Reading up on ai-infrastructure

Since early 2025, international markets have outperformed the U.S. — emerging economies up 68%, Europe 45%, Japan 44% — driven by the global AI infrastructure supply chain and corporate reforms.

Google raises $80B in equity, including $10B from Berkshire Hathaway, to fund AI capex, signaling demand vastly exceeds expectations and capital is becoming a commodity.

SpaceX's $2T IPO lacks financial justification, but data centers in space for agentic AI inference could make the valuation plausible.

The AI infrastructure buildout could require $7.5 trillion in spending by 2030, creating trillions in enterprise value for neoclouds like CoreWeave and Nebius.

AI inference workloads demand unprecedented concurrency and throughput, exposing cloud storage limits that require a decoupled, software-defined layer to avoid catastrophic failures.

Musk's TERAFAB plan bypasses ASML's EUV bottleneck by using Intel GaN chiplets on mature nodes and advanced 3D packaging to achieve 1 TWh compute capacity.

Falling launch costs and rising terrestrial constraints will make space-based AI data centers cheaper than Earth-based ones within a decade, potentially creating a trillion-dollar annual CapEx market for inference workloads.

SpaceX, long resistant to an IPO, now races to go public to fund orbiting AI data centers and help Musk's xAI compete with OpenAI.

LMCache is a vendor-neutral KV cache management layer that reduces time-to-first-token and improves throughput for LLM inference by enabling persistent storage and reuse of cached states.