Google: The AI Company

Unknown Source October 06, 2025 247 min

artificial-intelligence generative-ai ai-infrastructure investment startup google nvidia anthropic

200 Companies

382 Key Quotes

5 Topics

19 Insights

🎯 Summary

This 246-minute podcast episode, “Google: The AI Company,” provides a deep, historical, and strategic analysis of how Google became the epicenter of the modern AI revolution, focusing heavily on the development and impact of the Transformer architecture and the ensuing competitive landscape.

Here is a comprehensive summary structured for professional insight:

Comprehensive Podcast Summary: Google: The AI Company

1. Focus Area

The primary focus is the history, strategic positioning, and inherent dilemma facing Google as the “AI Company.” This includes tracing the lineage of key AI talent from Google, the foundational role of the Transformer architecture (published by Google Brain in 2017), and the internal tension between protecting the highly profitable Search monopoly and aggressively pursuing the less immediately profitable, but potentially revolutionary, Generative AI wave. Key technical discussions revolve around early probabilistic language models (like PHILL), the concept of data compression equaling understanding, and the engineering feats required to scale these models (e.g., parallelization).

2. Key Technical Insights

Compression as Understanding: The early hypothesis, posited by Google engineer George Herrick, that data compression is technically equivalent to understanding foreshadowed the core mechanism of modern Large Language Models (LLMs), which compress vast knowledge into dense vector representations.
The Transformer’s Genesis: The entire current AI boom (OpenAI, Anthropic, etc.) is predicated on the Transformer paper published by Google Brain in 2017, highlighting Google’s foundational role in creating the architecture that powers modern generative AI.
Engineering for Scale (The Jeff Dean Effect): Google’s early language models, like PHILL, were computationally prohibitive (12 hours per sentence). Jeff Dean’s intervention, by rearchitecting the system for massive parallelization across Google’s distributed infrastructure, reduced translation time to 100 milliseconds, proving the viability of large-scale production AI.

3. Business/Investment Angle

The Innovator’s Dilemma: Google faces a classic dilemma: launching superior AI products (based on the Transformer) threatens the massive, high-margin profits of its existing Search business, creating a strategic hesitation to fully commit resources.
Unique Asset Concentration: Google is uniquely positioned as the only company possessing both a frontier foundational model (Gemini) and proprietary, scaled AI chips (TPUs), positioning them as a top-tier player alongside Nvidia (GPUs).
Search as the Moat: Despite the AI disruption, Google still controls the “text box”—the primary entry point to the internet for intent-based queries—which remains the critical asset to leverage for AI monetization.

4. Notable Companies/People

Google Brain/DeepMind Talent: The episode emphasizes that nearly every major figure in modern AI—including Ilya Sutskever, Jeff Hinton, Alex Krizhevsky, Dario Amodei (Anthropic), and Demis Hassabis (DeepMind)—was once a Google employee, underscoring Google’s historical dominance in AI talent acquisition.
Jeff Dean: Portrayed as the engineering linchpin, responsible for critical infrastructure optimizations (like parallelizing PHILL) and a legendary figure whose efficiency is immortalized in “Jeff Dean Facts.”
Larry Page: His early vision defined Google as an AI company from its inception in 2000.
Noam Shazeer & George Herrick: Key early researchers who pioneered probabilistic language models at Google, leading to the development of PHILL.

5. Future Implications

The conversation suggests the industry is at an inflection point where having both a frontier model and custom silicon (AI chips) will separate commodity players from market leaders. The central question remains whether Google will overcome the profit protection instinct to fully capitalize on its foundational AI breakthroughs, or if competitors who have less legacy revenue to protect will seize the lead. The future hinges on Google’s ability to integrate Gemini effectively into its core services without cannibalizing search revenue.

6. Target Audience

This episode is highly valuable for AI/ML professionals, technology strategists, venture capitalists, and corporate executives interested in the competitive dynamics of the generative AI landscape, particularly those tracking Big Tech strategy and the history of foundational research.

🏢 Companies Mentioned

transformer ✅ ai_research

Wemo ✅ ai_application

Oracle ✅ other_business

Saudi Aramco ✅ other_business

YouTube ✅ big_tech

Amazon (implied via Azure) ✅ big_tech

Franz Axe ✅ ai_researcher

Epic ✅ company

TSMC ✅ ai_manufacturing

Erzholzul ✅ big_tech_leader

Jensen ✅ ai_infrastructure_leader

Apple ✅ big_tech

Stripe ✅ ai_application

Kate Matts ✅ ai_application

Versel ✅ ai_infrastructure

💬 Key Insights

"and to Brian Lawrence from Oak Cliff Capital for helping me think about the economics of AI data centers"

Impact Score: 10

"if a i isn't as good a business as search and they're choosing between two outcomes one is fulfilling our mission of organizing the world's information and making it universally accessible and useful and having the most profitable tech company in the world which one wins because if it's just the mission they should be way more aggressive on AI mode than they are right now"

