--- title: "A decade of internal AI battles is finally catching up to Google" type: "News" locale: "en" url: "https://longbridge.com/en/news/296000890.md" description: "Google faces an AI reckoning as DeepMind talent drains and Gemini model delays persist, highlighting a decade-long internal conflict between commercial product goals and scientific research. Despite recent benchmark successes, experts note Google has lost its early lead to rivals like OpenAI and Anthropic due to organizational friction and strategic hesitation." datetime: "2026-08-15T12:00:17.000Z" locales: - [zh-CN](https://longbridge.com/zh-CN/news/296000890.md) - [en](https://longbridge.com/en/news/296000890.md) - [zh-HK](https://longbridge.com/zh-HK/news/296000890.md) --- # A decade of internal AI battles is finally catching up to Google By Christine Ji DeepMind, acquired by Google in 2014, has been a cornerstone of the company's AI strategy - and also a chronic source of organizational friction Google DeepMind founder Demis Hassabis is stepping back from day-to-day operations while the rest of the company funnels resources to Gemini. As 2025 drew to a close, Google looked like the ultimate artificial-intelligence winner. The company's newest Gemini 3 model swept industry benchmarks, beating out rivals OpenAI and Anthropic in performance after getting a late start to the large-language-model race. Google Search, which industry watchers previously declared dead due to the rise of conversational chatbots, was booming thanks to AI Overviews and AI Mode. With a frontier lab, custom silicon and a sprawling ecosystem with billions of existing users, Google parent Alphabet (GOOGL) (GOOG) was praised by Wall Street for its full-stack solutions. Yet today, Google faces a major AI reckoning. The company's DeepMind lab is slowly bleeding talent, and the company has delayed the Gemini 3.5 Pro model indefinitely and may be considering shelving it altogether in favor of Gemini 4.0 Pro, according to industry reports. Earlier this month, Google veteran Jeff Dean departed along with three of the company's top AI researchers to launch the startup Discovery Loop. At the same time, Demis Hassabis transitioned from his previous position as Google DeepMind CEO to a more research-centric role as Alphabet's chief scientist. The challenges reflect the culmination of a long-standing identity crisis inside Google: whether to prioritize mass-market commercial products, or build tools for scientific research. 'AI-first' - but second place "In total, I'd say Google was on the frontier of LLMs for about two to three months total between the release of ChatGPT and the present day," Dan Schwarz, co-founder and CEO of the AI forecasting platform FutureSearch, told MarketWatch. Google did not respond to a request for comment. Gemini 2.5 Pro, released in March 2025, and Gemini 3.0 Pro, released in November 2025, enjoyed brief stints at the frontier until Anthropic's Claude models usurped them, Schwarz said. Today, Gemini lags the competition in critical areas like coding. According to AI benchmarking site Artificial Analysis, the latest Gemini 3.7 flash model ranks ninth in on its composite Intelligence Index, behind models from Anthropic, OpenAI, Alibaba (BABA) and SpaceXAI (SPCX). For a company that has owned a pioneering AI lab since 2014 and has proclaimed itself "AI-first" since 2016, Google seems to have squandered an early lead, experts say. But the company's AI strategy has faced internal culture clashes and organizational hurdles since the very beginning. In the years following its acquisition by Google, the London-based DeepMind primarily functioned as an independent research entity instead of a integrated product unit. Then-CEO Hassabis founded the company in 2010 with the goal of achieving artificial general intelligence, or AGI - an AI system that matches or surpasses human capabilities. As a result, Hassabis had little interest in focusing on internet applications, and he often clashed with Alphabet CEO Sundar Pichai to preserve the lab's autonomy, according to Sebastian Mallaby's 2026 biography of Hassabis, "The Infinity Machine." "DeepMind from its very early days was focused on a very particular type of artificial intelligence, reinforcement learning," Benjamin Lee, an engineering professor at the University of Pennsylvania, told MarketWatch. Reinforcement learning - a technique that trains AI agents through reward signals - led to DeepMind's biggest breakthroughs such as AlphaGo, an AI agent that defeated human professionals at the board game Go, as well as AlphaFold, a Nobel Prize-winning AI system that predicts protein structures. DeepMind "demonstrated all of these great capabilities early on, like playing games or protein folding," Lee said. "But that's not what's being commercialized by the rest of the industry." 