Hyperscalers Build AI Moats: Beyond LLM Hype to Physical Compute

By Korra · 1 September 2026

The AI race has quietly shifted from software to physical infrastructure. While retail investors chase LLM hype, the real battle is for compute power and the immense capital required to build it. This isn't just about code anymore; it's about who owns the foundational layers of the future.

Korra's read

My read is that hyperscalers and major tech firms are strategically deploying massive capital, estimated in the tens of billions, to construct an AI moat around physical compute and power. This isn't just about building better algorithms; it's about monopolizing the hardware cycle and infrastructure necessary to run them. They're effectively funding their own customers to ensure future AI development remains captive to their ecosystems, creating significant barriers to entry for smaller players.

The evidence

Apple's recent capital allocation signals this shift clearly. While exact capex deltas aren't publicly detailed, their R&D spending surged to 10.3% of revenue in fiscal Q2 2026, reaching $11.7 billion, almost entirely attributed to AI.

This investment is focused on next-generation custom silicon and 'Apple Intelligence' for on-device processing, rather than just expanding cloud services, demonstrating a silicon-centric approach to owning the hardware layer.

What to watch

To understand the deepening of this AI capital moat, watch for Apple's upcoming foldable iPhone and the new Siri, which are expected to showcase their intensified AI R&D. Any significant shift in Apple's capex towards building more proprietary data centers, moving away from its current hybrid approach with partners, would also signal further entrenchment. Ultimately, the market will be looking for a hardware upgrade cycle driven by these new AI-enabled devices to validate this capital-intensive strategy.

Information only, not investment advice.