Anthropic Poaches Four Top Scholars in Two Weeks, From Nobel Laureate to Berkeley Department Chair
The AI talent war in Silicon Valley reached a fever pitch in the final two weeks of June 2026, and this time, the battle has spread to the most fundamental theoretical strongholds of academia. On the afternoon of July 1, a brief announcement sent shockwaves through academic circles: Jelani Nelson, chair of the Computer Science division in the Department of Electrical Engineering and Computer Sciences (EECS) at the University of California, Berkeley, and a professor of theoretical computer science, announced he had joined AI company Anthropic as a Member of Technical Staff and would be taking a leave of absence from the university. Nelson's statement on social media was remarkably restrained: "I've joined Anthropic and am on leave from the university. Excited to work alongside many talented, mission-driven people on the defining technology of our time." He revealed no further details about his specific role, team assignment, or research direction. Yet this did not stop the outside world from viewing it as a landmark event—the person leading one of America's top-tier computer science divisions had simply walked away. This was merely the fourth top-tier researcher Anthropic had recruited in the past two weeks. Less than two weeks before Nelson's announcement, John Jumper, the 2024 Nobel laureate in chemistry and core developer of AlphaFold, announced on June 19 that he was leaving DeepMind after nearly nine years to join Anthropic. Then, according to a Bloomberg report on June 24, Jumper's two core collaborators in protein structure prediction—Jonas Adler and Alexander Pritzel—would also follow him from Google's DeepMind to Anthropic. Both were key contributors to Gemini model pre-training and coding capabilities. In just two weeks, a Nobel laureate, two core Gemini researchers, and a sitting Berkeley department chair all flowed to Anthropic. The speed and depth of this talent migration are rare in the history of the AI industry. From engineering heavyweights to theoretical scholars, the logic of AI talent acquisition is shifting Over the past three years, AI companies' talent competition has primarily focused on engineering implementation, product deployment, and multimodal capabilities. Nelson's move signals that the competitive focus is now sinking to a deeper, more theoretical level. Nelson's academic pedigree is a textbook template for theoretical computer science. He completed his undergraduate, master's, and doctoral studies at MIT, earning his Ph.D. in computer science in 2011, with research focused on efficient algorithms for massive datasets. He subsequently conducted postdoctoral research at Berkeley, Princeton University, and the Institute for Advanced Study in Princeton, joined Harvard University as faculty in 2013, and moved to UC Berkeley in 2019. In the fall of 2024, he formally assumed the role of chair of the Computer Science division in EECS. His research centers on streaming algorithms, dimensionality reduction, and randomized algorithms. In layman's terms, he is dedicated to solving a core problem: when datasets are too large to fit entirely in memory, how can critical tasks be completed with minimal computational resources? Several years ago, his team tackled the fundamental problem of "approximate counting" and provided rigorous mathematical proofs defining the minimum memory any algorithm requires to perform such tasks. This kind of work—"defining the physical lower bounds for computation"—is the core mission of theoretical computer scientists. In terms of academic contributions, Nelson and his collaborators jointly proved the optimality of the Johnson-Lindenstrauss lemma, laying a theoretical cornerstone for the field of data dimensionality reduction; he also co-developed an asymptotically optimal algorithm for the count-distinct problem. These achievements earned him numerous honors, including a Sloan Research Fellowship and the U.S. Presidential Early Career Award for Scientists and Engineers. So, what does a theoretical scholar researching streaming algorithms mean for a large model company? The problem domain Nelson has plumbed for two decades—processing maximum data with minimum memory and computation—hits precisely the most acute pain point of the current AI industry. Training frontier models is essentially compression and filtering performed on astronomical-scale data streams; memory management and context window optimization on the inference side are battles fought against memory and computational complexity at every turn. As model scale collides with the ceiling of computing power and data, the value of "saving" begins to surpass that of "piling on." The underlying logic of AI competition is shifting from "whose model is stronger" to "whose algorithm is more efficient." Anthropic's signing of Nelson appears to be