On June 22, Alphabet's stock price closed with a significant drop of 5.08%, with an intraday maximum decline surpassing 7%. This downturn wiped out approximately $225 billion in market value in a single day, marking the largest single-day market value loss in the company's history. The catalyst for this dramatic fall seemed straightforward: within a mere three days, Google DeepMind had lost two top-tier AI experts in succession.

One of the departing experts was Noam Shazeer, a co-author of the seminal Transformer paper, the core architect behind the Mixture of Experts (MoE) approach, and the co-lead of the Gemini project. In 2024, Google had invested a staggering $2.7 billion to repurchase his stake from Character.AI, a company he had founded, only to see him leave again less than two years later. The other was John Jumper, the core leader of AlphaFold and the recipient of the 2024 Nobel Prize in Chemistry, who opted to join Anthropic after nine years at DeepMind. With their departures, all eight original authors of the Transformer paper have now exited Google.

(Left: Noam Shazeer, Right: John Jumper)

On the surface, it appeared as though the market had assigned a value exceeding $200 billion to these two scientists. However, Wall Street's true panic stemmed not from the individuals themselves but from fears that this brain drain would impede the technological advancement of Gemini. Media reports revealed that some members of the Gemini team were dissatisfied with the model's progress, noting that the gaps in programming, reasoning, and Agent capabilities with competitors like OpenAI and Anthropic had not narrowed as anticipated. As AI competition enters an era of rapid monthly iterations, Google finds itself under unprecedented pressure, which partly explains the successive departures of these two experts.

Over the past year, akin to other Silicon Valley giants, Google has aggressively invested in AI infrastructure, with capital expenditures reaching a staggering $100 billion. Wall Street's valuation models were predicated on a single premise: investments in computational power must be coupled with top talent to maintain a leading position in the AI race. However, with hardware funds spent but star researchers departing and algorithms stalling, this logic has collapsed, justifying the capital market's panic.

This situation raises a recurring question in the AI industry: How vital are AI experts to tech giants?

If they are not crucial, why did Google's market value plummet by $225 billion following the two departures? Conversely, if they are indispensable, why has Google, which has amassed the world's largest concentration of AI scientists in recent years, still failed to decisively outpace OpenAI and Anthropic?

The answer likely lies somewhere between these two extremes.

AI experts are undoubtedly significant. For large model companies, they represent not just technical prowess but also the ability to attract talent, possess engineering experience, and employ effective R&D methodologies. The arrival of a top researcher often brings not just an individual but an entire team capable of developing first-class models, along with a suite of technical paths and organizational expertise. Meta's multimillion-dollar compensation packages and Tencent's and Xiaomi's costly recruitment of star scientists all aim to acquire this scarce resource—spending big to fill critical gaps.

Without Luo Fuli, Xiaomi's large model team might never have taken shape, nor could its outputs have rapidly climbed the model rankings. Recruiting star researchers is essentially buying experience, a point Liang Wenfeng understands well. During his first external funding round, he explicitly demanded "anti-poaching" clauses in investment agreements, prohibiting investors from poaching employees from his company, DeepSeek. Liang's aversion to poaching stems not from DeepSeek's inability to function without specific individuals but from a harsh lesson: each researcher poached by a rival carries away a set of experiences—core assets.

However, overemphasizing "AI stars" in large firms comes with significant drawbacks.

Frequent appearances at forums, interviews, and the glare of publicity can conflict with the solitary focus required to solve technical challenges. A more insidious cost is the impact on team morale. By 2025, anonymous workplace forums at multiple tech giants revealed a clear trend: resentment over "stars winning awards in the spotlight while teams take the blame" is accelerating mid-level talent attrition. When all honors and resources flow to one individual, how do those executing the details and working tirelessly feel?

Then there's the escalation of human resource costs. Meta's talent war pushed AI industry salaries to staggering heights: Mark Zuckerberg personally called 24-year-old researcher David Ditek with an initial offer of $125 million over four years, doubling it to $250 million after rejection. To poach Apple's AI model team lead, Ruoming Pang, Meta offered over $200 million. Beyond reckless spending, there was a level of sincerity reminiscent of "three visits to the thatched cottage" (a reference to a famous Chinese tale of a ruler seeking talent). Zuckerberg's actions set a toxic industry precedent, with Sam Altman publicly complaining that Meta tried to poach OpenAI employees with $100 million signing bonuses. Star researchers' soaring salaries have driven up talent costs across the AI supply chain.

The risks of single-point dependency are also visible. If a star departs due to ideological clashes, better offers, or other reasons, the company faces immediate technical discontinuity. Google's plight is a vivid case, akin to MCN companies' reliance on a single superstar streamer—without Li Ziqi, Weinian loses its appeal. Could tech giants' "AI star dependency" unfold similarly?

Some large firms downplay AI heroes—at least publicly. Anthropic avoids packaging individuals as stars; OpenAI's Sam Altman functions more as CEO than technical spokesperson. Deepseek, Zhipu, Baidu, and ByteDance similarly avoid spotlighting researchers. These companies certainly have stars, yet their AI technologies continue to iterate, with Zhipu's valuation exceeding HK$1 trillion.

Two years of industry development have proven one thing: AI competition is no longer decided by individual heroism. Meta AI's struggles demonstrate that poaching stars alone cannot solve technical challenges.

Many frame AI as an epic of genius: a prodigy proposes the Transformer, a scientist creates AlphaFold, a researcher writes world-changing code. However, reality differs, especially today.

Large model R&D has evolved into one of the world's most complex engineering systems. From pre-training to post-training, reinforcement learning to inference optimization, data governance to infrastructure, every stage demands massive team collaboration. A genius can propose new directions but cannot train a GPT alone; a scientist can design algorithms but cannot deploy tens of thousands of GPUs single-handedly.

More critically, AI's development phase has shifted. The industry's greatest challenges now lie in productizing AI technologies, commercializing products, and implementing technical scenarios—all requiring cross-disciplinary collaboration.

Thus, Google's ordeal may prompt tech giants to reflect on and even guard against "AI heroism," shifting their focus from star researchers to strengthening systematic capabilities. The industry will re-examine: In the AI era, is the rarest resource genius—or the ability to organize genius?