The latest AI Infrastructure posts we have indexed from venture capital blogs and newsletters, plus the highest-scoring pieces of the past year and the people who write about it most.
Decision models—a new AI category that makes bounded choices rather than generating text—launched at $100M revenue in a week, then got replicated for $95 in open-source form within five days, signaling that software commoditization has reached a new inflection where defensibility now rests on integration and operations rather than capability alone.
Venture capital is shifting from software to physical technologies as AI collapses software development costs and the SaaS model loses defensibility; the strongest opportunities now lie at the intersection of AI-accelerated atoms—chips, energy, robotics, biotech—where capital intensity and technical moats remain defensible.
Software AGI has arrived as of November 2025, but only in coding; the real frontier is verification, not generation. Meanwhile, capability gaps in medicine and security reveal the pattern—models can do the work, but humans haven't scaled to review it.
As AI coding agents accelerate code generation, continuous integration infrastructure becomes a bottleneck—validating code faster and more reliably than legacy CI systems is now a critical competitive opportunity, with the market projected to grow 20%+ annually through the 2030s.
AI labs are spending billions on data for post-training and RL environments; data companies are hitting unprecedented revenue scales, but the market is lumpy, concentrated among four customers, and masks stark differences in unit economics and durability.
The dominant AI interface is shifting from conversation to delegation as personal agents like Claude Duet enter the market; simultaneously, infrastructure bottlenecks are moving from GPU scarcity to power grids and physical constraints, while agent adoption outpaces web security standards designed for humans and bots.
AI infrastructure investment is reaching revenue payback twice as fast as the telecom internet build-out did, reaching 10% revenue-to-capex ratios in four years versus ten. The risk is circularity—whether venture-backed AI companies spending on compute are just recycling capital—but venture-backed firms now represent a sixth of US employment, suggesting a genuinely large addressable market beyond startup-to-startup spending.
As AI inference demands explode while power grows scarce, specialized hardware architectures are resurging after decades of general-purpose CPU dominance—enabled by new software tools, easier chip development, and heterogeneous workload requirements that favor tailored designs over one-size-fits-all solutions.
Tomasz Tunguz (22) · Lightspeed Venture Partners (18) · Saanya Ojha (15) · Investing in AI (10) · Ed Sim (10) · A16Z (10) · Radical VC (8) · USV (6) · Newcomer (6) · Sequoia (5)
Suggest a blog or newsletter for the SandHill index. We track the RSS feed and feature the best pieces in the weekly digest.
The latest theses & analyses from 800+ venture investors, curated into one email. Join 5,000+ investors, founders and LPs.