Sundar Pichai, CEO of Alphabet | The All-In Interview
Speakers
Sundar Pichai
— CEO, Alphabet
David Friedberg
— CEO, The Production Board
Summary
David Friedberg interviews Alphabet CEO Sundar Pichai at the Googleplex about whether AI is disrupting Google search. Pichai argues search is expanding rather than being cannibalized, citing AI Overviews at over 1.5 billion users, a sharp fall in cost per query, and AI Overview ads already monetizing at baseline, and frames Google's TPUs and full-stack infrastructure as a structural cost advantage behind Gemini. The conversation then covers Nvidia, DeepSeek and China, the electricity constraint on AI scaling, quantum computing, robotics, and Google's culture, talent and Alphabet's structure.
- Pichai says AI Overviews (1.5B users, 150+ countries) and the new AI Mode expand search queries instead of cannibalizing them.
- Cost to serve a query has fallen dramatically in 18 months; latency is the real constraint, and AI Overview ads already match the non-AI baseline.
- Seventh-generation TPUs, Ironwood at 40+ exaflops and infrastructure down to subsea cables put Google on the performance-cost Pareto frontier.
- Nvidia stays a core part of the mix: Google serves much Gemini traffic on GPUs while remaining long-term committed to TPUs.
- Roughly $75B of 2025 capex, about half of compute going to a supply-constrained Google Cloud; YouTube plus Cloud exited last year at $110B.
- DeepSeek showed China is closer to the AI frontier than assumed, though Google benchmarked Gemini Flash as equally or more efficient.
- Electricity is the likely binding constraint on AI; Pichai backs solar plus batteries, nuclear, geothermal, grid and transmission upgrades, and more electricians.
- Quantum is where AI was in 2015 with a practical breakthrough expected in about five years, while robotics is two to three years from its magical moment.
Ideas
AI expands search rather than disrupting it.
Pichai says quantum computing today is roughly where AI was around 2015 and that within a five-year time frame there will be an aha moment where a genuinely useful practical computation is done in a quantum way far better than on classical computers. Google has worked on it out of long-term conviction regardless of outside attention, expects more exciting milestones to share this year, and would give access through cloud. He compares his internal read to self-driving three years ago, when many players looked equivalent but he could tell how far ahead Waymo was, and says he feels the same about Google's quantum effort despite a lot of noisy announcements in the industry.
Nvidia stays world class and core.
Asked whether TPUs wholesale replace Nvidia, Pichai says Google uses both and remains a committed Nvidia customer: a lot of Gemini traffic is served on GPUs, Google is deploying GPUs internally, and it gives cloud customers the choice. He calls Nvidia a phenomenal company whose R&D pace and world-class software stack give it a lot of advantages, and says the two push each other forward, while noting that everyone in the industry will try to build custom silicon, so Google stays long-term committed to TPUs alongside GPUs.
AI pie grows; many winners.
Pichai rejects the winner-take-all framing of the AI race: he expects all the major participants to do well because the opportunity landscape is bigger than all previous technology waves combined, so the pie grows rather than being divided. He points out that when the internet arrived Google did not exist, so companies that have not been founded yet may end up among the biggest AI winners, and argues the differentiator will be the ability to innovate and execute with the best talent rather than who beats whom in the marketplace.
Power is AI's solvable bottleneck.
Pichai agrees electricity is the most likely constraint gating AI progress and therefore GDP growth, and says he worries about it a lot, but insists it is an execution problem rather than a physics barrier because the supply technologies already exist. He wants no innovator's dilemma in energy and lists the areas to lean into: the innovation going into nuclear and geothermal, upgrading the grid, solving transmission, and faster permitting. Today the constraint shows up as project delays from permitting and missing electricians, which is why Google itself is supply constrained in cloud, and he expects these to become far more visible if the AI buildout continues at this pace.
Solar plus batteries stay underestimated.
Within the energy build-out Pichai singles out solar as the piece people perpetually underestimate, saying solar plus batteries will end up being huge. He names it first among the innovations the industry should embrace to meet AI-driven power demand, ahead of nuclear and geothermal.
Market keeps underrating Alphabet's patient bets.
Friedberg argues Alphabet's long-horizon bets are systematically dismissed by outsiders and then work: the TPU, the DeepMind acquisition, the decades of infrastructure investment and the insane continued investment in Waymo were all treated as small or pointless bets, and Waymo now looks on track to be a hundred-billion-dollar business. He uses that track record to argue that the bets currently being ignored, quantum in particular, are underpriced by the market, and closes by saying people underappreciate how much innovation Alphabet drives across many lanes.
Quantum's useful moment arrives within five years.
Pichai says quantum computing today is roughly where AI was around 2015 and that within a five-year time frame there will be an aha moment where a genuinely useful practical computation is done in a quantum way far better than on classical computers. Google has worked on it out of long-term conviction regardless of outside attention, expects more exciting milestones to share this year, and would give access through cloud. He compares his internal read to self-driving three years ago, when many players looked equivalent but he could tell how far ahead Waymo was, and says he feels the same about Google's quantum effort despite a lot of noisy announcements in the industry.
AI plus robotics reaches its moment.
Pichai says Google tried the robotics application layer too early, before AI was influencing robots, but the combination of AI plus robotics is now the sweet spot. Google has one of the most advanced frontier robotics R&D teams in the world, its Gemini robotics vision-language-action models are world class, and Intrinsic is effectively the bet on being the Android for robotics by supporting robot manufacturers. He says humanoid progress is now hard to distinguish from fake video at a glance, puts the magical moment two to three years away, and promises more announcements while Google decides between first-party and third-party products.
Up Next
This All-In Podcast video, published May 16, 2025, features Sundar Pichai, David Friedberg discussing GOOGL, NVDA, AI-SECTOR, URA, GEOTHERMAL ENERGY, Electric grid infrastructure, SOLAR, Battery storage, QUBT, ROBO. 8 trade ideas extracted by AI with direction and confidence scoring.
Speakers: Sundar Pichai, David Friedberg · Tickers: GOOGL, NVDA, AI-SECTOR, URA, GEOTHERMAL ENERGY, Electric grid infrastructure, SOLAR, Battery storage, QUBT, ROBO