He Co-Invented The Technology Behind ChatGPT
Then Google Paid $2.7 Billion Just To Get Him Back

Noam Shazeer's Childhood: A Math Teacher's Son In Philadelphia

Noam Shazeer was born around 1976 in Philadelphia, Pennsylvania, into a family shaped by genuine resilience — his grandparents had survived the Holocaust. His father worked first as a math teacher before moving into engineering, while his mother stayed home to raise Noam and his siblings, in a household that placed heavy emphasis on education and creativity.

That environment nurtured an early passion for mathematics and computer science, which Shazeer went on to formally study at Duke University, graduating with a degree that would set him up for one of the most consequential engineering careers of the internet era.

Noam Shazeer Joins Google In 2000 — And Builds Its Spelling Correction System

Shazeer joined Google in 2000, at a company he'd actually overlooked at a job fair at Berkeley the year before, later deciding it seemed "like an interesting place to work." According to accounts of his hiring, he landed the job in part by building a working spell-corrector during an interview with Paul Buchheit, the engineer who would go on to create Gmail.

His early contributions at Google were foundational rather than flashy: he built a major upgrade to Google Search's spelling-correction system, and later developed the PHIL algorithm, which became a core component of Google AdSense — the advertising system that has generated a substantial share of Google's revenue for two decades.

Noam Shazeer Co-Invents The Transformer: "Attention Is All You Need"

In 2017, Shazeer co-authored a research paper with a deceptively simple title: "Attention Is All You Need." It introduced the Transformer architecture — a new way for neural networks to process sequential data using a mechanism called self-attention, rather than the older, more limited approaches that had come before.

The paper went on to become one of the most cited pieces of AI research in history. Shazeer had already laid some of the groundwork the year before, in 2016, with a Sparsely-Gated Mixture-of-Experts model, and he specifically helped design the Transformer's multi-head attention mechanism and residual architecture — the technical core that made the whole approach work at scale.

Meena: The Conversational AI Google Built, Then Wouldn't Ship

Shazeer's contributions continued well past the original Transformer paper. In 2018, he introduced Mesh-TensorFlow, a system for training much larger Transformer models, and in 2019 contributed to T5, an influential text-to-text model. He also played a central role in LaMDA, Google's large-scale conversational AI project.

In 2020, Shazeer worked with fellow researcher Daniel De Freitas on Meena, a conversational AI capable of human-level dialogue. Despite its technical promise, Google chose not to release Meena publicly, citing concerns about inappropriate responses — a decision that reportedly became a significant source of frustration for Shazeer, who felt the company's caution was holding back technology he believed was ready for the world.

Founding Character.AI: A Chatbot Platform That Predates ChatGPT

On November 1, 2021, Shazeer and De Freitas left Google to co-found Character.AI — a platform letting users create and converse with AI-driven characters, from impersonations of celebrities to entirely custom personas, spanning entertainment, education, and creative use cases.

The timing mattered enormously. Character.AI beat ChatGPT to market by roughly a year, giving it a genuine head start in consumer AI chat — and the platform's growth reflected that advantage. It reached a $1 billion valuation within about 18 months, and its user base swelled to somewhere between 20 and 30 million monthly active users, with reports describing teenagers in particular spending upwards of two hours a day in conversation with its AI characters.

Character.AI's Near-Miss Google Deal — And Its $150 Million Burn Rate

In November 2023, reports surfaced that Google was in talks to invest hundreds of millions of dollars in Character.AI at a $5 billion valuation. That specific deal never materialized. Meanwhile, Character.AI's own financial position was becoming more urgent — the company was reportedly burning through roughly $150 million a year, without a clear, sustainable path to profitability on its own.

Character.AI needed either significant new funding or a strategic partner. As it turned out, the deal that eventually arrived looked very different from a typical investment round.

Google's $2.7 Billion Deal To License Character.AI's Technology — And Bring Its Founders Back

In August 2024, Google announced an unusual arrangement: rather than acquiring Character.AI outright, it signed a non-exclusive licensing agreement for the startup's large language model technology, paying $2.7 billion in cash. Alphabet's own SEC filings later confirmed the exact figure, recording $2.7 billion in goodwill and $413 million in intangible assets tied to the transaction.

As part of the same deal, Shazeer and De Freitas — along with roughly 30 of Character.AI's researchers, mostly those focused on pre-training models — rejoined Google, specifically its Google DeepMind division. Existing Character.AI investors, including Andreessen Horowitz, had their shares bought out at a separate valuation of roughly $2.5 billion, reportedly delivering about 2.5 times their original investment. Character.AI itself remained an independent company, now fully employee-owned with no outside VC equity, refocused entirely on its consumer product rather than the broader pursuit of artificial general intelligence it had once talked about.

Why Regulators Took A Closer Look At The Google-Character.AI Deal

The unusual structure of the deal — licensing technology and hiring talent rather than buying the company outright — drew close attention from industry analysts and, eventually, regulators. Some observers described it as a "reverse acquihire": Google effectively removing one of its most credible consumer AI competitors from serious contention, without triggering the antitrust review a full acquisition would likely have faced.

In May 2025, the U.S. Department of Justice opened an antitrust investigation into the arrangement. Character.AI, for its part, continued operating independently — hiring a new Chief Product Officer, and later a permanent CEO, Karandeep Anand, a former Meta VP of Business Products and former President and Chief Product Officer at Brex — while narrowing its focus specifically to consumer-facing AI chat.

Noam Shazeer Returns To Google As VP Engineering And Co-Lead Of Gemini

Back at Google, Shazeer took on the role of Vice President of Engineering and Co-Lead of Google Gemini, the company's flagship AI initiative — overseeing the development of next-generation large language models and the scalable infrastructure behind them. In 2023, before rejoining, he had already received the WTF Innovators Award for contributions advancing emerging technology.

Noam Shazeer And Character.AI: The Full Timeline, Year By Year

Five Business Lessons From Noam Shazeer And The Character.AI Story

• Foundational technical work often looks unglamorous at first

  Shazeer's early Google contributions — spelling correction, an ad algorithm — weren't exciting on the surface, but they built the deep technical credibility that later let him shape the Transformer architecture.

• A big company's caution can be a startup's opportunity

  Google's decision not to release Meena over safety concerns left an opening that Shazeer filled directly by founding Character.AI — turning corporate hesitation into his own company's head start.

• Being early matters, even by a single year

  Character.AI's roughly one-year head start on ChatGPT gave it real time to build a devoted user base before the market became saturated with competitors.

• Running low on cash doesn't have to mean losing everything

  Character.AI's high burn rate could have ended in a distressed shutdown. Instead, careful positioning turned it into leverage for a deal that paid out investors, employees, and founders alike.

• Unconventional deal structures deserve scrutiny, not just celebration

  The "reverse acquihire" format let Google secure elite talent and neutralize a competitor without a full acquisition — a creative structure that also drew direct regulatory attention for potentially sidestepping normal antitrust review.

Frequently Asked Questions About Noam Shazeer And Character.AI