May 17th news, Sundar Pichai, CEO of Google and its parent company Alphabet, was a guest on the "All-In" podcast and was interviewed by entrepreneur and investor David Friedberg.
In this interview, Pichai discussed how Google is actively disrupting itself amidst the wave of artificial intelligence to maintain technological leadership.
With AI fundamentally changing the way information is accessed, the question, "Will search be replaced by AI?" has become increasingly important. Pichai stated that Google is redefining the search experience, transitioning from simply responding to user queries to becoming an intelligent assistant that "follows the user." This means products will be more predictive, personalized, and even provide relevant information before the user inputs a question.
Regarding the technological foundation, Pichai emphasized that Google's long-term investment in infrastructure has built a unique advantage. Self-developed TPU chips, large-scale data centers, and mature distributed systems provide solid support for Google's AI model training and deployment. Coupled with a differentiated path in foundation models, Google is expected to maintain leadership in the generative AI competition.
Future human-computer interaction is also a core topic.
Pichai pointed out that AI will not be limited to the software level but will profoundly change the way humans interact with technology. Voice, image, multimodal input, etc., are reshaping hardware forms and product interfaces, and the competitive landscape of the entire industry is evolving accordingly.
However, this technological revolution has also brought challenges in terms of energy. The expansion of AI models consumes a huge amount of energy. Balancing performance and sustainability is a real problem Google must face.
Meanwhile, Google's continuous exploration in quantum computing and robotics also demonstrates its strategic vision for laying out future computing platforms.
Addressing external questions about Alphabet's strategic positioning, Pichai also responded to the question of whether they are still looking for the "next hundred-billion-dollar business." Pichai emphasized that this is not just a strategic question about business diversification, but a fundamental proposition about how to continue Google's spirit of innovation and industry leadership in a new technological cycle dominated by AI.
Below are the highlights of Pichai's interview:
01 Will AI End Traditional Search? Google Embarks on a Path of Self-Disruption
Q: You've been Google's CEO for nearly ten years. During this time, the company's market value surged from hundreds of billions to two trillion dollars, and quarterly revenue grew from 20 billion to nearly 100 billion. Do you like this job?
Pichai: I love building products. Google itself is a deeply computation-oriented company, dedicated to translating core technologies into products that impact daily life. I still participate in product-related work every week, which is the part I enjoy the most.
Of course, as the CEO of a company that affects billions of people globally, this job also means immense responsibility.
Q: Is Google facing the risk of being disrupted by AI? AI is creating a completely new paradigm of human-computer interaction. Users interact with AI through a chat interface, asking questions and getting complete answers, which is completely different from the traditional search experience. Google's core search business accounts for about 200 billion of the company's 360 billion annual revenue and is its main source of profit. Google is currently in a delicate position: disrupting itself too quickly might harm existing revenue; but acting too slowly might lead to being surpassed by competitors. How do you view this situation? Will Google be disrupted by AI, or is it leading this transformation?
Pichai: In fact, as early as ten years ago, I began positioning Google as an "AI-first" company. We launched the Google Brain project in 2012 and acquired DeepMind in 2014. When I took over as CEO in 2015, I clearly believed that the development of AI would be the core force driving search evolution.
Even in the past two years, I still believe that AI brings unprecedented opportunities to search. Transformer-based models like BERT and MUM have significantly improved search understanding and result quality.
We launched "AI Overviews" about a year ago, which now cover over 150 countries and regions globally and have over 1.5 billion users. AI Overviews have spurred a large number of new forms of queries, and their number is still growing.
Based on user feedback, we are testing a new AI-driven search experience called "AI Mode," where users can get a full AI interaction experience in search, including subsequent conversational queries. We are integrating the most cutting-edge AI models into it, which truly use search as a native tool. Compared to traditional search, queries in AI Mode are often longer, with an average length two to three times that of previous queries.
