Compute & Market Power
Chips, cloud capacity, data centers, energy, capex, and the supply chains that determine who can build AI.
The AI race ultimately reaches the physical world: accelerators, memory, fabs, networking, data centers, electricity, cooling, and the financing required to assemble them. This topic explains how those constraints shape model availability, cloud competition, national strategy, and the bargaining power of builders.
Coverage follows the stack from semiconductor materials and export controls to hyperscaler capex and grid queues. The goal is to make infrastructure legible as an operating constraint and a source of market power—not to treat every chip launch or data-center announcement as an isolated event.
-
Intel's Agent Test Has a 50% Core-Coverage Gap
-
Reflection Beam: the Beta Migration Budget
-
Ghost Core's $3,499 local-agent test
-
Kolibri's FP8 Weights Fill About 98% of One H100
-
Claude India Inference Needs a 9.1% Budget Check
-
Lambda's 8.9% Larger Loan Does Not Prove Spare Capacity
-
Nebius's 10% GPU Saving Would Still Leave an 8% Rise
-
GMI Cloud's $668M Raise Still Needs a Delivery Test
-
Efficient's New Round Beats Its Series A by Over $37M
-
Samsung's $1B Is Under 10% of Helix's Launch Capital