Reading up on infrastructure

GitHub is rewriting its Git storage architecture, replacing the Spokes system with Azure Blob Storage to achieve a 35× write performance boost for agent-driven traffic.

Ethiopia has sustained over 8% annual real GDP growth since 2008 despite ongoing wars, driven by state-led infrastructure, hydropower, agricultural mechanization, and mining expansion.

The article explains how Hetzner Cloud’s network stack evolved from basic Linux bridges to Open vSwitch, enabling Private Networks and Cloud Firewalls for scalable, secure tenant isolation.

Texas halted new data‑center permits amid a flood of speculative filings, grid‑reliability worries, and disputes over who should pay for costly transmission upgrades needed to support AI‑driven load.

Hundreds of new data‑center projects are being launched nationwide to meet AI computing demand, and these facilities are rapidly expanding in size and scale.

celld is a self‑hosted, distributed implementation of Cloudflare Durable Objects that lets Workers apps run unchanged while storing state in a user‑owned bucket at dramatically lower cost.

The piece claims HBM is a costly mistake that will be replaced by optically linked LPDDR memory pools, forecasting DRAM stocks to double or triple before a eventual 70% drawdown.

Muse’s browser uses near‑stock Chromium driven by an 18 MB Rust broker that provisions warm VM pools, residential IP egress, and accessibility‑tree clicks for reliable automated browsing.

Vint Cerf chose a 32‑bit IPv4 address space in 1977, assuming 4.3 billion addresses would suffice for an experiment, but the protocol’s longevity exhausted the supply decades later.

Elon Musk said SpaceXAI will add 660,000 AI GPUs this year, reaching about 1.44 million total, while building a 1.2‑gigawatt power plant to run them.

Estimating Meta’s Muse agent for 100 M DAU, the analysis finds ~0.1 GW CPU/VM power, ~1‑2 GW inference power, requiring ~$800 M CPUs, ~$2 B DRAM, and up to 3‑4 GW total.

S3's architecture is built for disk-era limits, but with cheap SSDs and fast networks, a new SSD-based disaggregated storage service could replace it — cloud providers lack incentive.

ClickHouse Managed Postgres enables strict memory overcommit so that out‑of‑memory conditions fail a single query instead of triggering an instance‑wide OOM restart, with no measurable throughput loss.

This piece describes a technique or system change that freed an additional 100TB of RAM in data-center operations.

Neoclouds require a tiered storage architecture—flash for active training, cost-effective capacity for other stages—to avoid hyperscaler lock-in via data gravity.

Short & Long Tail presents a series of essays that examine everyday topics—cars, hurricane names, maps, water transport, and WWE moves—by tracing their common origins into obscure, fascinating details.

The article argues that in distributed systems, timestamps alone cannot capture causality, showing how approvals, messages, and actions can refer to outdated states despite correct delivery.

Mailflare is a self-hosted email inbox for custom domains, built on Cloudflare, that handles sending, receiving, and routing with its own D1 database and R2 storage.

A GitHub cleanup job saturated a database primary because its safeguard monitored only replica lag, not primary load, causing site-wide errors.

When machine intelligence becomes abundant, the bottleneck shifts from AI capability to enterprise architecture, requiring systems designed for AI as a first‑class participant to unlock lasting value.

The article argues that New York subway’s constant decay and patchwork repairs unintentionally create an ever‑changing underground art exhibit dubbed the MTA Modern.

TigerBeetle’s protocol-aware deterministic simulation testing validates safety and liveness invariants at each replica, enabling deep verification of consensus and storage beyond black‑box methods.

The article argues that creating reliable backups is deceptively complex, requiring careful planning, testing, and ongoing maintenance beyond simple copy‑paste solutions.

By reusing TLS connections with Undici keep‑alive, the webhook sender eliminated most handshakes, cutting latency and saving over $2,000 monthly.

Statichost.eu offers fully European-owned static site hosting, from git deploy to CDN, targeting developers who want infrastructure located and owned in Europe.

