Google CEO Sundar Pichai: AI’s Infinite Possibilities, Compute Scarcity Dilemmas, Future AI Products

Anthropic

Across the shows: 25 speakers in other episodes, 15 bullish and 36 bearish calls, past 30 days

• Sundar Pichai · Alphabet / GoogleBullish

  Anthropic has a world-class founding team that Google judged would reach the frontier, so it invested.

  Pichai says that early on Google saw Anthropic's founding team was world class, with members involved in some of the earliest GPT models. From that, Google quickly judged the company would reach the frontier. Google's approach is to identify and partner with companies that will be at the cutting edge, and he cites SpaceX and Anthropic as examples. Some of those partnerships have taken the form of TPU deals.

  “When we looked at the founding team of Anthropic, it was very clear to us. You know, this was a world class team, which was involved in some of the earliest GPD models.”at 35:24

AI infrastructure & data centers

• Sundar Pichai · Alphabet / Google

  Everyone is short of compute; demand runs well above what companies planned to build.

  Pichai says the compute Google has today reflects decisions made in 2024, because it builds its own data centers and faces long lead times. He says everyone is short, not just Google, and everyone wishes they had twice what they had planned for. Google leased some short-term compute last year to help customers. When asked whether Andy Jassy's claim that AWS is sold out through 2028 is familiar, he says customers asking for compute in 2027 is incredibly difficult, and he expects all providers to face similar timelines.

  “Everyone is short, not just us, other companies. And everyone wishes you were 2x of what you had planned for.”at 18:57

• Sundar Pichai · Alphabet / Google

  AI data center build-out makes sense long-term, though bottlenecks and macro shocks could cause overbuild.

  Pichai says whether this is overbuilding depends on the time frame. Over a long enough horizon the build-out makes complete sense, because AI is early in what he calls the most consequential technology ever. Unlike fiber, the capacity is being used and is constrained daily. Google plans from demand signals, committed contracts and internal evals, and is confident the compute it plans will find uses. Overbuild is still possible: the build needs many complements, including memory, electricity, electricians and land, so constrained components can cause bottlenecks, and oil shocks or other macro shocks are risks.

  “It depends on the time frames you're looking at... if I take a long enough timeframe, it makes complete sense to me. We are at early stages of the most consequential technology ever. And this is not like fiber, it's not dark, we are using it.”at 20:56

Consumer AI products

• Sundar Pichai · Alphabet / Google

  Technology will adapt to people rather than the reverse, though screens will remain for some uses.

  Pichai says computing has historically meant humans adapting to machines, and with AI the technology will adapt seamlessly to how people live. Assistants should be available on demand, have context and run persistently. He already uses Gemini voice input on his phone. He still expects some interface, because people want to watch and consume the content AI creates, such as video.

  “With AI, it'll be the other way about. The technology will adapt seamlessly to how humans live and go about their day to day lives.”at 25:26

• Sundar Pichai · Alphabet / Google

  Consumer AI at scale will be monetized; agents doing tasks will create commercial value and a clearing mechanism.

  Asked how an agent that recommends a product should share the sale, Pichai argues monetization follows value. Software developers already pay for coding models because those models make them much more productive. He says he has never seen a consumer product reach billions of users with intense use and no monetization. If agents deliver value, providers will want to reach users with commercial needs, and some mechanism will emerge to settle it, as in an efficient system.

  “I've never seen examples where consumer products really scale to billions of users where people are using it intensely, and hasn't had monetization to go along with it.”at 30:01

AI pricing & unit economics

• Sundar Pichai · Alphabet / Google

  Good products do not need the most frontier model, since frontier models are slower and costlier.

  Pichai agrees that very good product experiences can be built without the most frontier model, because frontier models tend to be slower and more expensive. That is why Google built Flash and other smaller models, and he says there is room for models of all sizes. He still says Google cares about the Pareto frontier of cost against capability, and that frontier work matters for innovation and for building the next models.

  “You can develop very good product experiences with not necessarily the most frontier model because they tend to be slower and more expensive too.”at 7:17

AI: overall outlook

• Sundar Pichai · Alphabet / Google

  Recursive self-improvement loops exist now, but Google is far from the end state; progress expected next year.

  Pichai says RSI is a critical contributor to developing Argon. He is clear that the industry is still far from an end state of RSI. What exists today is RSI loops alongside AI researchers in various parts of model development, where agents help researchers in automated ways to build models. He expects that continuum to keep making progress into next year.

  “In a continuum... we are pretty far from... the end state of RSI or something like that. ... you almost have agents helping AI researchers in some automated way, work together to build these models.”at 10:38

AI regulation & safety

• Sundar Pichai · Alphabet / Google

  Industry needs shared standards and independent evaluators on safety; regulation will eventually play a role.

  Pichai says safety comes from defense in depth: alignment work, real-time monitoring and telemetry to intervene when behavior is unexpected, and enterprise control over where models deploy and what data they access. He says the next step is industry standards with independent evaluators, citing financial controls, accounting standards and independent auditors as the model. He says regulation will play a role as the industry goes through this.

  “I think the next step would be developing better standards, so that across industry standards of how we do these things, including ability to evaluate using independent evaluators.”at 13:33

Robotics & autonomous vehicles

• Sundar Pichai · Alphabet / Google

  Google's world-model research is state of the art for embodied reasoning, which supports robotics.

  Pichai argues world models are key enabling blocks for intelligence because they let models understand physics and simulate the world in high fidelity. He says that across applications, robotics is an example: Google's embodied reasoning benchmarks are state of the art, and he says the model literally topped many of them. He says this comes from Google's experience building world models, and that the company wants to be at the frontier of all approaches, including the LLM path.

  “If you look at, you know, embodied reasoning benchmarks, [Argon] is state of the art, right? I think it literally sowed on many of these benchmarks.”at 15:06