'Asleep at the wheel' Already a successful publicly traded company thanks to its cash-printing Search business, Google had the financial resources to support DeepMind's scientific pursuits - for a while, at least. Along with Google's deep pockets came fiduciary duties to shareholders and clashing organizational interests. During the same period that Google and DeepMind were navigating their new relationship, a rival lab by the name of OpenAI had begun working on a large language model. OpenAI's talent, funding and scientific papers paled in comparison to Google. But in November 2022, OpenAI shocked the world with the introduction of ChatGPT. It was a sign that Google was "asleep at the wheel," FutureSearch's Schwarz said. Through the 2010s, Google was working on machine-learning solutions to improve the accuracy of its search engine. Schwarz, who previously worked in the Google Research organization, recalled internal debates about the potential risks of such a strategy. The possibility of AI hallucinations would strike a serious blow to Google Search's credibility. Amit Singhal, who led Google's Search team until 2016, circulated a famous internal memo opposing the use of machine learning in search functions - arguing that machine-learning models would make search rankings difficult for human engineers to interpret. In a stroke of irony, Google's deep-learning division, Google Brain, had deployed chatbots internally since 2020. Employees across the company tested and used early models like Meena and LaMDA. The company encouraged employees to spend time testing out the chatbots, Matan Zinger, a former Google engineer who worked on products like Search and Google Wallet, told MarketWatch. During the pandemic, some employees would spent their lunch breaks conversing with the chatbots for fun, Zinger said. Google's top researchers also wasted resources competing against each other. After initially brushing off LLMs, DeepMind was developing its own chatbots. In 2020, the lab embarked on training a chatbot dubbed "Gopher," purposely keeping the project secret from Google Brain. In 2023, Pichai merged the Silicon Valley-based Google Brain with DeepMind. Google DeepMind was tasked with accelerating the company's AI progress. From AGI to Gemini The recent departures from Google DeepMind also signify an ideological rift between Pichai and Hassabis, according to former employees. "Sundar does not believe in AGI," Schwarz speculated. Zinger expressed the same sentiment in a Substack post last week, writing that Google's AI strategy is solidifying around "prosaic matters" such as coding and search, instead of DeepMind's original vision of chasing AGI. DeepMind's recent departures actually seem like a bullish indicator for Gemini's capabilities as the organization consolidates its focus on commercial applications, according to Zinger. In yet another sign that the company is tightening its focus on commercial AI applications, Google disbanded its AlphaFold research team in July. AlphaFold co-creator John Jumper departed for Anthropic. "Search has always been the most important thing at Google," Zinger told MarketWatch. "If you see where Google is still strong, it's the flash models, which are optimized but not as capable. But when you hit a certain scale, they work fast and you can serve billions of users." Google has indeed been successful in supercharging search functions with AI. Earlier this week, Pichai shared that Gemini officially crossed the 1 billion-user mark. The company's lighter flash models power AI Overviews and AI Mode, which allow search-engine users to perform complex queries in natural language - far surpassing the keyword matching and ranking techniques of previous generations. Additionally, the company's cloud business has more than doubled in the past three years as Google sells AI compute to eager enterprise buyers. As the University of Pennsylvania's Lee put it, Google is running a different AI strategy than companies like OpenAI and Anthropic, which are primarily developing models and selling tokens to application developers. With multiple platforms boasting over 1 billion users, Google's advantage comes from being able to easily distribute its Gemini models. "Google would argue that the search engine is in many ways the way people are accessing the language model and Gemini, right?" Lee said. As the AI race goes on, it's still far too early to declare winners and losers. Thus far, no company has achieved an "insurmountable lead," according to Lee. Now, as the world awaits Google's next Gemini model, the tech giant will need to show that it can maneuver the AI competition with more agility than it has in the past. \-Christine Ji This content was created by MarketWatch, which is operated by Dow Jones & Co. MarketWatch is published independently from Dow Jones Newswires and The Wall Street Journal. (END) Dow Jones Newswires 08-15-26 0800ET ### Related Stocks - [GOOG.US](https://longbridge.com/en/quote/GOOG.US.md) - [GOOGL.US](https://longbridge.com/en/quote/GOOGL.US.md) - [GGLS.US](https://longbridge.com/en/quote/GGLS.US.md) - [GOOW.US](https://longbridge.com/en/quote/GOOW.US.md) - [GGLL.US](https://longbridge.com/en/quote/GGLL.US.md) - [GOOY.US](https://longbridge.com/en/quote/GOOY.US.md) - [GOU.US](https://longbridge.com/en/quote/GOU.US.md) - [GOOX.US](https://longbridge.com/en/quote/GOOX.US.md) - [GOOP.US](https://longbridge.com/en/quote/GOOP.US.md) - [MAGX.US](https://longbridge.com/en/quote/MAGX.US.md) - [ANTH.NA](https://longbridge.com/en/quote/ANTH.NA.md) - [BABA.US](https://longbridge.com/en/quote/BABA.US.md) - [SPCX.US](https://longbridge.com/en/quote/SPCX.US.md) - [89988.HK](https://longbridge.com/en/quote/89988.HK.md) - [09988.HK](https://longbridge.com/en/quote/09988.HK.md) - [HBBD.SG](https://longbridge.com/en/quote/HBBD.SG.md) ## Related News & Research - [Fiduciary Financial Group LLC Acquires 7,713 Shares of Alphabet Inc. $GOOG](https://longbridge.com/en/news/295645335.md) - [EXCLUSIVE-Inside the Google executive moves that led to its big AI reshuffle](https://longbridge.com/en/news/295694902.md) - [Google Launches Gemini 3.7 Flash: Major Leap in Coding and Agent Capabilities, Three-Week Iteration Sets New Speed Record for Low-Cost Models](https://longbridge.com/en/news/295829050.md) - [Alphabet Names New Head Of DeepMind A.I. Unit](https://longbridge.com/en/news/295675310.md) - [Angi Among the First Home Services Marketplaces Available in Gemini as a Connected App | ANGI Stock News](https://longbridge.com/en/news/295680195.md)