Tech giants Google and Meta have simultaneously unveiled their latest strategic adjustments in the artificial intelligence sector, drawing significant industry attention. Google announced a major leadership reshuffle: Nobel laureate and DeepMind founder Demis Hassabis is formally stepping down from day-to-day management to focus on long-term scientific research. At the same time, legendary engineer Jeff Dean, who spent nearly 27 years at Google, announced his departure and has co-founded AI startup Discovery Loop alongside several former senior Google scientists. On the other side, Meta released “Muse Code,” a terminal-based coding agent, aggressively positioning itself in the AI programming development arena.
According to Google’s announcement on August 5, Hassabis will step down as CEO of DeepMind, transitioning to the role of Chairman of DeepMind and Chief Scientist at Alphabet. He will dedicate more energy to long-term AI research and continue to lead Isomorphic Labs, a subsidiary focused on AI-driven drug discovery. Day-to-day operations at DeepMind will be taken over by long-time CTO Koray Kavukcuoglu, who will be promoted to Senior Vice President, reporting directly to Google CEO Sundar Pichai. He will be responsible for Gemini model development, frontier research, and related applications and developer teams.
This move is interpreted by outsiders as Google attempting to clearly bifurcate its AI development strategy into two tracks: a “product clock” pursuing speed to market, and a “civilization clock” focused on breakthroughs in artificial general intelligence (AGI) and fundamental science. The aim is to separate the “factory floor” of product development from the “observatory” of basic research. On the day of the announcement, Alphabet’s stock closed down more than 4%, having fallen over 5% intraday, wiping out over $180 billion in market value in a single day, signaling market sensitivity and skepticism regarding the search giant’s AI strategic pivot.
The personnel earthquake at Google extends far beyond this. Almost simultaneously, Jeff Dean—Google’s employee number 30, former Chief Scientist, and affectionately known as “Jeff” throughout the industry—announced his departure from the company he served for 27 years. He has co-founded a Public Benefit Corporation named Discovery Loop with long-time collaborator Sanjay Ghemawat, along with two other top Google AI scientists, Quoc Le and Oriol Vinyals.
Discovery Loop’s mission is ambitious: “Automate machine learning, science, and engineering to accelerate discovery and progress.” The core idea is to establish an “automated experimentation loop,” inserting AI into the entire scientific research process: proposing hypotheses, designing experiments, running them, analyzing results, modifying plans, and then automatically initiating the next cycle. The company will initially focus on machine learning research and engineering—the area the four founders know best—with future potential to extend this methodology to computer hardware, drug discovery, and other scientific and engineering fields. Google itself will continue to support the new company as a founding investor and cloud partner.
The startup’s fundraising process is the stuff of legend. Jeff Dean’s business plan was reportedly just three pages long, showcasing the team’s past brilliant achievements, the massive teams they’ve led and AI talent they’ve cultivated, and their enormous influence in academia. The products and infrastructure built by this team span nearly the entirety of Google’s core technology stack, from Google Search, Ads, and Gmail to TensorFlow and the Gemini models. The roster of talent they’ve mentored is a veritable who’s-who, including Anthropic founder Dario Amodei, OpenAI co-founder Ilya Sutskever, and Yang Zhilin, founder of Chinese AI startup Moonshot AI.
Faced with such a stellar lineup, Silicon Valley venture capital firms were practically “lining up to hand over money.” Confirmed investors include Khosla Ventures, Radical Ventures, Lightspeed Venture Partners, Kleiner Perkins, and Doerr Capital. Khosla Ventures and Radical Ventures are co-lead investors. Khosla Ventures founder Vinod Khosla even excitedly posted on social media: “Truly honored to be chosen to lead the Discovery Loop round!” He compared this investment to his firm’s 2018 investment in OpenAI. Although the specific funding amount has not been officially announced, sources familiar with the matter revealed the round size reaches “hundreds of millions of dollars.”
While Google was undergoing its leadership restructuring, social media giant Meta also dropped a bombshell in the AI field. Meta AI recently released a test version of Muse Code, a terminal-based coding agent. Unlike traditional code auto-completion tools, Muse Code is powered by the Muse Spark 1.2 model and can autonomously handle complex tasks such as change planning, code writing, and result verification within large code repositories. It can even distribute large tasks to multiple sub-agents for parallel processing and features action logging that allows for automatic recovery after a system crash. This signals Meta’s major push into the AI coding domain, attempting to build intelligent agents capable of autonomously writing software.
The investment frenzy in AI infrastructure also shows no signs of cooling. Cloud AI startup Anthropic has been confirmed to have signed a computing deal worth up to $10 billion with data center operator Volta. Additionally, Bitdeer has signed a 16-year colocation lease agreement with Volta Tydal AS to deploy 121 IT MW of Nvidia GPUs in Norway, exclusively for Anthropic’s use. Meanwhile, SK hynix’s board approved a massive investment of approximately 54 trillion won (approximately $38.1 billion) to build new DRAM and NAND fabrication plants in Yongin and Cheongju, South Korea, to meet the continuously growing demand for memory in the AI era.
In the Chinese market, competition and commercialization in the AI model space are also accelerating. DeepSeek announced plans to broadly raise its API service pricing soon, with the increase expected to be substantial, marking a strategic shift for large model vendors from “trading low prices for market share” to “value-based pricing.” Previously, the DeepSeek V4 Flash model had already processed 8 trillion tokens in a single day on overseas developer platforms, demonstrating robust market demand. On another front, the UK’s AI Safety Institute (AISI) released a report indicating that the latest models from Anthropic and OpenAI exhibited out-of-control behavior during safety testing, taking actions on the real internet directed at real individuals and organizations without authorization, once again sounding a safety alarm for the rapidly advancing AI industry.