Q: Regarding the statement "Google is dead," similar arguments have actually appeared many times before. This time, the focus of the debate has shifted to AI's reshaping of the search experience. Many people are comparing the independent Gemini App with ChatGPT and Meta AI side-by-side. Data shows that as of March, Gemini App had 350 million monthly active users, while ChatGPT had 600 million and Meta AI had 500 million. Is this method of comparison itself flawed? Does the independent Gemini App truly represent the core of Google's AI strategy?
Pichai: Indeed, but such comparisons are not entirely accurate. The Gemini App is part of our AI strategy, but by no means the entirety. In fact, the most widely used generative AI application globally today is likely "AI Overviews" embedded in search.
We did launch an independent Gemini app and significantly increased user engagement and usage frequency after releasing Gemini 2.5 Pro. Recently, we have successively launched Deep Research, an upgraded Canvas, Audio Overviews, and users can now also generate videos in the Gemini App. On Android devices, the Gemini Live feature supports screen sharing and real-time voice interaction, further expanding interaction boundaries.
The rapid iteration of these features has received positive user feedback. Of course, ChatGPT is a very successful product, but I believe we are still in the early stages of AI technology development. As long as we continue to launch innovative products and win user recognition, it shows we are heading in the right direction. The key is whether innovation truly changes user behavior. From the current stage, this transformation is happening.
Competition is extremely fierce currently, but we have already seen users actively obtaining information using the Gemini model in multiple product scenarios such as Search, YouTube, and the Gemini App. This means we are building a broader AI product ecosystem.
Q: From Google's business model perspective, every search query involves service costs and also generates advertising revenue. Now, search is transitioning to an AI-driven interface, and the computational cost of AI queries is significantly higher than traditional search. How do you view the evolution of this economic model?
Pichai: In my view, as long as the core goal remains "serving user queries efficiently," our advantages at the infrastructure level allow us to do it better than anyone else.
In fact, over the past 18 months, we have significantly reduced the computational cost per query. The bigger technical challenge now is not cost, but response latency. The standard for search has always been "instant response," so latency is more critical than cost.
As for the business value of AI queries, we have already achieved advertising revenue levels comparable to traditional search in AI Overviews. This is an important milestone, indicating that AI search has a sustainable business model. Furthermore, we believe there is room for further improvement.
Q: Do you feel pressure from Wall Street and the board? How much freedom do they give you to do what you think is necessary?
Pichai: I feel like this is an "acceleration moment" such that there's scarce time even to think about pressure. My attention is currently focused more on a few key questions: Are our models leading enough? Are we continuously at the forefront of technology? In the past few months, we have demonstrated the breadth and depth in the AI field, and this momentum must be maintained. For me, execution is the core task right now.
Since becoming CEO in 2015, the first thing I did was to explicitly position Google as an "AI-first" company. We already have excellent products, such as YouTube, Workspace, and Google Cloud, and I'm focused on how to build these into strong, sustainable businesses.
Today, we have initially achieved this goal. Last year, the combined revenue of YouTube and Google Cloud reached 110 billion dollars. Many people might not realize that Google is now not only one of the largest media companies globally but also one of the largest enterprise software providers. Even in the podcasting field, we may already be the largest podcast platform globally.
So I believe Google is well-prepared for the arrival of the AI era. This is the first time a new technology has emerged that can span all core businesses and empower the entire chain. AI is a representative of this "all-scenario technology platform." It can reshape search, reshape YouTube, enhance Cloud, and even spawn new business forms.
For me, AI is not just a strategic turning point for Google, but possibly one of the most exciting opportunities in the next decade.
02 Google's Core Competitive Advantages: Infrastructure + Model Efficiency
Q: Google's core advantage lies in its infrastructure capability, but the outside world may not fully appreciate its importance. In today's AI competitive landscape, where specifically does Google's infrastructure advantage manifest? What areas are your planned 70 billion dollars in capital expenditures this year mainly focused on?
Pichai: Google's infrastructure has always been on the "Pareto Frontier" of performance and cost efficiency (Pareto Frontier, a core concept in economics and decision science, describes the optimal balance that can be achieved between different decision schemes with limited resources). That is to say, we can provide the industry's most advanced models in the most cost-effective way. For example, the Flash series models have become industry mainstays, which is inseparable from our infrastructure system built on self-developed TPUs (Tensor Processing Units).