about driving the theoretical foundation one layer deeper, beyond models, engineering, and alignment. "Leave-of-absence hiring" goes mainstream as the academia-industry revolving door accelerates Nelson's move to Anthropic utilizes the "academic leave" model—his faculty position is retained, and he can return at any time. This system is already well-established in American academia. In 2017, Stanford professor Fei-Fei Li used her sabbatical to serve as Google's Vice President and Chief Scientist of AI/ML for Cloud, returning to campus two years later. Today, this channel between academia and industry is spinning faster and faster. For scholars, it's a ticket with a safety net, while industry can offer computing power, data, and real-world problems that academia struggles to match. For AI companies, signing a scholar means signing not just one person, but also their students, collaborators, and entire academic network. The traditional one-way path of "get tenure, work until retirement" is being replaced by a hybrid model of "keeping one foot in industry." And for universities, once this door is opened, it becomes very difficult to close. If the best theorists are all "on leave" at companies, what remains of academic institutions? This question is becoming a new source of anxiety for higher education. Google bleeds talent relentlessly, caught between brain drain and product breakthroughs Anthropic's aggressive poaching is only one corner of this talent war. Its direct competitor, Google, is experiencing a rare hemorrhage of talent. On June 18, Noam Shazeer, one of the eight core authors of the Transformer architecture and co-lead of the Gemini project, announced he was leaving Google to join OpenAI. Notably, Google had only re-hired him from Character.AI in 2024 in a deal valued at approximately $2.7 billion. Less than two years later, he was gone again. This was followed by the successive departures of Jumper, Adler, and Pritzel to Anthropic. These four individuals sit squarely on the most critical neural pathways of large models: architecture, science, coding, and pre-training. Capital markets reacted violently. According to a 36Kr report citing market data, Alphabet's stock price fell 5% to 6% over two days around June 22, wiping out hundreds of billions of dollars in market value. Investors are not worried about any single product, but about a more fundamental question: Can Google still retain the people who created its most important technologies? In the same month as these talent losses, Google launched its long-anticipated agent product, Gemini Spark. Unlike previous chatbots, Spark runs on dedicated virtual machines in Google Cloud, allowing it to continuously execute tasks in the cloud even when the user is offline. It deeply integrates with Gmail, Calendar, Docs, Sheets, Slides, Drive, as well as Maps and YouTube—Google's full suite of applications—enabling it to automatically complete multi-step tasks across apps. For example, it can automatically extract customer information from emails and populate tracking spreadsheets, or monitor school notifications and generate daily summaries. Google equipped Spark with an agent framework called Antigravity, supporting scheduled execution and conditional triggers. Through the MCP protocol, Spark can also connect to third-party services like Canva, OpenTable, and Instacart. However, the product is currently only available to Google AI Ultra subscribers paying $100 per month, a pricing strategy that has sparked widespread controversy. The emergence of Spark shows that Google has finally resolved to let AI step out of the chatbox and into the background to "do work for people." But industry observers note that Google holds the best productivity tool ecosystem in hand, yet due to the caution and fear of risk characteristic of large corporations, failed to play this card earliest. While it hesitated, open-source projects and startups were already redefining the boundaries of browser-based agents. The lure of equity ahead of expected IPOs Multiple sources indicate that OpenAI has confidentially filed IPO documents, and Anthropic is similarly pointed to by multiple sources as nearing a public listing. For top-tier researchers, joining now means pre-IPO equity incentives—a price that mature tech giants find difficult to match. From Nobel laureates to Berkeley department chairs, as the most elite theoretical scholars and engineering talent accelerate their migration into AI companies, these enterprises are, in effect, growing into a "second research institution system." The focus of the AI race has already sunk from model capabilities down to the foundational layer of algorithmic theory. And for the tech giants watching their core talent drain away, figuring out how to retain "the people who build the gods" is becoming a more urgent proposition than launching any new product.
Source: finance.biggo.com