Why Google Is the Dominant AI Company and a Historic Buying Opportunity

TechI Was Wrong. This Is A Historic Buying Opportunity.

TL;DR

Google is the only company owning every layer of the AI stack (models, chips, data centers, networks, browser), reduced inference costs by 78% last year, and processes 3.2 quadrillion tokens monthly across 13 products with over a billion users each, making its massive capex spending ($200B in 2026, $811B in future purchase agreements) a justified investment rather than reckless burning of cash.

Key Takeaways

1Google reduced Gemini serving costs by 78% last year and core AI response costs by 30% since November by owning and optimizing every layer of its AI stack, from custom TPU chips to private undersea fiber optic cables.
2Google Cloud revenue grew 82% year-over-year to $24.8 billion in Q2, with operating margins jumping from 21% to 36% in a single year—accelerating despite being a $24.8 billion business.
3Google processes 3.2 quadrillion tokens per month (equivalent to 800 full-length HD movies per second), distributed across Gemini app (950 million MAU), AI Overviews (2.5 billion monthly reach), and 8.5 million developers building on Google's models.
4Google signed $811 billion in future purchase agreements for chips, equipment, and electricity—including contracts running until 2041—and increased capex guidance from $185 billion to $200 billion for 2026, with 2027 spending expected to be even higher.
5Google is now directly selling TPU chips to other companies' data centers for the first time, marking a shift from only renting them through Google Cloud and expected to significantly ramp up product revenue given the $514 billion backlog.
6Despite processing 300x more tokens over two years, Google's free cash flow turned negative at negative $5.9 billion in Q2 (versus $24.6 billion two quarters earlier), causing the stock to drop 7% despite posting record growth metrics.
7Unlike Microsoft (which relies on OpenAI for heavy workloads) or Amazon (which uses Anthropic), Google is the only company controlling the entire AI stack end-to-end, enabling proprietary optimizations and rapid innovation without supplier dependencies.

Notable Quotes

“Google is the only company on Earth that owns every single part of the stack.”
— Alex

“Google Cloud generates product revenues primarily from the sale of TPU systems. This is the first quarter that Alphabet recognized revenues from selling their TPUs.”
— Alex

“Some of this money is for energy contracts that run until 2041, which means Google signed electricity bills that are due 28 years from now for data centers that don't even exist yet. That's how serious this AI race is becoming.”
— Alex

“While analysts see a company that went from being one of Wall Street's safest picks to one of the world's biggest burners of cash, I see the only company on Earth that can actually justify this level of spending.”
— Alex

Chapters

1. The Three Stages of AI: Training, Fine-Tuning, and Inference

AI models predict the next step across three lifecycle stages: training (showing models massive datasets to learn patterns), fine-tuning (teaching models to give helpful, formatted answers for specific use cases through feedback), and inference (the real-time execution in under one second that customers actually pay for). Fine-tuning costs a fraction of training and is where distillation occurs—teaching smaller, cheaper models by having them copy a larger model's work.

2. Google's Vertical Integration: Owning Every Layer of the AI Stack

Google uniquely owns all components: Gemini AI models, custom TPU chips (split into 8T for training and 8i for inference), proprietary data centers operational since 2015, private fiber optic networks including undersea cables, Chrome browser, YouTube, and Google Search. This vertical integration allows Google to optimize any layer without supplier permission, unlike Microsoft (which relies on OpenAI) or Amazon (which uses Anthropic).

3. The Technology Answer: Cost Reductions and Explosive User Growth

Google reduced Gemini serving costs by 78% last year and core AI response costs by 30% since Gemini 3 launch in November. These efficiencies enabled 13 products with over a billion users each: AI Overviews (2.5 billion monthly reach), AI Mode (1 billion MAU), Gemini app (950 million MAU, doubled year-over-year), and 8.5 million developers. Token processing exploded 300x in two years to 3.2 quadrillion monthly—equivalent to 800 HD movies per second—demonstrating Jevons Paradox: lower costs drove exponential demand.

