Google is sick

Author: Sleepy

In August 2024, Google spent $2.7 billion to buy back Noam Shazeer from his own company, Character.AI.

Shazeer is the core author of the 2017 paper "Attention Is All You Need" and a co-inventor of the Transformer architecture. Without his paper, there would be no GPT, no Claude, no Gemini, and no AI industry as we know it today. He joined Google in 2000, was one of its earliest employees, and stayed for over two decades. Later, because Google refused to release his chatbot Meena, he started his own business in 2021. Google spent a lot of money to bring him back, giving him the title of Vice President of Engineering and having him co-lead Gemini, hoping he could help Google win the AI ​​battle. Less than two years later, he left for OpenAI. According to Bloomberg, shortly before Shazeer announced his departure, computing resources for one of his projects were internally reallocated within Google to the DeepMind team. Sources said this adjustment was intended to improve team collaboration and integrate pre-training work. Shazeer left on June 18th. The following day, John Jumper also left. Jumper's story is different from Shazeer's. Shazeer was a veteran, having spent over two decades at Google, witnessing all the company's good and bad. Jumper, however, was nurtured by the company. Six months after graduating with his PhD, Hassabis made a risky decision, putting this young man with no management experience in charge of the entire protein structure prediction project. Jumper did not disappoint. He led his team to create AlphaFold, predicting the three-dimensional structures of over 200 million proteins, advancing the progress of structural biology research by a decade. In 2024, he and Hassabis stood together in Stockholm to receive the Nobel Prize in Chemistry. The first half of this story is about trust and fulfillment. Hassabis trusted a young man, and the young man repaid him with nine years of dedication, giving back to the entire field of biology. But the story has a second half. Two years after receiving the Nobel Prize, on June 19, 2026, Jumper posted a short tweet saying he was going to Anthropic. When the market opened on Monday, Alphabet's stock price plummeted. The stock price fell by about 7% at one point during the day and closed down about 5%, wiping out approximately $225 billion in market value—equivalent to the loss of Spotify. Alphabet's stock price had been declining since reaching an all-time high in early 2026, weighed down by antitrust lawsuits, exorbitant capital expenditures, and anxieties about AI competition for months. These two departures were the final straw. In the following days, news followed one after another. Jonas Adler and Alexander Pritzel were also leaving for Anthropic. These two were core contributors to Gemini and were also former collaborators with Jumper on AlphaFold. Adding to this the earlier departure of AI security researcher Arthur Conmy, Google lost at least five top researchers within a month, four of whom went to Anthropic. Hassabis nurtured Jumper from scratch, and now he's watching as half of the AlphaFold team walks into the door of their rivals. I don't know what he saw under Jumper's tweet, but I guess it was a familiar sense of fate. Every generation of the best tech companies eventually becomes a nursery for the next. Google itself grew up this way. Many of its earliest engineers came from Microsoft, IBM, Yahoo, and Bell Labs. When Microsoft was severely weakened by antitrust lawsuits in the 2000s, a large number of top talents flocked to Mountain View, including the young Shazeer. Going back further, Bell Labs invented the transistor, Unix, and the C programming language, practically laying the foundation for the entire information age. But what about Bell Labs itself? Its people scattered to every corner of Silicon Valley, becoming founding teams for other companies. Now it's Google's turn. In 2016, AlphaGo defeated Lee Sedol, and the world realized for the first time that AI could do something like that—that was Google's moment. In 2017, the Transformer paper was published, laying the foundation for the entire AI industry—again, that was Google's moment. In 2021, AlphaFold predicted 98% of the structure of human proteins; that was still Google's moment. Back then, no one asked, "Can Google win the AI ​​war?" because asking that question was as redundant as asking, "Will the sun rise in the east?" Google had the best researchers, the most data, the strongest computing power, and the most money—if it didn't win, who would? But look at who's standing against Google now. Ilya Sutskever, co-founder of OpenAI, worked on deep learning research with Geoffrey Hinton at Google early in her career. Anthropic's founders, the siblings Dario and Amodei, previously worked on security research at OpenAI, and many of OpenAI's early core teams came from Google. Jumper spent nine years at DeepMind, and Shazeer spent over twenty years at Google. The entire AI talent chain, tracing its origins, almost all have worked in Mountain View. SignalFire conducted a study in 2025 showing that DeepMind engineers were 11 times more likely to move to Anthropic than vice versa. Someone commented on this wave of resignations on Twitter, writing, "Google is becoming an Anthropic training ground." Google invests money, computing power, and a free environment to attract the world's brightest young people, giving them the best conditions for cutting-edge research. Once they've grown up, they fly away, go to the other side, create better products, and come back to beat you. Google's problem isn't just retaining talent. The moment it bought Shazeer back for $2.7 billion, it kept him. The question is, what happens after it's kept him? Shazeer left Google twice. The first time was in 2021, when Google refused to release his chatbot, Meena. ChatGPT hadn't even been created yet, and Google's attitude towards conversational AI was one of cautious observation. Shazeer couldn't wait and left. The second time is now; his computing power was reassigned, and he left again. Both departures were essentially because he wanted to do things, but the organization wouldn't allow it. Google's decision-making chain is too long. From R&D to launch, a new AI