Woofun AI reports that Google's AI division suffered a catastrophic structural collapse in late June 2024, marked by the simultaneous departure of its most critical scientific leadership. On June 20, 2024, John Jumper, the Nobel Prize in Chemistry laureate and lead researcher behind the AlphaFold project, officially announced his exit from Google DeepMind after nearly nine years to join Anthropic. Just days later, Jonas Adler and Alexander Pritzel, two pivotal members of the AlphaFold team, followed Jumper to the same competitor. Two days prior to Jumper's announcement, Noam Shazeer, a core author of the seminal Transformer paper and co-leader of the Gemini project, declared his move to OpenAI. This sequence of events confirmed that all eight original Google authors who contributed to the development of Transformer in 2017 had now completely left the company. Within a mere 72-hour window, two renowned scientific leaders and the core teams of two crucial projects departed in rapid succession. Behind these specific departures lay a broader, eight-year trend of significant AI talent loss at Google, ranging from the founders of Transformer to the leaders of AlphaGo and AlphaFold, and from the creators of BERT to the core development team of Gemini. The entity once described as the 'Huangpu Military Academy' of the global AI industry is now experiencing an unprecedented exodus of its top-tier intellectual capital.

The first and most visible sign of this talent crisis emerged from the field of AI life sciences, where John Jumper's departure carried profound implications for the entire sector. As the creator of some of DeepMind's most significant scientific achievements in the past decade, Jumper led a team that solved the long-standing problem of protein folding within just six months of graduating from university. His work resulted in the development of the AlphaFold, AlphaFold2, and AlphaFold3 models, which enabled the simultaneous analysis of over 200 million protein structures, fundamentally changing the research approach in structural biology. In 2024, John Jumper and Demis Hassabis, the head of DeepMind, jointly won the Nobel Prize in Chemistry, making Jumper the youngest Nobel laureate in chemistry in 70 years. For Google, Jumper represented the pinnacle of its efforts in AI for Science; for Anthropic, recruiting him was a calculated strategic move, as the company had already invested heavily in AI life sciences by acquiring biotech companies and developing specialized large-scale model products. Jumper's addition significantly enhanced Anthropic's capabilities in protein computation and drug development, instantly placing it among the top players in the industry. Demis Hassabis publicly acknowledged the value of their nine-year collaboration but expressed regret at losing such key talent. What was even more concerning was that these departures were not isolated individual actions but represented systematic transfers of entire teams, signaling a coordinated shift in the industry's power dynamics.

Jonas Adler and Alexander Pritzel, who followed Jumper, were not only core technical members of the AlphaFold project but also key contributors to the Gemini model, making their departure doubly damaging. Jonas Adler was a co-author of the AlphaFold2 paper and served as the main designer of the algorithm for predicting molecular interactions, a critical function for drug discovery. Alexander Pritzel was the third author of the AlphaFold2 paper and played a crucial role in the training architecture and optimization strategies of the Gemini model. Their departure meant that the core technical team behind AlphaFold had completely moved to Anthropic, not only transferring AI life science capabilities but also strengthening Anthropic's abilities in large-scale model engineering and code development. This migration effectively dismantled Google's technical barrier in one of its most prestigious scientific domains while simultaneously bolstering a direct competitor's infrastructure.

Just two days before Jumper's announcement, another significant departure occurred when Noam Shazeer joined OpenAI to focus on the research of next-generation model architectures. As one of the eight core authors of the Transformer paper and a pioneer in the MoE hybrid expert architecture, Noam Shazeer was one of the founders of Google's large-scale model technology. In 2024, Google spent $2.7 billion to acquire his team from Character.AI and appointed him as co-leader of the Gemini project.

However, less than two years after returning, he chose to leave again. Noam Shazeer's departure marked the complete departure of the original creators of Google's Transformer technology, leaving the company without any of the eight individuals who defined the modern era of deep learning. The events in June were merely the tip of the iceberg; over the past eight years, from Google Brain to Google DeepMind, Google has experienced a continuous loss of key talent in various areas, including basic models, AI for Science, large-scale model engineering, and product management. More than 20 core authors of milestone papers and several senior executives have left the company, creating a vacuum that competitors are eager to fill.

The erosion of Google's dominance begins with the departure of the Basic Model Founders, specifically the teams behind Transformer and BERT. Google is the birthplace of modern large-scale models, yet it failed to retain the people who defined this era. The eight authors who published the Transformer paper in 2017 have all left Google: Lukasz Kaiser joined OpenAI in 2021; Aidan Gomez, Ashish Vaswani, Niki Parmar, Jakob Uszkoreit, Llion Jones, and Illia Polosukhin started their own businesses; and Noam Shazeer's second departure dealt a final blow to Google's efforts in this area. The same trajectory applies to BERT. Jacob Devlin, the first author of the BERT paper, left Google in 2023 due to concerns about the compliance of Bard's training data and later returned briefly before leaving again. These individuals were the 'pioneers' of the large-scale model era, and their departures not only took with them their technical expertise but also the driving force for innovation at the fundamental level, leaving Google to rebuild its foundational research capabilities from scratch.

