Why Are Google’s Best AI Scientists Leaving? Inside the Talent War Reshaping Artificial Intelligence

Something unusual happened in the third week of June 2026. In the span of six days, four of Google’s most consequential AI researchers announced they were leaving. Not in the usual way a quiet resignation here, a retirement there but in rapid succession, for direct competitors, with the kind of public finality that makes investors nervous and rivals optimistic. The exits hit Google’s Gemini team, its DeepMind lab, and its coding AI efforts all at once.

In any other industry, losing a few senior researchers might barely register. In frontier AI, where the difference between a breakthrough and a dead end often lives inside a handful of minds, it’s a different kind of problem entirely.

Why Everyone Is Talking About Google

Google’s relationship with artificial intelligence is complicated by the fact that it essentially invented the modern version of it. Every major large language model deployed at scale today GPT-4, Claude, Gemini, Llama, Mistral is built on the Transformer architecture. That architecture came out of Google Brain in 2017. For years, Google was the place where the most important AI research happened. It attracted the best people, gave them enormous compute budgets, and produced work that defined the field.

Then came the generative AI boom, and suddenly the company found itself reacting rather than leading. Google spent much of the current AI boom playing catch-up with the likes of OpenAI and Anthropic before hitting its stride late last year with more capable models and chips. The gap has narrowed, but the narrative around Google’s ability to retain its best people has not recovered and the events of late June made that much worse.

The exits highlight the pressure Google faces from two startups that are on the cusp of going public, offering even well heeled employees at Big Tech firms the chance at a rare payday by signing on before an IPO. That pressure is not subtle. It is the kind that shows up in stock prices and internal morale simultaneously.

Who Left and Why It Matters

The week began with Noam Shazeer. Shazeer co authored the paper “Attention Is All You Need,” which introduced the Transformer architecture behind most of today’s large language models. That single paper, published in 2017, is arguably the most consequential piece of AI research of the past decade. Every chatbot you have ever used, every AI coding assistant, every image generator the underlying logic traces back to it. Shazeer didn’t just contribute to that paper; he proposed scaled dot product attention, multi head attention, and the parameter-free position representation, becoming involved in nearly every detail of the work.

He left Google once before, in 2021, to found the chatbot startup Character.AI. Google brought him back in 2024 through an unusual licensing deal that valued Character.AI at $2.5 billion, and once back, Shazeer co led development of the company’s flagship Gemini AI model. He stayed less than two years. Shortly before Shazeer announced his plans to join OpenAI, computing power dedicated to one of his projects was reassigned to a London based team at Google DeepMind, in an attempt to boost collaboration across teams and streamline pre training. Whether that reassignment was the deciding factor or merely the final straw, the result was the same.

Two days later, John Jumper announced he was leaving for Anthropic. Jumper led the AlphaFold project at Google DeepMind, shared the 2024 Nobel Prize in Chemistry for that work, and spent nearly nine years at the company. AlphaFold a system that predicts protein structures with extraordinary accuracy is widely considered one of the most important scientific contributions AI has ever made to biology. Jumper did not leave quietly. He said he planned to take time to recharge before starting at Anthropic, which suggests this was not a hasty decision.

Then came Jonas Adler and Alexander Pritzel. Both were viewed internally as key contributors to Google’s Gemini AI model. Adler worked on the company’s AI coding effort and Pritzel was involved in the process of training AI systems. Both also worked with Jumper on AlphaFold. Arthur Conmy, a senior research engineer who contributed to the Gemini 2.5 model and AI coding, also announced his departure to join Anthropic to work on AI safety.

Five departures. Six days. Three of Google’s own flagship model contributors heading to the same rival. The pattern is too concentrated to be coincidental.

Why Anthropic Keeps Attracting Top Talent

The most revealing data point in this entire story comes from a 2025 analysis by venture capital firm SignalFire: DeepMind engineers are nearly 11 times more likely to leave for Anthropic than the reverse. (Outlook Business) That number should stop anyone trying to frame this as normal industry churn. It describes a one directional gravitational pull.

Part of it is financial. Anthropic has raised $65 billion at a $965 billion post money valuation in its latest funding round, marking what could be the AI startup’s last private fundraising before debuting on the public markets. (TechCrunch) Pre IPO equity at a company approaching a trillion dollar valuation is a genuinely life changing offer, and it’s the kind of offer a public company like Alphabet simply cannot match with its own stock.

But equity alone doesn’t explain the pull completely. Researchers at this level are not primarily motivated by compensation they could extract enormous salaries anywhere. What they want is the ability to work on hard problems without bureaucratic friction, and the sense that their work will matter to the outcome. Top talent wants to land at the lab that wins the race for AGI. Right now, a significant portion of the world’s most capable AI researchers believe that lab is Anthropic.