Impact Score: 10

"google being definitively the low cost provider of tokens because they operate all their own infrastructure and because they have access to low markup hardware it actually makes a giant difference and might mean that they are the winner in producing tokens for the world"

Impact Score: 10

"normally like in historical technology eras it hasn't been that important to be the low cost producer google didn't win because they were the lowest cost surge engine apple didn't win because they were the lowest cost you know that's not what makes people win but this era might actually be different because these ai companies don't have 80 percent margins at the way that we're used to in the technology business"

Impact Score: 10

"I've seen estimates that over half the cost of running an a i data center is the chips and the associated depreciation the human cost that rnd is actually a pretty high amount because hiring these air researchers and all the software engineering is meaningful call it 25 to 33 percent the power is actually a very small part it's like two to six percent"

Impact Score: 10

"Google has all the capabilities to win an AI and it's not even close foundational model chips hyperscaler all this with self-sustaining funding I mean that's the other crazy thing as you look at the clouds have self-sustaining funding and video has self-sustaining funding none of the model makers have self-sustaining funding so they're all dependent on external capital"

Impact Score: 10

📊 Topics

#artificialintelligence 619 #generativeai 56 #aiinfrastructure 50 #investment 37 #startup 34

🧠 Key Takeaways

💡 so large that they keep pouring tens of billions of dollars into these competitors yeah plenty of other folks have made the sort of glib comment but there's merit to hey as flatfooted as Google was when Chatship T happened if the outcome of this is they avoid a Microsoft level distraction and damage to their business from a US federal court monopoly judgment worth it well there's a funny meme here that you could draw you know that meme of someone pushing the domino and it knocking over some big wall later yeah there's the domino of Ilya leaving Google to start open AI and the downstream effect is Google is not broken up yeah right exactly actually saves Google it actually saves Google it's totally wild totally wild all right so here's the business today over the last 12 months Google has generated three hundred and seventy billion dollars in revenue on the earnings side they've generated a hundred and forty billion over the last 12 months which is more profit than any other tech company and the only company in the world with more earnings is Saudi or Ramco let's not forget Google is the best business ever and we also made the point at the end of the health a bit episode even in the midst of all of this AI era and everything that's happened over the last 10 years the last five years Google's court business has continued to grow five X since the end of our alphabet episode in 2015 2016 yeah market cap Google surge past their old peak of two trillion and just hit that three trillion mark earlier this month they're the fourth most valuable company in the world behind Nvidia Microsoft and Apple it's just crazy on their balance sheet actually think this is pretty interesting I normally don't look at balance sheet as a part of this exercise but it's useful and here's why in this case they have 95 billion in cash and marketable securities and I was about to stop there and make the point wow look how much cash and resources they have actually surprised it's not more so it used to be a hundred and forty billion in twenty twenty one and over the last four years they've massively shift from this mode of accumulating cash to deploying cash and a huge part of that has been the capex of the AI data center build out so they're very much playing offense in the way that met a Microsoft and Amazon are into playing that capex but the thing that I can't quite figure out is the largest part of that was actually buybacks and they started paying a dividend so if you're not a finance person the way to read into that is yes we still need a lot of cash for investing in the future of AI and data centers but we still actually had way more cash than we needed and we decided to distribute that to shareholders yeah that's crazy best business of all time right that illustrates what a crazy business their core search ads businesses if they're saying the most capital intense race in business history is happening right now we intend to win it yep and we have tons of extra cash lying around on top of what we think plus a safety cushion for investing in that capex race yeah yes well so there are two businesses that are worth looking at here one is Gemini to try to figure out what's happening there and two is a brief history of Google Cloud I want to tell you the cloud numbers today but it's probably worth actually understanding how did we get here on cloud yep first on Gemini because this is Google and they have I think the most obfuscated financials of any of the companies we've studied they anger be the most in being able to hide the ball in their financial statements of course we don't know Gemini specific revenue what we do know is there are over 150 million paying subscribers to the Google one bundle most of that is on a very low tier that's on like the five dollar a month ten dollar a month the AI stuff kicks in on the twenty dollar a month tier where you get the premium AI features but I think that's a very small fraction of the hundred and fifty million today I think that's what I'm on but two things to know one it's growing quickly that hundred and fifty million is growing almost 50 percent year over year but two is Google has a subscription bundle that a hundred and fifty million people are subscribed to and so I've kind of had it in my head that AI doesn't have a future as a business model that people pay money for that it has to be ad supported like search but hey that's not nothing that's like a that's almost half of America I mean how many subscribers does Netflix have Netflix is in the hundreds of millions yeah Spotify is now a quarter billion something like that yeah we now live in a world where there are real scaled consumer subscription services I owe this insight to Shishir Morodo we chatted actually last night because I named dropped him in the last episode and then he heard it and so we reached out we talked and that's made me do a 180 I used to think if you're going to charge