Blog

Introducing The Hidden Supply Podcast

AI Infrastructure, Exposed

July 08, 2026

|Written by Elan Neiger

  • Cost per Token
  • Distributed Inference
  • GPU Utilization
  • Inference Operating System
  • Inference Orchestration
  • LLM Serving
  • Token Factory

The Hidden Supply Podcast explores the constraints shaping AI infrastructure: compute, energy, capital, data centers, and the production systems needed to make AI work in production.

In our first episode, Why AI Factories Are Bleeding Money, NeuReality CEO Moshe Tanach sits down with Varana Capital CIO Ezra Gardner to unpack the economic reality behind the AI infrastructure boom. Demand is growing fast, but GPUs, power, space, and capital are finite. The result is a market where buying more capacity is no longer enough.

They discuss why installed GPUs do not automatically translate into usable AI supply, why GPU utilization has become a business issue, and why AI changes the software economics that defined the SaaS era. The conversation moves from energy scarcity and data center buildouts to heterogeneous compute, open models, inference as a service, and the operating layer needed to turn expensive infrastructure into efficient production inference.

Read Next

  • Reproducing (and Beating) a Published Inference Benchmark: A DeepSeek V4 Pro Debugging Story

    • Cost per Token
    • Distributed Inference
    • GPU Scale-Out
    • GPU Utilization
    • Inference Networking
    • Inference Operating System
    • Inference Orchestration
    • LLM Serving
    • Token Factory
    Read More
  • AI Coding Agents Are Exposing the Missing Operating Layer in Production AI

    • Cost per Token
    • Distributed Inference
    • GPU Utilization
    • Inference Operating System
    • Inference Orchestration
    • LLM Serving
    • Token Factory
    Read More
  • Why Standard NICs Fail AI Inference — And What an AI NIC Does Differently

    • AI NIC
    • AI-SuperNIC
    • GPU Scale-Out
    • In-Network Compute
    • Inference Networking
    • LLM Serving
    • SuperNIC
    • Ultra Ethernet
    Read More