Concerns are mounting that Alphabet (ASX: GOOG) is losing the battle for top-tier artificial intelligence talent.
On June 22nd, the company experienced its worst single-day stock performance in a year. Following the market open, shares of Google's parent company, Alphabet, tumbled sharply, at one point plunging over 7% to mark the largest intraday decline in nearly a year.
At its lowest point, the tech behemoth saw its market capitalization evaporate by more than $200 billion.
The market panic was not triggered by slowing performance, but by the departure of key personnel. Within a 48-hour window, the company lost two of its AI aces.
One is Noam Shazeer, a technical co-lead for the Gemini project, a co-author of the seminal Transformer research paper, and the founder of Character.AI, a company for which Alphabet paid approximately $2.7 billion to bring him back. He has now moved to OpenAI.
The other is John Jumper, a 2024 Nobel laureate in Chemistry and a Vice President & Engineering Fellow at DeepMind. He has departed for Anthropic.
Within the AI industry, these two individuals represent two of Alphabet's most critical technological moats: large language models and AI for Science. Their back-to-back departures within days have sparked concerns that the company is losing the war for elite AI talent.
The High-Priced Return That Didn't Last
Over the past year, the war for top AI talent has fundamentally reshaped Silicon Valley's power structure. Alphabet, the company that once defined AI, is becoming a significant source of talent outflow in this conflict.
Noam Shazeer is a prominent name in AI circles. In 2017, he co-authored the Transformer paper while at Google, helping ignite the large language model revolution. Today, nearly all major language models are built upon the Transformer architecture.
This shows Alphabet was not a latecomer to the AI race; it was there at the beginning. However, a pivotal turn began here. Even before ChatGPT's explosive debut, Shazeer and colleagues had developed an internal chatbot named Meena. Shazeer wrote an internal memo predicting such chatbots could replace Google Search and generate trillions in revenue.
Google executives reportedly blocked its release, citing safety risks and fairness concerns. What may have been caution for the company felt like a missed monumental opportunity to Shazeer. He left in 2021 to found Character.AI.
The story then took an ironic turn. In 2022, ChatGPT's launch made the entire industry realize chatbots were not fringe toys but the next-generation AI interface. What Google had kept in the lab became OpenAI's weapon to rewrite industry rules.
Consequently, Google sought Shazeer out again. In 2024, it struck a deal worth around $2.7 billion to license Character.AI's technology and bring Shazeer back. The entire Silicon Valley understood the real purchase was the person, not just the tech.
Upon his return, Shazeer was given a high-profile role as a technical co-lead for the Gemini project. Silicon Valley lore suggests he identified and fixed a critical bug that significantly improved training efficiency, helping Gemini surpass ChatGPT on some benchmarks. Some employees indicated Shazeer "saved" Gemini.
Yet, even the heavily incentivized return did not secure his long-term stay. In June 2026, Shazeer left Alphabet again, announcing on X his move to OpenAI. OpenAI's CEO Sam Altman promptly replied, calling Shazeer one of the people he most wanted to work with since OpenAI's founding.
An AI commentator noted this was "the most significant AI talent move of the year," raising questions about what is happening internally at Google. Shazeer's exit led the market to question whether Alphabet is still missing out on top talent.
The Slide from the Top Spot
Two days later, on June 20th, John Jumper also announced his departure on X. Unlike Shazeer, Jumper had spent nine years at DeepMind. In 2024, he won the Nobel Prize in Chemistry for leading the AlphaFold project, which predicted over 200 million protein structures.
In his post, he thanked DeepMind CEO Demis Hassabis for giving him the chance to lead the entire AlphaFold team just six months after earning his PhD. This statement served as a reminder of how attractive DeepMind once was, where a young researcher could be entrusted with a mission to change scientific history.
Now, Jumper, a soul of DeepMind, is leaving for Anthropic. Following his departure, media reports revealed a mood of "extreme frustration and widespread dissatisfaction" within DeepMind. Employees reportedly feel the once world-leading AI lab has slipped to an "awkward third or even fourth place" in the industry.
One internal employee posed a pointed question: "In text, image, video, speech, and even vision, we no longer have a single model at the industry frontier. If, with so many resources and after more than four months of effort, we can't produce a true frontrunner model, what are we even doing?"
Resource allocation is another sore point. Since the merger of Google Brain and DeepMind in April 2023, the two teams and cultures have failed to integrate fully. Competition for computing resources between internal teams has reportedly become intense.
In October 2025, Google reportedly allocated precious Google Cloud TPU compute to a major client and direct competitor, Anthropic, further fueling external doubts about its internal prioritization. A sentiment expressed earlier by Llion Jones, another Transformer co-author, seems prophetic: "I feel like Google's bureaucracy has grown to a point where I can't get anything done."
