Aurelia Han & Siddharth Ramakrishnan — Scale Venture Partners - The Big Data Center Squeeze: Getting More Compute From Existing Infrastructure - March 2026
FreeGetting more compute from existing data center infrastructure
FreeFree tier
About Aurelia Han & Siddharth Ramakrishnan — Scale Venture Partners - The Big Data Center Squeeze: Getting More Compute From Existing Infrastructure - March 2026
A blog post by Aurelia Han and Siddharth Ramakrishnan of Scale Venture Partners analyzing the data center capacity squeeze caused by surging AI inference demand and hyperscaler capex constraints. It argues that new software improving GPU utilization (tokens per GPU) and facility power efficiency (tokens per watt, GPUs per megawatt) will be critical to extract more compute from existing infrastructure, rather than relying solely on new data center builds.
Key Features
Focus on inference workload growth (33% of AI compute in 2023 to 67% in 2026)
Analysis of hyperscaler capex commitments ($200B Amazon, $180B Google, $135B Meta, $120B Microsoft)
Identifies two software opportunity areas: utilization/throughput (tokens per GPU) and facility efficiency/power delivery (tokens per watt, GPUs per megawatt)
Covers constraints: hardware shortages, energy supply, regulatory challenges
Authored by Aurelia Han and Siddharth Ramakrishnan, published March 2026