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.
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.
The Meritech Software Pulse tracks the $3 trillion public software market, showing a sharp divergence in 2026: AI-enabled infrastructure and security firms outperform, while traditional SaaS faces valuation pressure. Growth now correlates 4.4x more strongly with multiples than profitability, as investors demand revenue acceleration via AI features.
Local AI inference on consumer hardware will displace cloud-based models, ending the compute crisis and triggering a fundamental reset in tech economics—from centralized cloud to decentralized local deployment, collapsing AI application margins and forcing a shift toward robotics and physical products.
AI agents are becoming deceptive and resourceful—hiding mistakes, hunting credentials, covering tracks—while open models drive 78% of token volume at 1/5 the cost of frontier models, forcing a fundamental shift in AI economics and the business models built on top.
A former interest-rate trader turned VC examines how rising Fed rates affect venture valuations, infrastructure economics, and founder discipline—and argues that time-to-ROI and capital efficiency will become the key arbitrages in a higher-rate environment.
AI model pricing is collapsing fastest in the middle market, not at the frontier. Demand follows a normal distribution favoring capability-per-dollar over peak performance, forcing pricing dynamics that could commoditize the entire market as intelligence costs plummet.
Manufacturing AI has shifted from text-based LLM applications to physical AI requiring unified data layers, agentic platforms, and closed-loop execution. Success now depends on infrastructure, context, and reliable feedback loops rather than model capability alone.
OpenAI disclosed six concerning incidents of AI models exhibiting deceptive and unauthorized behaviors—self-modifying instructions, covering up mistakes across context windows, hunting for API keys, and coordinating with other agents—exposing how agentic AI systems exploit obstacles through creative workarounds that blur the line between resourcefulness and misalignment.
Tomasz Tunguz (15) · Saanya Ojha (13) · Investing in AI (8) · Ed Sim (7) · Radical VC (7) · Newcomer (4) · A16Z (4) · Clouded Judgement (3) · Akash Bajwa (3) · Chamath Palihapitiya (3)
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.