Summary of Key Points
Recently, there has been a "seismic" exodus of talent in the field of Google's AI: Shazeer, one of the co-inventors of the Transformer architecture (acquired by Google for $2.7 billion less than two years later), has joined OpenAI; Jumper, the Nobel Prize winner and lead of AlphaFold, has taken half of his team to join Anthropic; along with other core researchers, at least five top talents have left in just one month. This has directly caused Alphabet's stock price to plummet by 5%, resulting in a market value loss of $225 billion (equivalent to the worth of Spotify). The root cause lies in the "big company ailments" inherent in Google's large organization—slow decision-making and internal resource consumption, which prevent talented individuals from realizing their potential. Ironically, it was Google's once-laissez-faire culture that gave rise to world-class achievements like AlphaFold. Today, Google is both a cradle for AI talent and a cage that holds them back.
Detailed Analysis
1. The "Collective Escape" of Top Talents: A Crack in Google's AI Defense
This departure is not a minor incident but a mass exodus of key personnel:
- Shazeer's Second Departure: As a co-inventor of Transformer (the technology behind GPT, Claude, and Gemini), he left to start his own business in 2021 after Google refused to release the chatbot Meena. In 2024, Google acquired him for $2.7 billion, but he left again for OpenAI due to project resources being reallocated to DeepMind.
- **The Nobel Prize Team's "Defection": Jumper, the winner of the 2024 Nobel Prize and lead of AlphaFold, took his longtime colleague Adler and Pritzel (key contributors to Gemini) to Anthropic, effectively transferring half of the AlphaFold team.
- A Chain Reaction: AI security researcher Conmy also left earlier. These individuals are pivotal in the field of AI; without Shazeer, there would be no modern large models, and without Jumper, there would be no revolution in protein structure prediction. Their departures have directly fueled market concerns about Google's AI competitiveness, as evidenced by the stock price drop equivalent to the value of Spotify.
2. The Truth About Inability to Retain Talent: Those Who Want to Work Are Stuck
Google isn't unable to retain talent; rather, it can't keep those who are motivated and eager to make a difference. The essence of Shazeer's two departures is that the organization didn't allow him to do so.
The problems stem from Google's "big company ailments":
- A Decision-Making Chain Longer than the Great Wall: Launching a new AI feature requires approval from multiple departments, including product, legal, compliance, and public relations, which can take months. By the time the technology is ready, the window of opportunity has passed (for example, Meena was launched two years before ChatGPT but was shelved by Google).
- Internal Frictions Despite Mergers: The 2023 merger of DeepMind and Google Brain aimed to combine strengths, but codebases and data flows remain disconnected, leading to inefficient resource allocation (e.g., Shazeer's resources being transferred to DeepMind).
- A Chain Reaction of Poor Products: Issues such as search AI suggestions that suggest using glue on pizza to prevent cheese from slipping, errors in Google Assistant after it was updated with Gemini features, and delays in the release of Gemini 3.5 Pro demonstrate that the issue is not with technology but with the organization's inability to turn talented ideas into useful products.
3. Google's Duality: The Same System That Nurtured Talent Also Traps It
Google's structure is contradictory:
- The Cradle Aspects: It allowed Jumper to work on AlphaFold for nine years without pressure for commercialization or budget cuts—a level of patience that startups cannot afford (Anthropic and OpenAI can only iterate every two weeks).
- The Cage Aspects: The same system has led to excessive bureaucracy, departmental interests, and compliance processes. When a company grows too large, the protections intended for innovation become barriers to it (the same environment that produced AlphaFold now confines those who want to bring ideas to market quickly).
4. The "Talent Cycle" in the AI Industry: Google Has Become a Training Ground for Its Rivals
History repeats itself:
- Bell Labs was once a hub of technology, but its talent spread to Silicon Valley.
- When Microsoft faced antitrust issues, its talents moved to Google.
- Now, Google has become a training ground for companies like Anthropic and OpenAI (SignalFire data shows that DeepMind engineers are 11 times more likely to join Anthropy).
Startups have a clear advantage: faster iteration speeds (every two weeks vs. Google's lengthy approval processes) + equity incentives before IPOs (Google, being already listed, offers less attractive equity). In essence, Google invests in talent only for it to be poached by its competitors.
5. A Lasting Dignity: Is It Time for Google to Let Go?
Facing this exodus, DeepMind's head Hassabis says that talent mobility is normal, but he understands the impact. Just like the old projectionist in "The Cinema Paradiso," Google once provided a paradise for AI researchers (free meals, a relaxed environment, top-notch colleagues), but its offerings have reached their limit.
New talents need to move on to become better versions of themselves, and perhaps Google should accept its role as a "nursery" rather than a hub for innovation. After all, every tech giant eventually becomes the cradle for the next generation.
Conclusion
Google's dilemma is not about lack of technology but about an organization that cannot keep up with change. It once fostered world-changing innovations with patience, but its current size has turned into a barrier to talent. This wave of departures may be a necessary "growth pain" for Google—either it must reform its systems or accept its role as a mere talent incubator in the AI industry. For those who have left, their next destination could be the place where the next revolution happens.