Skip to main content
AI Talent WarsIndustry ShiftJun 25, 2026, 3:56 PM· 4 min read· in ai

Transformer Co-Author Noam Shazeer Leaves Google DeepMind for OpenAI in Major Talent Shift

Less than two years after Google paid $2.7 billion to bring him back, the foundational AI researcher is departing to lead architecture research at OpenAI.

By Logan Price

Industry Analysts 40%OpenAI Strategists 35%AI Research Community 25%
Industry Analysts
Focus on the staggering cost of the AI talent war and Google's retention challenges.
OpenAI Strategists
View the hire as a massive victory that could accelerate OpenAI's architectural breakthroughs.
AI Research Community
Emphasizes Shazeer's legacy and anticipates his next technical contributions.

The short answer

  1. Transformer co-inventor Noam Shazeer is leaving Google DeepMind to join OpenAI as Lead for Architecture Research.
  2. The move comes less than two years after Google paid $2.7 billion to license Character.AI and bring Shazeer back.
  3. Shazeer co-authored the seminal 2017 paper that introduced the architecture powering almost all modern AI models.
  4. OpenAI CEO Sam Altman celebrated the hire, stating he had wanted to work with Shazeer for a decade.
  5. The departure coincides with AlphaFold lead John Jumper also leaving Google DeepMind for Anthropic.

The architect behind the fundamental technology powering the generative AI boom is changing teams. Noam Shazeer, the co-inventor of the Transformer neural network architecture, has announced his departure from Google DeepMind to join OpenAI as Lead for Architecture Research.[1]

The move marks a seismic shift in the AI industry's ongoing talent wars. Shazeer, who most recently served as vice president of engineering and co-lead of Google's flagship Gemini models, shared the news on the social platform X, calling it a "difficult decision" but expressing excitement to work with OpenAI's "exceptional team."[1][2]

OpenAI CEO Sam Altman publicly celebrated the hire, stating that Shazeer is one of the people he has most wanted to work with since OpenAI was founded, adding that the recruitment "only took 10 years." The high-profile acquisition arrives as OpenAI prepares for its anticipated initial public offering, signaling to investors that the lab remains the premier destination for generational AI talent.[2]

For Google, the departure represents a staggering return on investment failure. Less than two years ago, in August 2024, the search giant paid a reported $2.7 billion to license technology from Character.AI—the consumer chatbot startup Shazeer co-founded—in a deal widely understood by industry analysts as an elaborate acqui-hire designed specifically to bring him back into the Google fold.[1][2]

Shazeer's career traces the arc of the modern artificial intelligence boom.

Shazeer's legacy in the field is virtually unmatched. In 2017, he was one of eight Google researchers who co-authored "Attention Is All You Need," a seminal paper that introduced the Transformer architecture. By allowing models to process sequential data in parallel and weigh the context of different words simultaneously, the Transformer replaced older recurrent neural networks and became the engine for almost all modern large language models, including ChatGPT, Claude, and Gemini.[1][3][4]

In 2017, he was one of eight Google researchers who co-authored "Attention Is All You Need," a seminal paper that introduced the Transformer architecture.

His career has been characterized by a push for rapid deployment and consumer-facing innovation. During his first two-decade stint at Google, Shazeer and colleague Daniel De Freitas developed a highly capable conversational chatbot named Meena. When Google, exercising caution over safety and reputational risks, declined to release the product to the public, the frustrated duo left the company in 2021.[2][4]

That departure led to the creation of Character.AI, which quickly achieved a $1 billion valuation by offering users the ability to converse with customized AI personas. The startup's massive popularity forced Google's hand, leading to the multi-billion-dollar 2024 licensing agreement that returned Shazeer to the company to help close the performance gap between Gemini and OpenAI's GPT-4.[1][2]

During his second tenure, Shazeer was instrumental in refining the Gemini architecture, working alongside Google veterans Jeff Dean and Oriol Vinyals. His engineering expertise, particularly in scaling Mixture-of-Experts (MoE) models, helped Google DeepMind stabilize its foundation models and deploy them across the company's vast product ecosystem.[4]

At OpenAI, Shazeer will lead research into the next generation of neural network architectures.

However, the corporate structure of a legacy tech giant often clashes with the agile, research-driven culture top AI scientists prefer. Industry analysts note that Shazeer's exit highlights a broader retention crisis at Google. In the same week as Shazeer's announcement, John Jumper—the Nobel-winning researcher who led the AlphaFold protein-prediction project—also announced his departure from Google DeepMind to join rival lab Anthropic.[3]

At OpenAI, Shazeer is expected to tackle the industry's most pressing technical bottlenecks. As AI labs push the limits of current scaling laws, researchers are increasingly looking beyond the standard Transformer to new architectures that can handle massive context windows and complex reasoning more efficiently. Shazeer's new role as Lead for Architecture Research places him at the vanguard of this effort.[4]

The 2017 Transformer architecture revolutionized AI by allowing models to process entire sequences of text simultaneously.

The financial mechanics of the AI talent war have reached unprecedented heights. With Meta reportedly offering $100 million signing bonuses to poach OpenAI staff, and Google spending billions on licensing deals that function as recruitment tools, the market for elite AI researchers has decoupled from traditional tech compensation structures.[1][2]

Ultimately, Shazeer's move underscores a fundamental truth of the generative AI era: while compute power and data are critical, the architectural breakthroughs required to reach artificial general intelligence still rely on a remarkably small pool of human visionaries. By securing one of the field's founding architects, OpenAI has fortified its technical leadership for the next phase of the AI race.[3]

Why it matters

The movement of foundational AI architects dictates which companies will build the next generation of frontier models. Shazeer's shift to OpenAI consolidates top-tier talent at the ChatGPT maker while highlighting the staggering difficulty legacy tech giants face in retaining the visionaries who invented the modern AI era.

Jargon, explained

Transformer Architecture
A type of neural network introduced in 2017 that processes sequential data in parallel, forming the foundation of modern large language models.
Acqui-hire
A corporate acquisition where the primary motivation is to recruit the target company's talent rather than its products or services.
Mixture-of-Experts (MoE)
An AI architecture technique where different parts of a neural network specialize in different tasks, improving efficiency and performance.
Large Language Model (LLM)
An AI system trained on vast amounts of text data to understand and generate human-like language.

Sources

Source coverage

4 outlets

3 viewpoints surfaced

Industry Analysts 40%OpenAI Strategists 35%AI Research Community 25%
  1. [1]Fast CompanyOpenAI Strategists

    Google AI leader Noam Shazeer leaves company for OpenAI

    Read on Fast Company
  2. [2]QuartzIndustry Analysts

    Google's top Gemini engineer is leaving for OpenAI less than two years after a $2.7 billion return

    Read on Quartz
  3. [3]MintAI Research Community

    Transformer Architect Noam Shazeer Leaves Google for OpenAI

    Read on Mint
  4. [4]WikipediaAI Research Community

    Noam Shazeer

    Read on Wikipedia

Comments

Stay informed

Every angle. Every day.

Get ai stories with full source coverage and perspective breakdowns delivered to your inbox.