The disintegration of the AI for Science Teams represents a second critical failure, involving the collapse of two key projects that once built an unshakable technical barrier for DeepMind. In the field of protein folding, with the departure of John Jumper, Jonas Adler, and Alexander Pritzel, the core development team of AlphaFold has essentially disbanded, diluting Google's technical advantages in this area. In the field of reinforcement learning, David Silver, the 'father of AlphaGo,' left Google in January 2026 to focus on his company, Ineffable Intelligence, which develops reinforcement learning systems that do not rely on human data. The seed funding for Ineffable Intelligence amounted to $1.1 billion, setting a new record for European AI startups. Mustafa Suleyman, co-founder of DeepMind, left in 2022 to start his own business and later joined Microsoft as CEO of Microsoft AI, bringing with him Karén Simonyan, a key contributor to AlphaZero. From AlphaGo to AlphaFold, the core leaders of DeepMind's two most prestigious projects have either started their own companies or joined competing organizations, effectively neutralizing Google's historical lead in scientific AI applications.

In the competitive field of general large-scale models, the Gemini team has also suffered significant losses of top talent, further weakening Google's position. In the area of reasoning capabilities, Jason Wei, the co-founder of Chain-of-Thought, joined OpenAI in 2023 and became a key figure in the development of large-scale model reasoning. In the field of competitive reasoning, Dustin Tran, co-founder of Gemini DeepThink, and Ashish Kumar joined xAI in September 2025, helping to enhance the reasoning capabilities of Grok 4. In the areas of security and efficiency, Nicholas Carlini and Neil Houlsby joined Anthropic, weakening Google's technical strength in these areas. These departures indicate that Google is losing not just individual researchers but entire functional pillars required to build a state-of-the-art general-purpose model, forcing the company to rely on external hiring rather than internal organic growth.

Beyond technical talent, there have been significant losses in product and management positions, reflecting a mass departure at all levels of management. Mustafa Suleyman, co-founder of DeepMind, left Google taking with him his entire productization and commercialization team, stripping the company of its ability to translate research into market-ready products. Daniel De Freitas, former head of Google's LaMDA project, started his own company, Character.AI, and became a pioneer in the field of generative dialogue AI. From fundamental research to product development, Google's AI talent pipeline is undergoing significant weakening, creating a disconnect between its vast resources and its ability to execute on strategic initiatives.

The top talent leaving Google has not dispersed across the entire industry but has focused on four main destinations, reflecting the current competitive landscape of the AI industry. Google's main rival, OpenAI, is the largest recipient of Google's top talent, targeting the definition of fundamental architectural concepts and core capabilities. At the architectural level, OpenAI has acquired Lukasz Kaiser and Noam Shazeer, further strengthening its leadership in basic model architecture. At the capability level, it has recruited Jason Wei for reasoning and Jacob Devlin for pre-training and data systems. Each of these acquisitions directly targets Google's technical foundation, allowing OpenAI to maintain control over the evolution of large-scale models and ensuring that Google remains a follower rather than a leader in architectural innovation.

The fastest-growing rival, Anthropic, is systematically building a complete technical stack that can compete with Google. In the early stages, it recruited Niki Parmar to strengthen model architecture, Neil Houlsby to improve model efficiency, and Nicholas Carlini and Milad Nasr to enhance AI safety and privacy. By acquiring John Jumper's team, Anthropic quickly entered the AI life science field, establishing a dual-track approach of 'general large-scale models + vertical scientific computing.' This strategy has allowed Anthropic to rapidly climb to the top tier within five years, transforming from a niche player into a dominant force capable of challenging Google's legacy in both general intelligence and specialized scientific domains.

Multiple competitors, including Meta, xAI, and Microsoft, are focusing on their own weaknesses and acquiring talent accordingly to plug specific gaps in their portfolios. Meta targeted Llama's reasoning shortcomings by bringing away the core members of Gemini's math research team, enabling rapid improvements in Llama's reasoning capabilities. xAI is focusing on extreme reasoning, recruiting competition-level experts like Dustin Tran to build a strong foundation in this area. Microsoft has strengthened its product management and technical capabilities in consumer AI by acquiring Mustafa Suleyman's Inflection team. These targeted acquisitions demonstrate that the industry is moving away from broad, undifferentiated hiring toward precise, strategic talent acquisition designed to neutralize specific competitive advantages held by Google.

