Alphabet Inc. Class C (GOOGL.US) AI faces "two extremes": cloud business surges by 82%, while the departure of top scientists raises concerns about cutting-edge research and development.

In the past two weeks, Google's "ice and fire" pattern in the field of artificial intelligence has been vividly displayed: on one hand, cloud business revenue surged by 82%; on the other hand, Chief Scientist Jeff Dean announced his departure after 27 years with the company. For this technology giant, which established the foundation for generative AI with the 2017 Transformer paper and now has a market value of $4 trillion, recent events highlight its core strategic dilemma: where to invest money? Building cutting-edge models requires substantial upfront investment in computing power and research and development, with uncertain future returns; meanwhile, the efficiency of cloud services has been proven, outpacing similar offerings from Amazon and Microsoft. Alphabet CEO Sundar Pichai stated in last month's earnings call that 90% of Fortune 100 companies are using Gemini Enterprise, showcasing Google's ability to sell AI services to corporate clients. Venture capital firm Theory Ventures founder Tomasz Tunguz pointed out that meeting the needs of most enterprises doesn't require top-tier models. "I think, especially in many white-collar work scenarios in the AI field, many models with acceptable performance are sufficient," Tunguz said. "Next-generation models will likely be used more in specific areas that require high-performance computers." Google's comprehensive AI strategy has been a major driver behind its 16% stock price increase this year (up 65% in 2025, outperforming all major tech competitors). However, recent trends have been somewhat rocky: after the latest earnings report, Alphabet's stock faced pressure due to market concerns over capital expenditures; on Wednesday, news of Dean's resignation and Demis Hassabis stepping down as CEO of Google DeepMind to become chairman further drove the stock price down. Despite a generally positive attitude on Wall Street, not everyone at Google is rejoicing. According to several unnamed insiders, some researchers are increasingly dissatisfied with the accessibility of computing powerthey struggle to obtain the computational resources necessary for advancing cutting-edge projects, while seeing Google sell its self-developed TPUs (tensor processing units, competing with NVIDIA's GPUs) to external clients, including Anthropic. Moreover, Google's cumbersome internal hierarchical approval processes make it difficult for research outcomes to be transformed into products, making teams like OpenAI, Anthropic, and even younger startups more attractive to AI researchersthey prefer lab work over financial figures. Dean's departure alongside Google veterans Sanjay Ghemawat, Oriol Vinyals, and Quoc Le has led to the establishment of Discovery Loop. Dean stated on the X platform that this Google-backed startup will position itself as a public-benefit corporation with the mission of automating machine learning, science, and engineering to accelerate discovery and progress. Several high-profile researchers have previously left, including Noam Shazeer, one of the authors of the landmark 2017 paper "Attention Is All You Need"the paper that laid the groundwork for generative AI, and now all eight authors have left Google. Shazeer joined OpenAI in June of this year, less than two years after Google brought him back with nearly $3 billion through "acqui-hiring." Shortly after he left, Nobel laureate John Jumper also departed DeepMind to join Anthropic. "Being a part of history" D.A. Davidson analyst Gil Luria pointed out that the trend of top talent leaving is apparent. "They are not eager to commercialize AI but want to be a part of history," Luria (who recommends holding Alphabet stock) stated. "So they see Anthropic, OpenAI, or other startups as places where they can write history." At Google, Dean was one of the few executives willing to openly criticize the Trump administration. Earlier this year, he strongly opposed the Pentagons decision to list Anthropic as a supply chain risk, warning that it could harm the overall interests of the U.S. AI industry. From a technical standpoint, he also built the foundational computing infrastructure and neural network systems that established Google's leading position in modern AI. Hassabis co-founded DeepMind in 2010 and sold it to Google four years later. He will now serve as chairman of that department and assume the newly established role of Chief Scientist at Alphabet, focusing on the long-term research and societal impacts of artificial general intelligence (AGI) while planning to devote more time to Isomorphic Labs, an AI drug discovery company spun out of DeepMind. The Chief Technology Officer of DeepMind and Alphabet's Chief AI Architect, Koray Kavukcuoglu, will take over the department's