The Infrastructure Layer
How the three hyperscalers — AWS, Azure, and Google Cloud — are building the physical and software infrastructure that AI runs on.
The three hyperscalers — Amazon (AWS), Microsoft (Azure), and Alphabet (Google Cloud) — are building the physical and software infrastructure that AI runs on. Their massive AI capital expenditure is often characterised as risk, but it is actually a signal about where economic value will be generated.
What the Infrastructure Layer Includes
The infrastructure layer encompasses:
- Data centres — Purpose-built facilities housing thousands of GPUs, with specialised cooling, power delivery, and networking
- Compute capacity — The raw processing power available to train and serve AI models
- Networking — High-bandwidth, low-latency connections between GPUs within data centres and between data centres globally
- Software stacks — The cloud platforms, APIs, orchestration tools, and managed services that make AI accessible to developers and enterprises
Together, these components form the substrate of the AI economy.
The Capex Signal
All three hyperscalers report increasing positive operating income from AI services. Their capital expenditure is not speculative — it is driven by customer demand that is already generating revenue. When companies with the analytical sophistication of Amazon, Microsoft, and Alphabet all independently decide to spend tens of billions on AI infrastructure, that convergence is informative.
Hyperscaler AI capex is one of the strongest demand signals available. These companies have direct visibility into enterprise AI adoption through their cloud platforms. Their spending reflects what they see in their order books, not what they hope will happen.
Investment Characteristics
For investors, the infrastructure layer offers exposure to AI growth through established, profitable companies with proven business models. Key advantages:
- Existing revenue streams that fund AI investment
- Direct customer relationships with enterprises adopting AI
- Diversified businesses that reduce single-point-of-failure risk
- Track records of turning infrastructure investment into high-margin recurring revenue (as demonstrated by cloud computing over the past decade)
The hyperscalers have done this before. They invested heavily in cloud infrastructure when it was unproven, faced scepticism about returns, and ultimately built some of the most profitable business segments in technology. AI infrastructure follows the same playbook.
Related
- AI Stack Framework — The five-layer model and how infrastructure fits within it
- The Models Layer — The frontier AI companies that depend on this infrastructure
- Inference Scaling — Why growing inference demand sustains infrastructure investment