Startups have also emerged as a significant force in the global AI ecosystem, with many top talents choosing to start their own companies. The eight founders of Transformer formed a powerful group; Aidan Gomez's Cohere is now one of the three largest basic model companies in North America, valued at over $20 billion. Character.AI, founded by Noam Shazeer and Daniel De Freitas, was once a unicorn in the dialogue AI field. Adept, founded by Ashish Vaswani and Niki Parmar, is a pioneer in the field of AI agents. Jakob Uszkoreit's Inceptive is dedicated to developing RNA-based AI drugs. Llion Jones' Sakana AI in Japan is a leader in the local AI startup scene. NEAR Protocol, co-founded by Illia Polosukhin, is a leading player in Web3 infrastructure. Entrepreneurs affiliated with DeepMind, such as David Silver's Ineffable Intelligence, have also achieved significant success, attracting investment for their innovative technologies. These startups are not merely side projects but are becoming primary drivers of innovation, often outpacing the very giants that once employed their founders.

The reason why Google is struggling to retain top talent is not related to salary but rather to the fundamental needs of these researchers—control, stability, and a sense of value. These aspects are gradually being eroded within Google's organizational structure. The difference in autonomy is a primary driver; the hierarchical structure of large companies undermines the autonomy that startups provide. For top researchers capable of achieving groundbreaking results, what drives them is not rank or money but the freedom to pursue their ideas. At Google, even senior scientists often face complex bureaucratic procedures and restrictions on resource allocation and direction setting. Llion Jones has criticized Google's increasing bureaucracy, stating that it makes it difficult to achieve anything meaningful. In contrast, startups allow researchers to independently lead their projects, from defining objectives to allocating resources, providing a level of agency that is impossible within a corporate giant.

Strategic uncertainty and internal frictions further exacerbate the retention problem. The merger of Google Brain and DeepMind into Google DeepMind did not lead to synergies but instead caused internal conflicts, resource allocation issues, and strategic disagreements. The hasty launch of LaMDA and Bard, followed by negative feedback on Bard and strategic adjustments to Gemini, demonstrated that Google's large-scale model efforts were constantly shifting. Researchers had to adapt to new priorities, teams, and goals, lacking a stable research environment. For scientists focused on fundamental research, such strategic instability was far more detrimental than insufficient funding, as it prevented the long-term planning necessary for breakthrough discoveries.

Ineffective incentive systems also play a crucial role in the exodus. Money alone is not enough to retain top talent. Google spent $2.7 billion to acquire Noam Shazeer's team, demonstrating its willingness to invest heavily in talent.

However, cash and high salaries are no longer sufficient to retain these individuals. In startups, top talent receives equity, which is tied to the company's long-term success. The industry reputation and personal influence gained from successful projects are far more valuable than promotions and bonuses within large companies. The potential for massive financial upside through equity ownership in a successful startup often outweighs the security of a high salary at a corporation, driving researchers to take risks on their own ventures.

Differences in philosophies regarding safety, compliance, and commercialization are also common reasons for talent departures. Jacob Devlin left Google due to concerns about the compliance of Bard's training data, highlighting the conflict between academic rigor and commercial speed. Many researchers in the fields of AI safety and privacy have joined Anthropic, which emphasizes safety as a core principle. When large companies prioritize commercialization, scientists committed to long-term research often feel disconnected and choose to join companies that share their values. For example, Tencent's recruitment of Yao Shunyu illustrates this point. Yao Shunyu, introduced at the age of 27 as Tencent's youngest senior executive, was given significant organizational authority, allowing him to report directly to both Tencent's chairman and the president of its Technology Engineering Group, which is extremely rare within Tencent. This level of direct access and authority is something Google's bureaucratic structure struggles to provide.

The eight-year talent exodus at Google is having a profound impact on the global AI industry, reshaping the competitive landscape in fundamental ways. For Google, its technical advantages are being eroded. The loss of key talent in AI for Science, especially in AlphaFold, has weakened Google's technical position. The continuous departure of core researchers in general large-scale models has slowed down the development of Gemini, and Google's ability to lead innovation in this area is at risk. The company is now forced to compete against entities that possess the very architects who built its original moat.

For the industry, the competitive landscape is becoming more defined. OpenAI, Anthropic, and Google DeepMind are the three dominant players, but Anthropic is rapidly closing the gap, especially in AI life sciences. Many top-tier startups are also challenging the giants in specific areas, further diversifying the technological landscape. The concentration of talent in a few key competitors and successful startups suggests a future where innovation is driven by agile, focused teams rather than monolithic corporate structures. This marks a definitive shift in the power dynamics of the AI industry, where the ability to attract and retain top talent has become the single most critical determinant of success. The era of Google's undisputed dominance in AI research appears to be ending, replaced by a more fragmented and competitive ecosystem where the former leaders are now the primary targets of recruitment.