day-to-day management and the development of the next-generation Gemini model. According to individuals close to the DeepMind team, Kavukcuoglu has gradually assumed more responsibilities previously held by Hassabis over the past year, including guiding model development and overseeing major Gemini releases; meanwhile, Hassabis has spent more time away from the lab, focusing on regulation and the long-term implications of advanced AI. Morgan Stanley's internal insights: computing power allocation Google's investment scale in global data centers, chips, and related infrastructure is almost unmatched, yet computing power remains scarce. Each TPU allocated for training models, supporting Google products, or fulfilling cloud client contracts represents a choice among multiple priorities. Insiders stated that researchers' dissatisfaction with the accessibility of computing power becomes particularly pronounced when Google announces major infrastructure commitments to competing labs like Anthropic (whose models directly compete with Gemini). One such individual noted that Google has long-term forecasts for demand across various areas, including research, model training, search, Gemini products, and cloud partnerships, with these demands modeled years in advance. However, if a particular product's growth exceeds expectations or priorities shift, short-term adjustments to computing power allocation may occur. Pichai emphasized in the last two earnings calls that even with growing demand from cloud clients, Google still prioritizes meeting DeepMind's computing power needs. In July, when asked about TPU allocation, he stated that the "top priority" is ensuring computing power to maintain leadership in AGI frontiers, calling this work "the foundation of all our businesses." He also added that Google would balance the aforementioned needs with the computing power required for consumer products and AI models, and alleviate external demand by directly deploying TPUs in third-party data centers. Dan Niles, founder of Niles Investment Management and a Google shareholder, believes that computing power allocation presents a natural contradiction. "Google has all these other businesses, and they have to decide who to allocate resources to," Niles said. "In this situation, someone is always going to be dissatisfied." DeepMind accelerates integration with cloud business At the World Economic Forum in Davos this January, Hassabis and Google Cloud CEO Thomas Kurian shared the stage to discuss enterprise-level products and application scenarios. According to someone familiar with Google Cloud's operations, this was a rare instance; historically, the two organizations have long operated independently, and Hassabis has been particularly distant from other company operations. This individual believes that their joint appearance marks Hassabis's more active involvement in AI enterprise use cases (particularly in programming, customer service, and other fields), reflecting Google's overall strategy to accelerate the deep integration of research and business operations in the face of competitive pressure from OpenAI and Anthropic. In the following months, Google's models faced several setbacks, the most notable being the delay in launching its latest flagship model, Gemini 3.5 Pro. Meanwhile, Kurian's cloud business recorded record explosive growth. Kurian, who previously served as an executive at Oracle, has led Google Cloud since 2019, building an active enterprise sales organization within a company accustomed to consumer internet. Google Cloud designs its own AI chips, operates a global network of data centers, and sells models, databases, security software, and AI agent development tools. This strategy enables Google to profit from AI demand from multiple dimensions: selling infrastructure to labs like OpenAI and Anthropic while providing Gemini to enterprises and integrating AI into its own products such as search, YouTube, Workspace, and others. The generative AI boom has lasted nearly four years, and a new question emerges: does Google need to develop the very top-tier AI models itself? Or would it be wiser for other companies to bear the high costs while it reaps the benefits? Kavukcuoglu stated in May at Googles developer conference that the company simultaneously emphasizes efficiency while pushing the frontiers of AI. He noted that the Flash model provides cutting-edge capabilities while running four times faster and more efficiently than similar models, allowing Google to extend advanced AI to enterprise and consumer services. Similar to Tunguz, Niles believes that most commercial scenarios do not require the strongest models. "Existing models can meet 90% of regular demands," he said. "You don't need a Ferrari; a Ford will suffice." However, for scientists and researchers dedicated to achieving the next Transformer-level breakthrough, "sufficient" is often far from adequate.