Good morning. It's Thursday, September 17, and we're covering OpenAI’s potential $1.5 trillion valuation, new U.S.-China talks on AI risks, and Google Home’s move to let AI agents control smart devices.

Plus: yesterday’s poll on whether AI development should slow down, and a new Real or AI challenge.

Google is giving AI agents the ability to control smart-home devices, not just answer questions about them. An early-access Google Home MCP server lets Claude, ChatGPT, OpenClaw and other AI systems control supported devices and review activity using natural language. The rollout begins September 16 for U.S. subscribers to the $20/month Google Home Premium Advanced plan, with Nest and Matter devices supported.

Google, NVIDIA and Emerald AI are betting that data centers can ease grid pressure by changing when they consume electricity. Their new AI Energy Management Alliance includes 20 AI and energy companies, including Anthropic, National Grid and Constellation. Google says it can reduce 1 GW of demand when needed, while 84% of Americans express concern about data centers’ impact on local power prices.

Apple’s new camera system signs pixel data before software processing, creating a cryptographically protected “digital negative” at capture. The opt-in Apple Reference Image feature on the iPhone 18 Pro and Pro Max is designed to verify that photos came from the camera sensor and weren’t AI-generated or altered. Secure Enclave metadata and cryptographic timestamps further protect capture details before development via Private Cloud Compute.

OpenAI is considering a new funding round that could value the artificial intelligence company at about $1.5 trillion, roughly double its latest private valuation of $730 billion. Investors recently proposed a $1.2 trillion valuation, but OpenAI believes its stronger customer traction and latest models support a higher figure, according to people familiar with the discussions.

The company says its annualized revenue exceeded $40 billion in August, about twice its level at the end of 2025. No deal has been finalized, and OpenAI’s confidential June filing with the U.S. Securities and Exchange Commission could complicate how it shares information with prospective investors. → Read the full article here.

Treasury Secretary Scott Bessent told Axios on September 16, 2026, that the U.S. is open to discussing shared artificial intelligence risks with China during high-level talks in New York this weekend. The discussions are expected to cover AI, trade and rare earths, with Bessent saying both open- and closed-weight AI models could be addressed.

The talks come ahead of President Donald Trump’s planned September 24 summit with Chinese President Xi Jinping and follow calls from AI executives for greater U.S.-China coordination on safety standards. China has separately called for open, inclusive AI development and warned that confrontation could hinder global AI governance. → Read the full article here.

SK Hynix Eyes Intel’s Ohio Fabs: The companies are discussing a lease or joint venture to produce memory chips in the US for the first time.

Google DeepMind Launches AGI Institute: The new forum will study artificial general intelligence’s societal impacts and publish accessible research.

May Mobility Goes Public: The autonomous vehicle company will merge with a SPAC at a $1.4B valuation, potentially raising over $300M.

OpenAI Launches AI Scam Tour: OpenAI and AARP will visit 10 US cities to teach older adults how AI can help spot scams.

Apple Eyes Server Market Return (Paywall): Apple is considering reentering the server market and has discussed using NVIDIA networking technology.

Here's how you voted: Opinions were split, with 44% backing continued progress and 43% saying AI safety needs to catch up.

Thanks for reading. See you next time!

— Matthew Berman, Nick Wentz & the Forward Future Team