Can Google Strike Back? DeepMind Restructuring, AI Talent Exodus, and Gemini Delays

Google (GOOGL-US) is at a critical juncture in the AI race. On August 5, Google CEO Sundar Pichai and DeepMind co-founder and CEO Demis Hassabis jointly issued an internal memo announcing that Hassabis will step down as CEO of DeepMind and transition to Chairman, while also serving as Chief Scientist at parent company Alphabet.

Minutes after the announcement, another heavyweight in the AI field revealed their departure. Jeff Dean, DeepMind’s Chief Scientist who had been with Google for 27 years, announced he would co-found a new company focused on 'AI for scientific research' alongside several colleagues.

The sudden shift of two senior DeepMind figures—one stepping up, the other stepping out—has prompted renewed scrutiny of Google’s position in the AI race: Is Hassabis gaining influence or exiting the front lines? Is decision-making power at DeepMind gradually reverting to Google’s headquarters? And is this tech giant, once at the forefront of AI breakthroughs, losing its competitive edge?

Talent Drain at DeepMind Raises Alarm Over Brain Drain

DeepMind’s talent exodus didn’t begin in August.

On June 18, one of the authors of the seminal paper 'Attention Is All You Need' and a core technical leader of Gemini announced he was joining OpenAI.

The next day, John Jumper, who co-won the 2024 Nobel Prize in Chemistry with Hassabis and contributed to AlphaFold’s development, announced he was leaving DeepMind after nine years to join Anthropic.

The news quickly impacted Alphabet’s stock price. On June 22, Alphabet closed down approximately 5%, its worst single-day performance in nearly a year. Days later, two more members of the Gemini team were reported to be following Jumper out the door.

By the end of June, all eight authors of 'Attention Is All You Need' had left Google—seven either founding new startups or joining competitors.

This talent movement drew significant market attention because these individuals were not ordinary engineers but key figures in the foundational architecture of modern large language models (LLMs) and AI research.

Third-party talent analytics also show that DeepMind engineers are about 11 times more likely to move to Anthropic than vice versa. While such data may have methodological limitations, talent flow remains a key signal of competitive dynamics in the AI industry.

Anthropic and OpenAI are both currently under intense capital market scrutiny. Anthropic raised $6.5 billion in May, reaching a valuation close to $96.5 billion. OpenAI has reportedly filed a confidential IPO application with the U.S. Securities and Exchange Commission (SEC).

Hassabis Moves to Chairman Role—Power Shift at DeepMind

Reports indicate that the August 5 leadership reshuffle directly affects DeepMind’s long-standing organizational structure.

According to three anonymous engineers cited by Fortune, this reorganization effectively represents a further shift of power from London to Google’s headquarters in Mountain View, California.

Day-to-day operations will now be led by DeepMind’s Chief Technology Officer, who also serves as Alphabet’s Chief AI Architect. He will be promoted to Senior Vice President and report directly to Google CEO Sundar Pichai.

To some employees, the lack of a new CEO appointment suggests DeepMind is being gradually absorbed into Google’s main operations, further eroding its former independence.

In 2023, following the launch of ChatGPT, Google merged Google Brain and DeepMind into Google DeepMind. Since then, more teams—including model training, machine learning infrastructure, and Gemini applications—have been consolidated under one umbrella, blurring the lines between DeepMind and its parent company.

Hassabis’s move to Chairman further underscores this organizational transformation.

Some engineers have questioned Hassabis’s involvement in Gemini’s daily operations, comparing him to Google co-founder Sergey Brin, who has recently re-engaged with AI work. Google has pushed back, stating Hassabis is based in London but frequently travels to Mountain View.

Workload and Military AI Controversy—Morale Issues Surface

Beyond leadership changes, DeepMind also faces internal challenges related to employee burnout and policy disputes.

Engineers have reported consistently working 60-hour weeks. Despite Google offering substantial retention bonuses and stock incentives, these measures have not fully stemmed the talent outflow.

Google’s AI contract with the U.S. Department of Defense has also sparked internal backlash. Under the agreement, Gemini models can be used on classified military networks for 'any lawful purpose.' Over 580 Google employees, including researchers, directors, and vice presidents, signed a public letter opposing the deal.

In May, around 1,000 employees at the London office pushed to form a union, with 98% of Communications Workers Union members voting in favor. This marks one of the first unionization attempts at a leading-edge AI lab globally.

Employee demands include halting the use of Google AI by the U.S. Department of Defense and the Israeli military, and reinstating the 'no development of weapons or surveillance AI' commitment that was withdrawn in February 2025.

Gemini Model Delays—Google’s Product Competitiveness in Question

Beyond talent issues, Google’s AI product timeline has also drawn market attention.

At the Google I/O developer conference on May 19, Pichai announced that Gemini 3.5 Pro would be publicly released 'next month.' However, the launch target was missed in June, June 30, and again on July 17.

Prediction market Polymarket ultimately settled with a 97% probability of 'not launched,' with trading volume nearing $230,000.

During this period, Google only released a few smaller-scale Flash models. Bloomberg reported that the delays have frustrated Google engineers, AI researchers, and management, with internal concerns growing that the company is falling behind Anthropic and OpenAI.

Multiple sources indicate that DeepMind ultimately decided to abandon its existing base model and restart with longer pre-training rather than rush an incomplete product to market.

One engineer noted that while Google was preoccupied with protecting search traffic, competitors had already shifted focus to AI agents and programming tools. Pichai himself admitted in May that Google was 'slightly behind' in AI agent programming.

Changes in model rankings further highlight the pressure Google faces.

Fortune cited the founder of an independent AI evaluation firm, who stated that Google once held the world’s most powerful model, but its top model now lags behind those of Anthropic, OpenAI, Alibaba’s (09988-HK) Qwen, xAI, and Meta (META-US).

As a result, a growing market sentiment suggests Google’s AI battle has shifted from 'how to win' to 'how to avoid losing.'

Google Still Holds the Strongest Cards in the AI Industry

However, Fortune argues that judging Google as having already lost the AI race based solely on model rankings may be premature.

After ChatGPT launched in late 2022, Google reportedly issued a 'code red' alert due to falling behind OpenAI. But by November 2025, following the release of Gemini 3, the market widely believed Google had closed the gap with Anthropic and OpenAI—so much so that OpenAI reportedly issued its own 'code red' in response to Google’s counteroffensive.

Analysts point out that Google’s biggest advantage lies in not being just a model company. From AI chips and data centers to models and applications, Google has full vertical integration. Its in-house TPU has evolved to the 8th generation, Iron