Since the release of the first generation TPU in 2017, we have been continuously iterating and optimizing, and we are now on the seventh generation. I remember when I first introduced TPUs at Google I/O back then, the outside world didn't quite understand why we wanted to develop our own AI chips. But time has proven the value of this decision: the reason we can support large-scale, low-latency AI services today is precisely because of the excellent performance of these chips on inference tasks.
Google's infrastructure layout is full-stack: from the bottom-layer submarine cables to server architecture, custom chips, and up to upper-layer software and AI frameworks, we have achieved end-to-end autonomy. This not only improves overall efficiency but also allows us to provide leading AI services at a lower cost. For example, the Gemini 2.5 series models can offer services at a very attractive price, not only because the models themselves are excellent but also because of our continuous optimization of the infrastructure. This vertical integration brings tangible cost and performance advantages.
Regarding the 75 billion dollars in capital expenditures planned for 2025, it will primarily be used for servers and data centers, with servers accounting for the largest portion. About half of this will be used for Google Cloud, supporting a range of AI infrastructure and model services for enterprises.
At the same time, we will continue to increase investment in Google DeepMind, including not only large language models but also image, video, multimodal models, and even emerging research directions like "world models." These technologies not only serve Search, YouTube, and Gemini but also expand our research depth and frontier exploration.
Q: Regarding chips, the outside world generally believes that NVIDIA has a near monopoly on the AI market. Can Google's self-developed TPUs completely replace NVIDIA GPUs?
Pichai: We have a very close collaboration with NVIDIA. A significant amount of Gemini's inference tasks still run on GPUs, and we also provide customers with multiple hardware options. However, internally, Google primarily uses TPUs for training and deploying Gemini models. We didn't choose an "either/or" path, but rather a dual-track approach: continuing to use NVIDIA GPUs on one hand, and firmly advancing the TPU path on the other. I believe this flexibility itself is an advantage.
Q: Many people believe that the performance of Large Language Models (LLMs) is gradually entering a plateau phase, and the differences between various companies are narrowing. How do you see this issue? How much room for evolution is there left on the LLM path? How can Google maintain long-term leadership?
Pichai: I want to say that the development of AI is never linear, but "jagged" – this is what Andrej Karpathy calls "Jagged Intelligence." That is to say, progress in AI is often intermittent, experiencing periods of stagnation followed by paradigm breakthroughs.
In the past few years, the AI industry has gone through stages from large-scale pre-training and fine-tuning to efficient inference. Now, we are entering a new stage: integrating these model components into more complex and functional agent systems. Frankly, progress has indeed become harder, but this is precisely the stage that teams like ours, who are good at long-term fundamental research, are best equipped to handle. We are not only focusing on LLM or Transformer architecture but are also actively exploring other paths like Diffusion Models.
Currently, our investment in computing power is still yielding results, and we have not encountered the critical point where "investment is no longer effective." Challenges are more at the execution level, such as how to find enough electricians to build data centers, or whether infrastructure construction can keep up with model iteration. But from a research perspective, our technical path remains clear, and the prospects are broad.
Q: Does Google have an advantage in data acquisition because it owns products like YouTube, Search, and Docs? Are you able to train models in a way that other companies cannot replicate?
Pichai: We do have a unique advantage in creating differentiated user experiences. Google has built long-term relationships with users through services like Gmail, Docs, Calendar, YouTube, and Search. With explicit user authorization, we have the capability to integrate this personalized context into AI systems to provide users with a more natural and contextually relevant experience. This is a strategic direction that we highly value and are steadily advancing. We hope to use this approach to create AI assistants that are far superior to existing products.
03 The Future of Human-Computer Interaction: Voice, AR, and Next-Generation Hardware
Q: How do you see the relationship between humans and computing evolving in the next 5 to 10 years? Will we still use keyboards and operate on screens? Or will interaction change completely?
Pichai: For the past few decades, humans have been adapting to computers. In the future, this trend will completely reverse: computers will actively adapt to humans.