Landfills, when properly managed, are cleaner, cheaper, and more environmentally sensible than recycling most everyday materials, except metals, according to lifecycle analyses.

Cloudflare rebuilt the Workers module registry to use URL-based resolution, lazy compilation, shared caches, and Node.js-compatible module handling.

The TSA is considering allowing passengers with TSA PreCheck to bring bottled water through security, ending a two‑decade‑old liquid ban.

The essay contends that the real AI risk lies in industry shifting costs to the public, urging accountability for today’s AI companies rather than focusing on hypothetical superintelligence.

New York’s subway system, with 47,000 openings, is increasingly prone to flooding as heavy rains overwhelm its aging infrastructure across the city.

After 25 years, Ground Zero remains underdeveloped despite the World Trade Center's success, and we should transform it into a vibrant neighborhood.

ovlive provides a real-time map showing all buses, trams, metros and trains across Belgium, operated by De Lijn, STIB-MIVB, TEC and NMBS/SNCB, with regional access links.

AI datacentre expansion will consume massive resources, trigger societal collapse, and render money obsolete as shared values shift beyond capitalist constructs.

The author argues that massive AI investment yields minimal profit while consuming excessive power, water, and hardware, calling the boom a toxic, self‑reinforcing aberration.

Route 53 Files exposes AWS Route 53 hosted zones as an NFS file system, allowing standard Unix tools to edit DNS records with near real‑time synchronization.

Tailscale lets you securely connect any internet resource using WireGuard‑based, zero‑trust, end‑to‑end encrypted tunnels that are fully configurable across any infrastructure.

Vercel's Fluid compute architecture dynamically assembles tailored machines for builds, sandboxes, and functions, enabling instant provisioning and efficient resource use.

One person died, about 15 may be missing and dozens evacuated after a flash flood destroyed bridges and knocked out the Grand Canyon's South Rim water pipeline.

The article traces how power and telegraph wires have been depicted in Japanese and global art, showing their evolution from symbols of progress to romantic, aesthetic subjects.

Store and Forward enables networks to treat waiting as an operation, allowing communication across time-varying links by storing data until future connectivity appears, extending connectivity beyond constant paths.

SpaceX's Starbase expansion in Los Angeles aims to boost local aerospace industry but faces regulatory and community hurdles.

Cloudflare cut over 100 terabytes of RAM used by its 1.1.1.1 DNS cache by redesigning cache layout and eviction policies without hurting query latency.

walgit is a git server implemented as a single binary that runs in front of an object store (S3/GCS), requiring no local state beyond a cache.

American opposition to data centers is driven less by their physical impacts and more by distrust of AI, big tech, and elites, fueled by bandwagon effects and rising AI visibility.

Vedansh, a backend and AI engineer, builds storage engines, network stacks, and real-time media systems from scratch, grounding his work in academic papers and AI/ML research.

Solid-state transformers, spurred by AI data center demand, offer mass-manufacturable, efficient power conversion that could also benefit EV charging and household grids.

The AI boom mirrors 2008's housing crisis as take-or-pay compute contracts create a 2027-2028 'reset wall' where payments start despite labs' revenue being insufficient to cover obligations.

The article explains how Kubernetes uses liveness, readiness, and startup probes to monitor container health and manage pod lifecycle effectively.

GitHub’s eight‑hour outage shows that reactive scaling fails when hidden concurrency limits are hit and client retries amplify demand, proving that agents need proactive traffic pacing.

Loudoun County, Virginia, earns substantial tax revenue from over 250 data centers, but officials and residents warn that reliance on this income may be risky.

Cursor argues Git’s packfile‑based design hinders scalable hosting and presents Continuity, a S3‑backed write‑ahead log system that enables linear, consistent replication.

The piece outlines a six‑stage AI infrastructure journey, maps three predictable cost/latency crises, and demonstrates how ACE Gateway’s control‑plane features cut spend and prevent hardware failures at each stage.