Anthropic was founded in 2021 by former OpenAI researchers, including Dario and Daniela Amodei, who left over concerns about AI safety and organizational direction. That founding ethos serious about safety, serious about frontier research, unwilling to treat those two goals as contradictory appears to remain a genuine draw. John Jumper is reportedly joining to work on AI in life sciences and healthcare, an area where Anthropic has been building quietly. Arthur Conmy explicitly said he is joining to work on AI safety. These are not people chasing the biggest paycheck. They are people choosing where they want the next chapter of this technology to be written.

Is OpenAI Winning Too?

Shazeer’s move to OpenAI was, in some ways, the sharper symbolic blow. The co-inventor of the architecture that powers every major AI model is now at the company most people associate with that technology in the public imagination. OpenAI CEO Sam Altman welcomed the hire publicly on X. (Search Engine Journal)

OpenAI filed confidential IPO paperwork earlier this month, giving it a similar pre-public equity story to Anthropic. The two companies have been running parallel recruitment campaigns, often targeting the same pool of people and often drawing from the same source: Google DeepMind. Earlier this year, Microsoft hired around two dozen employees from Google’s DeepMind AI research lab. (CNBC) Meta has reportedly offered signing bonuses of up to $100 million to retain and attract senior AI researchers.

Google has not been passive. In 2025, some 20% of AI software engineers Google hired were former employees, and the company saw a jump in AI researchers coming from major competitors compared to 2024. (CNBC) It has launched an internal incubator for departing employees, brought Sergey Brin out of retirement to personally recruit candidates, and restructured its management layers to reduce bureaucracy. None of it stopped this week’s departures.

The honest picture is that OpenAI and Anthropic are both winning this talent war. What Google is losing is the perception and perhaps the reality that it is the place where the most important work is being done.

What This Means for AI Users

If you use Claude, ChatGPT, or Gemini, the movement of individual researchers might feel irrelevant. The product still loads. The answers still come. But research talent is not a background variable in this industry it is the primary input. The quality of future AI models is inseparable from the quality of the people building them.

Consider what Anthropic is assembling. A Nobel laureate who built one of the most celebrated AI systems in scientific history. Multiple researchers who contributed directly to the architecture and training of Google’s best model. A safety researcher who helped shape the Gemini 2.5 model. These are not peripheral figures arriving to do incremental work. They are people who know, at a deep level, how frontier AI systems are built and they will apply that knowledge to building Claude’s next generation.

Anthropic has an AI for Science event scheduled for June 30, and Jumper’s hire in particular signals where some of that future effort is pointed: real-world scientific applications, the kind that don’t just answer questions but contribute to drug discovery, protein biology, and the kind of research that historically required decades of laboratory work. If Anthropic delivers on that, the Claude you use in two years may be meaningfully more capable in domains that matter beyond conversation.

For Gemini users, the more immediate concern is continuity. Google’s Gemini 3.5 Pro has reportedly been delayed until July for final adjustments. Losing this many senior researchers right before a major model launch doesn’t help the narrative. Google has enormous resources and a vast research bench, and Demis Hassabis is right that the company still employs far more AI researchers than any rival. But depth of bench and concentration of exceptional talent are different things.

What It Means for Investors

Alphabet’s stock dipped on the news, as it tends to do when high profile exits make headlines. The market is probably right to register concern, even if the reaction is measured. Talent losses at this level compound over time: the researchers who leave take knowledge with them, the models they would have improved get less of their attention, and the rival models they do improve become stronger comparisons.

Anthropic’s valuation has risen more than fifteenfold in just over a year from $61.5 billion in March 2025 to $965 billion in the Series H round of May 2026. That is a remarkable trajectory, and the talent story is not incidental to it. Investors are, in part, betting on the people building these systems. Every high profile researcher who chooses Anthropic over Google is a small validation of that bet.

With its confidential S-1 filing, Anthropic is poised to potentially beat OpenAI to the public market, setting itself up to attract attention and capital from a greater pool of investors. An IPO at or near a trillion dollar valuation would be among the largest in technology history. The talent narrative the idea that the people who built Google’s best models are now building Anthropic’s next ones will be part of how that story gets told to public market investors.

For Alphabet, the long term strategic question is harder. Google built the foundation of modern AI. It published the research that the entire industry now runs on. If it cannot retain the people responsible for that research, the question is not just about stock price it is about whether the organization that created this technology will be the one that benefits most from it.

FAQs

Further Reading on BEXORN

• The AI IPO Race 2026: SpaceX, Anthropic and OpenAI Are All Going Public at the Same Time

• Claude AI Growth 2026: The 306% Number That Should Worry Every ChatGPT User

• Claude Fable 5 vs ChatGPT 2026: The Battle Forcing Millions to Finally Ditch One AI for the Other

• What Is Google Gemini and Should You Use It in 2026? (Honest Review)

• How to Use Claude AI for Beginners: 7 Simple Steps to Powerful Results 2026