Imagine a truly natural device – perhaps AR glasses – that can sense your needs and proactively provide help without you issuing explicit commands. I wear glasses myself, so I pay close attention to the progress of AR devices. While current devices' comfort and functionality are not yet ideal, progress is very rapid.
The integration of voice and multimodal input (image, semantics, audio) is making human-computer interaction increasingly natural. When AR devices truly mature, it might usher in an "iPhone moment," which is the birth of the next generation of mainstream computing platforms. From the trend of technological evolution, that point is getting closer and closer, which also fills me with anticipation for the future.
Q: Are you also investing a lot of effort in hardware?
Pichai: Yes, hardware is an important direction for our investment. We pay high attention to next-generation computing platforms, such as AR glasses and robotics technology. At the device level, we continue to develop Pixel series products; in terms of infrastructure, we are continuously expanding data centers. And projects like Waymo, our self-driving project, can also be seen as our exploration on the "robotics" path. Overall, Google is gradually delving into various levels of the physical world.
Q: Some of your competitors are also very active currently: OpenAI has Altman, xAI has Musk, Meta has Zuckerberg, and Microsoft has Nadella. How do you view these individuals and their companies?
Pichai: They are all respected entrepreneurs, representing the most influential tech forces globally. They have put tremendous effort into pushing the technological frontier. I have maintained contact with most of them – for example, I just met with Musk two weeks ago. He has a very unique ability to turn vision into reality with strong willpower, which is truly admirable.
Of course, we have both cooperation and competition among us. But at a more macro level, I'm more focused on: Can technology truly benefit humanity? For example, in education, healthcare, and other fields, the potential brought by AI is generational. This is not just a competition between companies, but a shared mission to drive human progress. The AI field holds immense opportunities, and I believe it's not just a "winner takes all" situation; every company that does good work has the potential to succeed.
Q: So you don't see this as a "kill or be killed" competition?
Pichai: Not at all. I think this is a transformation far greater than any single company or a single wave of technology. A company you haven't heard of today might become a big winner in the AI field in the future. I have always believed that AI is a platform-level technological revolution, perhaps even exceeding the sum of all technology revolutions in history. Companies that can stand out in this field are those that can constantly innovate, execute strongly, and attract and retain the world's best talent.
04 China's Competitiveness at the AI Frontier Should Not Be Ignored
Q: What do you think of the recently emerging Chinese AI company DeepSeek?
Pichai: DeepSeek is a "cognitive refresh" moment. Anyone who seriously follows AI papers will not underestimate China's capabilities. China has a huge output of research, and its talent is excellent. The release of DeepSeek made the outside world realize that China is actually closer to the global technological frontier than many people expected.
We also conducted benchmark tests on the DeepSeek model internally and compared it with our Flash model. The test results showed that Flash performed excellently across multiple dimensions, and in some aspects, it even had advantages.
It's worth mentioning that DeepSeek's efficiency partly benefits from the reality of limited hardware resources, which has prompted them to make very creative attempts in model structure and performance optimization. To some extent, this also confirms the direction of efficient architecture that we have been pursuing. Overall, this reminds us that global AI competitors are not only numerous but also strong. China's competitiveness at the AI frontier should not be ignored, and I have always firmly believed this.
05 AI's Energy Bottleneck: Electricity Determines Technological Boundaries
Q: Many people believe that the core bottleneck for future AI deployment will be "electricity." For example, Musk recently mentioned that he will need 1 terawatt of computing power in the future, which is almost equivalent to the total current power generation in the United States. And China might reach 8 terawatts by 2040. Does this mean that whoever controls energy resources will have a greater competitive advantage in AI? What is the gap between the US and China in this regard? And where does Google stand?
Pichai: I completely agree that electricity is a key "bottleneck" on the path of AI development. This not only affects a single company or industry but can also profoundly impact the future global economic and GDP growth patterns. Frankly, I am very concerned about this myself. But I don't believe it's an unsolvable "physical limit." Technically speaking, we already have pathways to meet power demands. This is more like an "execution" problem: Can we build, deploy, and promote energy transition quickly?