Cursor explains why Git's packfile design hinders scalable hosting, reviews past solutions like Spokes, and introduces Continuity, a S3‑based write‑ahead log system that provides linearizable, elastically scalable Git storage.

At least 10 new skyscrapers, including five supertalls, are planned for Miami over the next five years, raising concerns about whether such rapid vertical growth is sustainable or wise.

The primer explains how data center power demand from AI is straining the US grid, details power plant economics and marginal pricing, and outlines the development process using Homer City as a case study.

U.S. airlines and airports are testing new passenger and baggage screening programs designed to significantly cut airport security wait times.

Kentucky’s data‑center sales‑tax exemption could cost the state over $1 billion in forgone revenue if awarded to just a few large, AI‑optimized facilities.

The article argues DuckDB outperforms SQLite for analytical queries due to its columnar storage and vectorized execution, recommending it for data analysis tasks over SQLite's transactional strengths.

The article explores the often-overlooked complexity of elevator systems, from scheduling algorithms to user experience.

The Times investigation reveals Oracle incurred massive debt to build global AI data centers, with repercussions extending beyond the company itself.

Cursor rebuilt its monorepo to support cloud agents by matching local dev, simplifying the interface, and adding a self‑healing environment, boosting agent‑authored PRs from 10% to over half.

AI chip count will double roughly every nine months, reaching about 200 million H100‑equivalent units by 2028 as massive data‑center construction ramps up worldwide.

The article explains the foundational concepts of networking and the Internet, breaking down protocols, layers, and principles from first principles to build intuitive understanding.

Claude surpasses a compiler by enabling engineers to make cross‑stack decisions—from strategy to machine code—through iterative agent loops, turning vibe‑engineering into real engineering.

Roblox is pairing its game engine with a video world model (Super Upsampler) to deliver photorealistic, scalable multiplayer games while solving latency, consistency, and creator‑control challenges.

Cloudflare launches EmDash, a serverless TypeScript CMS built on Astro that sandboxed plugins via Workers isolates to solve WordPress security issues while offering MIT licensing and x402 payments.

Designing reliable job queues requires careful handling of ordering, duplicate processing, failure retries, backpressure, and visibility, revealing many hidden pitfalls that developers often underestimate.

Cloudflare Research introduces Meerkat, a global consensus service using the QuePaxa algorithm to provide strong consistency and high availability for control‑plane data across its 330+ data centers.

Mozilla's Current AI project maps the open source AI stack, evaluating 24,626 projects to identify gaps and seeking collaborators to close them.

Europe's new smart border system (EES) failed at launch due to poor planning, technical glitches, and inadequate testing, causing major delays.

OpenAI unveiled its first custom inference chip, Jalapeño, designed with Broadcom and assisted by OpenAI's own models, claiming better performance-per-watt.

Co-locating workflow state and application data in Postgres enables atomic transactions that provide exactly-once execution without application-level idempotency logic.

Vercel now supports deploying any HTTP server from a Dockerfile on Fluid compute with autoscaling and CPU-based pricing.

1970s environmental regulations, not geography or transportation, halted U.S. urban land reclamation despite high land values and housing needs.

Rural areas in the southern United States are expected to host the most planned data centers, despite their noise and power demands.

SpaceX plans to deploy Starmind, a constellation of one million AI compute satellites in orbit, aiming to make terrestrial data centers obsolete.

sandboxd is an open-source backend that creates isolated Linux containers with AI coding agents and live preview URLs, all self-hosted on one machine using Docker and Go.

The U.S. government is investing $150 million in technologies that give buildings immune-like systems to fight airborne diseases, following the pandemic.

SpaceX's Starlink generated more revenue in 2023 than all other commercial space companies combined, demonstrating the financial power of vertical integration and rapid iteration over traditional aerospace contracting.

Kubernetes is overly complex for small teams, requiring numerous add-ons and frequent upgrades that outweigh its benefits.

Residents near AI data centers report health problems from constant low-frequency vibrations, highlighting an overlooked cost of infrastructure buildout.