Executive Briefing: Your Cloud Provider Is Your Competitor for AI Compute

Source: Nate’s Newsletter Date: 2026-02-08 URL: https://natesnewsletter.substack.com/p/executive-briefing-the-global-inference

Summary

The piece argues that enterprises face a structural conflict with their cloud providers: hyperscalers (Google, Microsoft, Amazon) will prioritize their own AI products over customer workloads when GPU compute is constrained. The supply case is concrete — DRAM contract prices rising 90–95% in a single quarter, new fabrication capacity not arriving at scale until late 2027, and agentic workloads (AI calling AI in loops) driving demand that traditional capex models weren’t built for. Google’s own consumption is cited at 1.3 quadrillion tokens per month, up 33% in months.

Implications

• Feeds the inference infrastructure thread: the 2027 capacity timeline and the hyperscaler conflict-of-interest are the two facts most worth tracking — they set the floor for when cost curves can plausibly inflect downward.
• The agentic-loop demand multiplier is underappreciated in most capacity planning; this is an early, data-anchored articulation of why agent architectures hit infrastructure constraints faster than single-turn workloads.
• For any team running inference at scale: committed-use agreements negotiated before the agentic demand signal was visible may already be mis-sized, and the geopolitical concentration (Taiwan, South Korea, Netherlands) adds a non-linear risk factor that doesn’t appear in standard vendor SLAs.