Four companies outspend the US government on social services while V8 quietly doubles JSON.stringify performance.
> The AI infrastructure bill is now larger than most government budgets. And somehow, a JSON.stringify optimization might matter more to your daily work.
Meta, Microsoft, Amazon, and Alphabet have collectively spent $155 billion on AI infrastructure in 2025 — exceeding the entire US government budget for education, training, employment, and social services combined. Meta doubled year-to-date capex to $30.7 billion. Alphabet hit nearly $40 billion. Amazon topped them all at $55.7 billion.
The forward numbers are even wilder: Microsoft plans $100 billion next fiscal year, Meta $66-72 billion, Alphabet $85 billion, Amazon $100 billion. Combined, over $400 billion in AI capex — more than the EU's quarterly defense spending.
This is the industrialization phase. These companies aren't experimenting anymore; they're building power plants, pouring concrete, and buying every GPU they can manufacture. The question isn't whether AI will be transformative — it's whether the economics of this buildout make any sense at current revenue levels.
While headlines chase billion-dollar AI deals, Google's V8 team shipped something that will touch more production code this month than any foundation model release: JSON.stringify is now more than twice as fast.
This matters because JSON serialization sits in the critical path of nearly every web application. Every API response, every logging call, every cache write. A 2x improvement here compounds across millions of services.
The optimization targets one of JavaScript's most fundamental operations. If you're running Node.js microservices that serialize large payloads — think analytics pipelines, real-time dashboards, or API gateways — this is a free performance win on your next V8 upgrade.
The broader lesson: foundational runtime improvements often deliver more aggregate value than flashy new features. The V8 team's unglamorous optimization work benefits every JavaScript developer on Earth, which is roughly 20 million people.
Frigate.video — Open-source AI camera monitoring with local processing. No cloud needed, full object detection on-premises. Perfect for privacy-conscious home and business security setups. Growing fast on GitHub.
D-Wave Quantum AI Toolkit — Open-source tools for integrating quantum computing into ML training workflows. Still specialized, but the developer-focused approach signals quantum is maturing toward practical use.
Open-Source ERP (216 HN points) — Three years of solo development on a full enterprise resource planning system. Enterprise-grade features without licensing costs. The HN community response shows demand for open alternatives to SAP and Oracle.
We're watching AI infrastructure become the new utilities sector — massive capital expenditure, long payback periods, and a belief that demand will eventually justify the investment. Whether that belief is right determines if 2025's spending spree looks visionary or reckless in hindsight. Place your bets.
— Aaron, from the terminal. See you next Friday.
Compare three approaches to AI agent browser automation. Browser Use, Stagehand, and Playwright MCP tested with code examples, benchmarks, and architecture trade-offs.
AI EngineeringHow OpenClaw routes messages across Discord, Telegram, and Slack with an 8-tier priority cascade, then isolates agent execution in pluggable Docker/SSH sandboxes.
AI EngineeringSide-by-side comparison of how OpenClaw and Hermes Agent build system prompts, manage token budgets, and compress long conversations without losing critical context.
AI Engineering