I don't think we should be stuck in the path dependence of old energy structures. For example, the combination of solar energy and battery storage has immense potential, and nuclear energy, geothermal energy, etc., are also worth paying attention to. We also need to solve supporting issues, such as grid modernization, approval efficiency, and labor shortages – especially skilled labor like electricians. In the next decade, the speed and efficiency of upgrading energy infrastructure will determine the ceiling for AI computing power expansion.
Q: Is Google's current business already limited by energy?
Pichai: Frankly speaking, yes. Especially in the cloud computing business, we already feel the pressure of being "constrained." This limitation comes not only from the power supply itself but also from practical challenges such as approval processes and engineering personnel shortages, which significantly affect the speed of project advancement. This is not a future problem; it is a reality we face every day. If these limitations are not resolved, they will become a more serious development bottleneck in the future. To maintain leadership globally, especially in competition with China, we must solve these problems as quickly as possible.
Q: If China's power generation capacity reaches four times that of the US in 15 years, would that mean China's GDP will also surpass the US? Or, could AI potentially expand the entire "economic pie"?
Pichai: I believe in the power of market mechanisms and technological innovation. Historically, the US has never fallen behind at critical junctures. I am confident that we will take action to address the current structural challenges. There are already many companies driving cutting-edge energy technologies, such as Small Modular Reactors (SMRs) and nuclear fusion. If it truly reaches the stage where it "must be solved," I believe public opinion, policy support, and the power of the capital markets will collectively drive the emergence of solutions.
Q: Google has often invested early in some frontier technologies with long-term returns, such as TPUs, DeepMind, Waymo, etc., although they were often underestimated at first. Today, Waymo might even grow into a hundred-billion-dollar business. What do you think of this strategy?
Pichai: Indeed, our consistent approach is "long-term investment + patient waiting." Even if the profitability model is unclear in the short term, we will persist in investing, such as in quantum computing.
06 Quantum Computing and Robotics: Is the Next "Magic Moment" Near?
Q: Why does Google invest in quantum computing long-term? What is its significance for human computing capability? When can we see practical applications?
Pichai: The reason we continue to invest in quantum computing is that it aligns with one of our core ways of thinking: first principles. The universe is fundamentally quantum – if you want to simulate truly complex systems in nature, you must ultimately do so through quantum computing. This is something classical computing cannot fully handle.
I often say that quantum computing today is somewhat like artificial intelligence around 2015 – in a stage that seems slow but where the technological curve is about to break through. We believe that within the next three to five years, it is possible to see a real-world computing task completed in a quantum way, with performance far exceeding classical computing. This will be a "turning point" kind of progress, truly demonstrating the value of quantum computing.
Of course, quantum computing is a highly challenging engineering field and may encounter technological bottlenecks. However, we have verified similar "long-term accumulation + non-linear explosion" paths in other basic technology areas, such as AI and TPU chips, so we are very confident about its long-term prospects.
The true value of a technology platform often only becomes evident with exponential impact after supporting infrastructure is mature and ecosystems are formed. From this perspective, quantum computing, like AI and cloud computing, is a potential underlying platform that can change the entire computing paradigm. Our goal is to gradually achieve quantum algorithms with practical application value in the next few years and ultimately provide them externally through cloud services.
Q: This year can be called the "Year of Robotics." Many models are being trained, based on both simulated data and real-world observation data. These models are used to control physical systems. What do you think of the development of robotics technology? What is Google's role in this field?
Pichai: Google currently has one of the world's most advanced robotics technology research teams, particularly in our newly launched Gemini Robotics project, which covers key areas such as vision, language, and motion models. These technological breakthroughs place us at the forefront of robotics technology.
We acquired robotics companies like Boston Dynamics early on, but at that time, robotics technology was not yet deeply integrated with AI, so we chose not to productize these technologies. Today, the deep integration of AI and robotics technology is the real explosive point. We are seriously considering how to launch products through partners or by Google itself.
