The Baseline

The Pure-Play Illusion

The market assumes the AI war is just a battle of foundation models. But the landscape has quietly split into two tiers: Closed APIs (renting access per token) and Open Weights (running models on your own infra).

Pure-play AI labs like OpenAI and Anthropic are trapped fighting for direct API subscriptions. They only play in the Closed Tier. Alphabet is the only major company competing in both. They charge for Gemini, but give away multi-million dollar Gemma models for free. Why? Because they don't just own a model—they own the ecosystem.

The Macro View

The "Three Payoffs" Playbook

Giving away frontier-level models for free isn't charity. Google runs three strategic payoffs stacked on top of each other that pure-play competitors cannot match:

monetization_on 1. Commercial Capture

The model is free, but the rails it runs on are not. Google captures immense value by acting as the cloud funnel for enterprise AI workloads, and by protecting their Android on-device ecosystem.

block 2. Competitive Denial

Releasing world-class open models blocks Chinese labs (like DeepSeek) from becoming the Western enterprise default. Simultaneously, it heavily squeezes the premium API margins that pure-play labs depend on to survive.

psychology 3. Portfolio Reinforcement

Every developer who builds on Gemma becomes fluent in Google's AI stack. The engineer learning today becomes the executive making a $5 million Cloud procurement decision tomorrow.

Value Extraction (Micro View)

The Technology Stack Moat

Let's double-click on that first strategy: Commercial Capture. Here is exactly how Google extracts billions in value across their stack without needing an API subscription.

Swipe to explore swipe

cloud

1. Cloud Infrastructure

Google wins no matter whose model you use. They extract value at the compute bedrock by renting out custom TPUs in the cloud, leasing massive NVIDIA server clusters, and even charging to run competitors' open-source models natively on Google Cloud infrastructure. The open model is the funnel; Cloud is the business.

smartphone

2. Endpoint Devices

Distribution is the hardest hurdle for AI startups, but Google owns Android—the leading mobile OS in the world. They don't have to beg users to download a standalone app; they can deploy natively integrated AI to billions of global devices via a simple software update, protecting their hardware ecosystem and core advertising revenue.

apps

3. Billion-User Software

Search, Maps, YouTube, Gmail, Workspace. Google possesses multiple properties with over a billion active users. By seamlessly layering AI on top of these existing habitual workflows, they create immense sticky value and extract massive data without forcing users to adapt to a completely new standalone interface.

The Takeaway

Beyond the Subscription

Google doesn't need to make money solely on AI subscriptions.

If a pure-play lab's model isn't worth $20/month, they face existential risk. The model is their business. Give it away, and there is nothing left to sell. Google, however, possesses an infrastructure endurance that pure labs simply cannot match. They can afford to give away frontier AI purely to drive enterprise cloud adoption, sell more hardware, and protect their ecosystem dominance.