feature has to go through product, legal, compliance, public relations, and approvals from various business lines; any delay at any level can take months. By the time the technology developed in DeepMind's labs actually enters consumer products, the window of opportunity has already passed. In 2023, Google merged DeepMind and Google Brain, and everyone was optimistic about the merger of these two strongest AI teams. However, merging does not equal integration. The two teams' respective codebases, data flows, and work habits remain largely unchanged to this day. The transfer of Shazeer's computing power to the DeepMind team is a microcosm of this lack of true integration. Nominally, they are one department, but in reality, resource allocation and priority setting remain internal power struggles. When an organization fails to utilize its talent effectively, its products naturally deteriorate. Google Search's AI summary feature once suggested users put glue on pizzas to prevent cheese from slipping, claimed that running with scissors is aerobic exercise, and confidently replied, "No, it's 2025," when asked, "Is it 2026 now?" Studies show that it generates tens of millions of incorrect answers per hour. In early 2025, Google announced the complete migration of Google Assistant to Gemini. Its basic functions, used for nearly ten years, suddenly stopped working; setting alarms and controlling smart home devices all malfunctioned. The migration, originally scheduled for completion by the end of the year, had to be postponed to 2026. In July of the same year, Google's newly released Gemini CLI coding tool malfunctioned again. A user asked it to organize folders, and it hallucinated a series of non-existent operations, deleting all project files and then admitting, "I have completely and disastrously let you down." At the May 2026 I/O conference, Pichai confidently stated that Gemini 3.5 Pro would be released "next month," but it was later delayed until July. None of these were complex technical issues. Access control and feature regression testing are standard practices; a well-functioning engineering team wouldn't encounter these problems. A bad product and losing talent are two sides of the same coin; the organization has lost the ability to turn a genius's impulse into a product. The technology is still there, the people are still there, but the path from idea to launch is blocked. However, I think attributing this problem to "systemic issues" is too simplistic. What allowed Jumper to spend nine years polishing AlphaFold was precisely Google's system. They don't rush you to commercialize, they don't cut your budget, they don't ask when you'll produce results. This patience, this depth, is something no startup can offer. Anthropic and OpenAI might allow you to iterate every two weeks, but they can't let you spend nine years working on something whose success is uncertain. AlphaFold couldn't have been born in a place where weekly iterations are the norm. The problem is, with the same level of complexity, while it protects you in developing AlphaFold, it also accumulates layers of approval processes, departmental interests, and compliance procedures. While it gives you nine years of freedom, it also grows those twelve layers of backstabbing that prevent you from obtaining computing power. The soil that nurtures genius and the soil that traps genius are the same soil. This is something that almost no organization can escape once it grows to this size and achieves this level of success. Anthropic and OpenAI offer a place where ideas can be directly turned into action, plus pre-IPO equity. People leave not because Google isn't good to them, but because they've become the kind of capable, ambitious, but incapable people they least wanted to be. But who knows? Maybe twenty years from now, a young person at Anthropic will tweet that they're leaving for a company that's only three years old. If you don't go out and explore… On June 23, Hassabis was interviewed at the Lions Festival in Cannes and asked about his views on the recent talent drain. He said, "The flow of talent between major laboratories is normal. We have our own pool of top talent. We have the largest research team in all laboratories, covering the broadest range of research areas." Hassabis is one of the smartest people in the industry. He personally mentored Jumper from a recent PhD graduate to a Nobel laureate. He knows better than anyone what he has lost and why he couldn't retain him. So I think he wasn't just being stubborn when he said that. Perhaps this is someone who has seen the ending clearly, leaving themselves a final act of dignity. I'm reminded of what the old projectionist Alfredo says to young Toto in *Cinema Paradiso*: "If you don't go out and see the world, you'll think this is all there is." When Alfredo says this, he's pushing Toto away. He's more reluctant to part with this child than anyone else, but he knows that if Toto stays in this small-town cinema, he will never become the person he should be. The cinema gave Toto everything: a love of film, an understanding of light and shadow, and an initial curiosity about the world. But that's all the cinema can give. The rest of the journey can only be found by going out. Google was once the cinema of paradise for all AI researchers. The best equipment, the most relaxed environment, the most knowledgeable colleagues. You could spend nine years building a model to predict protein structure, without rushing to commercialize it, without writing PowerPoint presentations for executives. When you succeeded, the entire biological community applauded you, you stood in Stockholm to receive the award, and the whole company celebrated for you. Back then, everyone felt that Google was the world. But perhaps the best time for a place is precisely when it's time to let people go. The free cafeteria in Mountain View still serves three meals a day, and the colorful bicycles in the park are still parked in front of every building, open to anyone. Every week, a new batch of Nooglers puts on their signature propeller caps, takes a group photo, and their eyes shine. He's exactly the same as Shazeer when he first walked into Google twenty years ago, and Jumper when he joined DeepMind nine years ago.