China increasing shipments of critical rare earth materials.

• The H20 chip is a less powerful alternative to Nvidia's top-tier processors, designed specifically to comply with earlier U.S. export controls before sales were halted in April.

• Nvidia's CEO Jensen Huang argued that restricting sales would only compel China to develop its own chip industry, potentially undermining America's long-term AI leadership.

Bottom line: This move provides Nvidia access to a market worth billions and eases some supply chain tensions for Chinese tech firms. It also marks a strategic pivot in U.S. policy, balancing immediate trade interests against the long-term goal of slowing China's technological advancement.

Mistral's Voice Challenger

The Recap: French AI leader Mistral has released 'Voxtral', a new family of open-source speech models designed to take on competitors like OpenAI by combining top-tier transcription with powerful language understanding at a lower cost.

Unpacked:

• Voxtral goes beyond simple transcription by integrating native Q&A, summarization, and even function-calling, allowing developers to build more interactive voice-powered apps.

• The models are positioned to undercut the market, with Voxtral Small matching ElevenLabs Scribe's performance and the API version beating OpenAI's Whisper for less than half the price.

• True to its roots, Mistral released both the 24B and 3B parameter models under a permissive Apache 2.0 license, making them available to download and run.

Bottom line: Mistral is applying its successful open-source playbook to the voice AI space, directly challenging the cost and performance of proprietary systems. This move gives developers and businesses powerful new building blocks to create more capable and accessible voice-driven products.

The Shortlist

Runway launched Act-Two, its next-generation motion capture model that translates a single video performance into a fully animated character with detailed head, face, body, and hand tracking.

Researchers published a paper co-signed by leaders from OpenAI, DeepMind, and Anthropic, calling for the preservation of "chain-of-thought" monitoring in AI models as a vital but potentially fleeting safety tool.

Google revealed its AI security agent, Big Sleep, discovered and helped patch a critical software vulnerability in the wild before it could be exploited by attackers.

Google is investing $25B in new data centers and AI infrastructure across the PJM electric grid, which covers 13 U.S. states and the District of Columbia.