# Hyperscalers The tech giants spending hundreds of billions on infrastructure to dominate cloud computing and AI. Currently: Microsoft, Meta (Facebook), Amazon, Alphabet (Google), Tesla, Oracle. ## 2025 AI Arms Race Collectively spending $400B+ in capex next year pursuing compute supremacy. This is [[AI Capex Super-Cycle|individually rational, collectively irrational]] behavior. Each hyperscaler reasoning: - "If I don't invest massively, I lose AI market position" - "Winner takes most in platform economics" - "Must build defensible moat through scale" Problem: They can't all win. Most gross margin will accrue to Nvidia (selling GPUs) and the hyperscalers themselves (owning distribution). Rest face potential misallocation. ## Characteristics What defines a hyperscaler: - **Scale**: Operating millions of servers across global data centers - **Capex**: Willing to spend tens of billions annually on infrastructure - **Vertical integration**: Own the stack from hardware to applications - **Distribution**: Direct access to billions of users - **Monetization paths**: Multiple ways to recoup infrastructure spend ## Business Models Different paths to recoup AI capex: - **Microsoft**: Copilot subscriptions, Azure AI services - **Google**: Search ads, Workspace, Cloud - **Meta**: Better ad targeting, reduced content moderation costs - **Amazon**: AWS AI services, Alexa improvements - **Tesla**: FSD, Optimus, Dojo competitive advantage - **Oracle**: Database + AI integration for enterprises ## Investment Implications Questions to ask: - Who has clearest path to monetization? - Who owns critical distribution (users, developers, enterprises)? - Who can afford to give AI away to protect core business? - What happens if open-source models get "good enough"? If Google offers Gemini for free (like they did with Google Docs), who survives? Only those with different monetization models. Related: [[AI Capex Super-Cycle]], [[Future of Compute]], [[Data Center MoC]], [[Alpha in Atoms - Reflection on the Physical Inversion]] --- Tags: #deeptech #systems