Where is the Talent Going?
The departures of Shazeer and Jumper mean that years of accumulated, undisclosed technical secrets and training intuition are materially transferring to OpenAI and Anthropic. An industry insider warned, "You can lock model weights in a data center, but the people who built them take with them tacit knowledge, training intuition, safety trade-offs, architectural patterns, and experience avoiding pitfalls."
Despite awareness of the problem, talent continues to flow out at a startling rate. A 2025 Talent State Report by SignalFire indicated the ratio of DeepMind talent flowing to Anthropic versus the reverse was 10.8 to 1. The probability of an engineer moving from OpenAI to Anthropic was over 8 times the reverse, and nearly 11 times from DeepMind.
The report also noted two-year employee retention rates: approximately 80% at Anthropic, 78% at DeepMind, 67% at OpenAI, and 64% at Meta. A SignalFire partner stated that when asking candidates about their dream company to join, Anthropic was mentioned more than any other.
The 2026 talent movement timeline is dense. Key figures from OpenAI's safety, research, and pre-training teams, including co-founder Andrej Karpathy, joined Anthropic in the first half of the year. Jumper's arrival signifies Anthropic is now absorbing the core scientific prestige from Google DeepMind.
All signs point to Anthropic becoming a talent magnet, yet intriguingly, it reportedly does not offer the highest market salaries. Founder Dario Amodei has suggested that mission alignment, not just money, is key. Many who moved from Google to Anthropic cite "focus" as a major factor.
At Google, models must serve a vast commercial ecosystem—Gemini chases GPT, models serve Search, capabilities enter Cloud, and products must cater to Workspace, Android, advertising, and developer tools. This creates a paradox: Google has the most resources, but researchers may lack focus. At Anthropic, the central question is simply: can the next model be more powerful?
Meanwhile, OpenAI is also absorbing Google talent. Shazeer's move is the most symbolic, with a Transformer pioneer joining OpenAI's architecture research team to lead exploration of next-generation model architectures.
Both OpenAI and Anthropic are preparing for IPOs, making the timing of these high-profile hires non-coincidental, as employee equity stands to be realized in public markets. With both rivals challenging Google on talent and capital markets simultaneously, Alphabet investors have reason for concern.
What Truly Retains Talent?
However, Google is far from the only company facing talent raids. The pursuit of top AI talent has reached a nearly frenzied stage industry-wide. A simple truth is understood: the value of a top researcher is more irreplaceable than any chip in a data center. Such brain drain essentially transfers an entire technical roadmap to competitors.
A no-holds-barred talent war has erupted, with compensation packages shattering ceilings. In the summer of 2025, Meta's Superintelligence Labs reportedly hired several core OpenAI researchers, with one package rumored to be worth $1.5 billion over six years.
Research indicates the top 1% of AI researchers command an average salary of about $1.94 million, significantly higher than academic peers. At the very pinnacle, packages from companies like Google DeepMind can reportedly reach $20 million annually.
The trend is mirrored in other regions, with reported monthly salaries for top AI scientists reaching exceptionally high levels and recruitment packages for key figures estimated in the hundreds of millions.
Headhunters note that budgets at major firms in 2026 seem to have "almost no upper limit." Yet, exorbitant pay is only one side of the story. The other side involves CEOs personally recruiting talent—from Mark Zuckerberg reportedly writing code and delivering soup to potential hires, to Satya Nadella personally calling candidates and approving exceptional offers.
However, when all top players can offer sky-high compensation and CEOs are personally involved, a paradox emerges: money suddenly becomes less effective. There have been instances of new hires returning to former employers shortly after joining a rival, and high-profile departures from newly formed labs.
A report examining these cases identified three core variables for retaining top researchers: mission alignment, compute resource supply, and organizational stability. It concluded that "when all leading players can offer top dollar, the offer itself ceases to be a differentiator."
This encapsulates Google's dilemma. The departures of Shazeer and Jumper are less a failure of compensation and more a loss of mission. Money can open the door, but it cannot keep people at the table.
The sharp drop in Alphabet's stock price on June 22nd serves as a stark lesson—the loss of over $200 billion in market value in a single day far exceeds the cost of any compensation package. More importantly, capital markets are beginning to recognize that AI talent is a signal for technological direction, product speed, and future valuation.
Thus, the AI war returns to its most fundamental question: who can retain the few individuals who truly know how to build the next generation of models? Alphabet once had the most such talent. Now, it is watching them leave for OpenAI, Anthropic, and other AI labs.
Comments