for something your total address will market shrunk by 90 to 99 percent but he kind of has this point that if you build a really compelling bundle and Google has the digital assets to build a compelling bundle oh my goodness YouTube premium NFL Sunday ticket yes stuff in the play store YouTube music all the Google one storage stuff they could put AI in that bundle and figure out through clever bundle economics a way to make a paid AI product that actually reaches a huge number of paying subscribers totally so we really can't figure out how much money Gemini makes right now probably not profitable anyway so what's the point of even analyzing it yep but okay tell us the cloud story so we intentionally did not include cloud in our alphabet episode Google part two effectively Google part two yes because it is a new product and now very successful one within Google that was started during the same time period as all the other ones that we talked about during Google part two but it's so strategic for AI yes it is a lot more strategic now in hindsight than it looked when they launched it so just quick background on it it started as Google App Engine it was away in 2008 for people to quickly spin up a backend for a web or soon after a mobile app it was a platform as a service so you had to do things in this very narrow Googley way it was very opinionated you had to use this SDK you had to write it in Python or Java you had to deploy exactly the way they wanted you to deploy it was not a thing where they would say hey developer you can do anything you want just use our infrastructure it was opinionated super different than what AWS was doing at the time and what they're still doing today which the whole eventually realized was right which is cloud should be infrastructure as a service even Microsoft pivoted Azure to this reasonably quickly where it was like you want some storage we got storage for you you want a VM we got a VM for you you want some compute you want a database we got to fundamental building blocks so eventually Google launches their own infrastructures service in 2012 took four years they launched Google compute engine that they would later rebrand Google cloud platform that's the name of the business today the knock on Google is that they could never figure out how to possibly interface with the enterprise their core business they made really great products for people to use that they loved polishing they made them all as self service possible and in the way they made money it was from advertisers and let's be honest there's no other choice but to use Google search right it didn't necessarily need to have a great enterprise experience for their advertising customers because they were going to come anyway right so they've got this self serve experience meanwhile the cloud is a knife fight these are commodities all about the enterprise it's the lowest possible price and it's all about enterprise relationships and clever ways to bundle and being able to deliver a full solution you say solution I hear grows margin yes but yes so Google out of their natural habitat in this domain and early on they didn't want to give away any crown jewels they viewed their infrastructure as this is our secret thing we don't want to let anybody else use it and the best software tools that we have on it that we've written for ourselves like big table or board how we run Google or dist belief these are not services that we're making available on Google cloud yeah these are competitive advantages yes and then they hired the former president of Oracle Thomas Kerian yes and everything kind of changed so 2017 two years before he comes in they had four billion dollars in revenue 10 years into running this business 2018 is their first very clever strategic decision they launch Kubernetes the big insight here is if we make it more portable for developers to move their applications to other clouds the world is kind of wanting multi cloud here right where the third place player we don't have anything to lose yes so we can offer this tool and kind of counter position against AWS and Azure we shift the developer paradigm to use these containers they orchestrate on our platform and then you know we have a great service to manage it for you it was very smart so this kind of becomes one of the pillars of their strategy is you want multi cloud we're going to make that easy and you can sure choose AWS or Azure to it's going to be great so David as you said the former president of Oracle Thomas Kerian is hired in late 2018 you couldn't ask for a better person who understands the needs of the enterprise than the former president of Oracle this shows up in revenue growth right away in 2020 they crossed 13 billion in revenue which was nearly tripling in three years they hired like 10,000 people into the go-to-market organization I'm not exaggerating that and that's on a base of 150 people when he came in most of which were seated in California not regionally distributed throughout the world the funniest thing is Google kind of was a cloud company all along they had the best engineers building this amazing infrastructure right they had the products they had the infrastructure they just didn't have the go-to-market organization right and the productization was all like Google it was like for us for engineers they didn't really build things that let enterprises build the way they wanted to build this all changes 2022 they hit 26 billion in revenue 2023 they're like a real viable third cloud they also flipped a profitability in 2023 and today they're over $50 billion in annual revenue run rate it's growing 30% year over year they're the fastest growing of the major cloud providers 5x and 5 years and it's really three things it's finding religion on how to actually serve the enterprise it's leaning into this multi cloud strategy and actually giving enterprise developers what they want and three AI has been such a good tailwind for all hyper scalars because these workloads all need to run in the cloud because it's giant amounts of data and giant amount of compute and energy but in Google Cloud you can use TPUs which they make a ton of and everyone else is desperately begging Nvidia for allocations to GPUs so if you're willing to not use CUDA and build on Google Stack they have an abundant amount of TPUs for you

💡 do this a lot

💡 bring in AI professors academics

💡 just start a whole new division within Google and it becomes Google X, the moonshot factory

💡 each have 30 and that's how it ends up breaking down