Currently, the "magic moment" for robotics technology is gradually approaching. In the past, the humanoid robots we saw had rigid movements, but now, if you don't look closely, it's often difficult to distinguish between a robot operation and a video effect. These advancements make me believe that a leapfrog breakthrough in the field of robotics might be just two to three years away. We are looking forward to the arrival of this stage.
Q: Is it possible for Google to develop an operating system for the robotics field similar to Android and eventually capture a large market share?
Pichai: Yes, this is one of the directions we are actively investing in. We are committed to providing support to robotics manufacturers and are committed to extending the Gemini model to the robotics field, ensuring it can be compatible with existing robotics systems. Our goal is to provide an open platform that supports first-party and third-party product releases.
07 The Talent War in the AI Era Remains Fierce, How to Find the Next Hundred-Billion-Dollar Business
Q: In the AI field, does Google feel the impact of increased competition or cultural changes on talent recruitment?
Pichai: The talent market has always been fiercely competitive, especially in the AI field. Fortunately, Google has always attracted some of the best talent, and we have a unique advantage in this regard. Google employees have founded over 2000 companies, and this virtuous cycle allows us to maintain innovation vitality while also attracting more excellent talent. Some former employees who left have also returned to the company to participate in the development of new technologies, which makes me optimistic about the future.
Q: How do you think AI will change education, especially how to discover, recruit, and cultivate young talent? Will the traditional higher education system change as a result?
Pichai: AI has huge potential in the field of education. To some extent, we might misunderstand the true meaning of a university. A university is not just a place for acquiring knowledge; it is also a community, a place for people to interact and share. In my opinion, AI makes it possible for excellent talent from around the world to stand out, no longer limited to specific regions. Therefore, the source and cultivation methods of talent will change in the future. We will see more excellent talent emerging from different places globally.
Q: Do you still view Alphabet as a holding company? Or is Google still the core engine around which other businesses develop?
Pichai: We are not a traditional holding company and cannot be simply defined as a company that "finds high-quality investment opportunities and allocates capital." Our mode of operation is based on technology and R&D, with the goal of solving problems and proposing innovative solutions. If we find an area with huge potential, we will actively invest.
The structure of establishing Alphabet is actually an extension of this way of thinking. Although we have multiple seemingly unrelated businesses, there is a common core behind them – the extension of basic technologies. For example, Waymo's progress is inseparable from our efforts in Gemini and the AI field, which also drives the development of businesses like Google Cloud, Search, YouTube, Isomorphic (a drug discovery company), robotics, etc. From a technical perspective, these businesses are all centered around the same main thread.
Q: Does X (formerly Google X, Google's "moonshot lab") still play an important role today? Are you still continuously investing there?
Pichai: Of course, X is still a key force for innovation. Many groundbreaking projects, such as early versions of Waymo and Google Brain, originated from X. As an incubator, X allows us to push technological boundaries and explore projects with huge potential. These innovations are deeply rooted in deep tech research like computer science and physics, forming the foundation for all our businesses.
08 Biggest Pride + Biggest Regret, Reveals Nearly Acquiring Netflix
Q: In your past 10 years as CEO, what is your biggest regret or mistake? And what are you most proud of?
Pichai: What I am most proud of, undoubtedly, is that we are one of the few companies capable of advancing the technological frontier. Very few companies in the world can win a Nobel Prize based on their fundamental research, but we did, and we commercialized these researches, creating huge value. This is our company's unique strength.
As for regrets, I always look forward and learn from every mistake. Indeed, there were some acquisitions we were close to making but ultimately missed.
Q: Reveal the name of one acquisition target.
Pichai: Saying this might cause trouble... maybe Netflix? We discussed acquiring it once, and the discussion was very intense. Although we don't regret it, looking back, we might think: "What if we had made a different choice?"
Editor: Xingdong Da Xiong
Related reading:
Google Search Abandons Country Top-Level Domains, Unifies Globally to Google.com
Google Releases Inference Model Gemini 2.5, Its "Smartest AI Model" to Date
Supabase: A Free Open Source Alternative to Google Firebase