4. Q2 Earnings: Record Growth Masked by Free Cash Flow Collapse

Alphabet reported $120 billion revenue (up 24% YoY, 12th straight quarter of double-digit growth), but $99 billion of the $112 billion net income came from paper gains on SpaceX shares worth $94 billion; stripping these out reveals earnings per share of $2.85 (below Wall Street expectations). Google Cloud revenue hit $24.8 billion, up 82% YoY with operating margins jumping from 21% to 36%. However, free cash flow turned negative at negative $5.9 billion (versus $24.6 billion two quarters earlier), causing a 7% stock drop.

5. The Capex Deluge: $200B Annual Spending and $811B in Committed Purchases

Google spent $45 billion in Q2 alone (double year-ago levels) and raised full-year 2026 capex guidance from $185 billion to $200 billion, with 2027 expected to be much higher. Alphabet signed $811 billion in future purchase agreements (up from $332 billion three months earlier) for chips, equipment, data centers, and electricity—including 28-year electricity contracts through 2041 for data centers that don't exist yet. Share buybacks halted for the first time since 2017, and long-term debt more than doubled in six months.

6. New Revenue Stream: Direct TPU Chip Sales to Enterprise Customers

Q2 marked the first time Google Cloud recognized revenue from directly selling TPU systems to other companies' data centers—a shift from the decade-long rental-only model through Google Cloud. With a $514 billion backlog, these product revenues are expected to ramp significantly in coming quarters as enterprises install custom Google chips alongside competing infrastructure.

7. The Investment Thesis: Why Google's Spending Isn't Reckless

While Wall Street views Google's negative free cash flow and doubled capex as a warning sign, the speaker argues this is justified investment, not cash burning. Google Cloud grows faster than Azure and AWS combined, maintains a $514 billion backlog, serves billions monthly across every device type, and is the only company that can justify this spending because it owns the entire stack—enabling proprietary optimization, rapid iteration, and ultimate scale economies that no competitor can match.

Key People & Entities

Alex (TickerSpy): Content creator and investment analyst; MIT electrical engineer and AI researcher with 8 years experience; host of the video
Alphabet/Google: Parent company of Google, controlling AI models (Gemini), custom chips (TPUs), data centers, Chrome browser, YouTube, Google Search, and Google Cloud
OpenAI: AI model company that Microsoft relies on for heavy workloads; competitor to Google's Gemini
Anthropic: AI model company that Amazon uses; Google invested approximately $900 million; also mentioned as Amazon's primary AI partner
Microsoft: Has custom chips (Maya), data centers (Azure), Windows and Office ecosystems, but depends on OpenAI for heavy AI workloads
Amazon Web Services (AWS): Cloud infrastructure competitor to Google Cloud; uses Anthropic for AI workloads rather than owning its own models
SpaceX: Aerospace company in which Google invested $900 million in 2015; stake worth $94 billion at quarter-end but subject to lockup periods
Ground News: News analysis platform that rates sources for bias and factuality; offers media literacy tool featured in the video (sponsor)

Glossary

LLM (Large Language Model): An AI model that predicts the next word in a sequence by learning patterns from enormous amounts of text data.
Parameters: Billions of numerical values that describe everything a trained AI model has learned; a model with 400 billion parameters contains 400 billion of these adjustable numbers.
Fine-tuning: The second stage of an AI model's lifecycle where it receives feedback (thumbs up/down, additional instructions) to learn how to give helpful, formatted answers for specific use cases.
Distillation: A technique where a large, expensive AI model teaches smaller models by having them copy not just the answers but also the reasoning process, creating cheaper models for specific tasks.
Inference: The third stage of an AI model's lifecycle where it processes a user's new query and generates an answer in real-time; the step that happens between hitting enter and receiving results.
TPU (Tensor Processing Unit): Google's custom-designed AI chip optimized for training and running machine learning models; Google's 8th generation split into 8T (for training) and 8i (for inference).
Jevons Paradox: An economic principle where increased efficiency and lower costs for a product actually lead to higher total consumption and spending because demand rises faster than supply improves.
Vertical Integration: A company owning multiple layers of its supply chain or production process; Google controls models, chips, data centers, networks, and distribution channels end-to-end.
Capex (Capital Expenditure): Money a company spends on physical assets like buildings, equipment, and infrastructure rather than day-to-day operations; Google's 2026 capex is $200 billion.

Explore