Topic: Infrastructure
140 stories
- Tinybird distills years of lessons from running ClickHouse at scale
- PlanetScale launches Neki, a sharded Postgres database
- OpenJDK's JEP 544 cuts Java startup time up to 80% with AOT caching
- ON.energy pitches medium-voltage UPS fix for AI data center outages
- Google signs 22-year nuclear power deal in €13bn Finland AI investment
- Samsung shows zHBM stacking memory on AI chips, claims up to 8x performance
- GitHub faces new Git rivals built for AI coding agents
- Patagonia draws AI data center plans, including a 500 megawatt site
- Miles v0.1 ships as open-source RL stack for frontier post-training
- ASML locks in TSMC, Samsung and Intel for High-NA EUV lithography
- TeraWulf's $3.2bn AI data center exposes gaps in fire safety
- Crawler bots cost git.kernel.org more CPU than real traffic
- Claude Code, Instinct run mobile agents in Firecracker VMs
- Broadcom pulls VDDK downloads, blocking VMware migration tools
- Arm unveils Mali G2-Ultra NX, its first AI-native mobile GPU
- NetBSD 9.5 released, ending support for the 9.x branch
- An ARM64 hypervisor bug shows the NX bit isn't just about security
- Imbue launches Cloud in a Bottle, an open-source personal cloud
- Val Town demos 3,613 app-to-app OAuth connectors via MCP
- Micron-backed report: AI inference makes memory and storage the bottleneck
- Deepseek plans 160,000-chip Huawei cluster in Inner Mongolia
- AMD's Threadripper Halo packs 576GB of HBM3e for local AI research
- OpenAI, Anthropic and xAI go down together, no shared cause
- Cerebras adds Qwen 3.8 27B to public API at ~1500 tokens/s
- Wasmi 2.0 ships a 2.2x faster WebAssembly interpreter
- IBM ships Granite Time Series models on Confluent Cloud
- Cloudflare prototypes Zstandard cache transcoding to save petabytes of storage
- Startups pay you to rent out spare compute for AI inference
- Baseten maps the efficient frontier of LLM inference serving
- ravynOS builds a pre-alpha, open source alternative to macOS
- Nvidia invests $3.5bn in MediaTek to lock in NVLink Fusion
- Cheap GPS jammers are creating navigation dead zones worldwide
- VMware set to lose its 20-year virtualization lead as deadlines loom
- SpaceX builds in-house foundry to cast gas turbine blades for AI power
- OpenAI and rival labs buy tens of thousands of Mac minis for agents
- NAT's 1994 quick fix for IP scarcity reshaped the internet
- AI crawlers now consume 20% of git.kernel.org's CPU
- Virtual power plants pay you to let a utility adjust your thermostat
- Nvidia extends its AI edge beyond GPUs into data orchestration
- FreeCORE continues TrueNAS CORE, ships stable 15.0-U1
- AWS's network redesign is up to 40% more energy efficient, but keeps the savings
- Artificial Analysis benchmarks small AI models on iPhone 17 Pro
- The Twelve-Factor App (2025) resurfaces on Hacker News
- Samsung's LPDDR5X-PIM does math inside DRAM, software isn't ready
- Monzo built a full backup banking platform on GCP
- Meta tests robots to automate data center maintenance
- Nvidia and Cerebras tout inference speeds that won't scale
- Cloudflare cuts DNS cache memory by over 50%, frees 100TB
- atproto answers X's Nitter crackdown by sharing the database
- Meta's MTIA 400 chip trains AI models and serves ads
- Amazon triples Nvidia GPU orders with 2 million more chips
- OpenAI's Jalapeño chip beats Nvidia Blackwell in early lab tests
- Hugging Face details hybrid search built for Papers with Code
- walgit turns Git hosting into one binary over object storage
- SiFive's BigSky SF-2U870 brings RVA23 support to RISC-V servers
- Jabber/XMPP turns 25 with an essay arguing it beats Matrix on openness
- IBM announces chip that runs Arm and Z instructions together
- Cerebras unveils CS-4, doubling CS-3 performance on the same chip
- AIREP protocol logs AI governance decisions as signed, tamper-evident records
- AI data centers drive investment in solid-state power transformers
- Xiaomi's Xring O3 chip matches Apple cores single threaded, much faster multithreaded
- Shipyard winds down its IPFS work as Protocol Labs ends funding
- SELF replaces ELF with a SQLite-based executable format
- Nvidia outlines CUDA support requirements for RISC-V
- TigerBeetle tests replica internals with protocol-aware DST
- Qwen3.6-27B tests show attention backend and quantization change output
- FlashPrefill V2 speeds up long-context LLM prefill up to 47x
- Bluesky releases atproto spaces alpha for non-public data
- Waymo unveils custom 5nm ASIC for self-driving compute
- OpenTelemetry's slow releases traced to maintainer scarcity and rigid stability rules
- Nari Labs hits sub-50ms TTS latency on a single H100
- Liquid AI ships DSpark draft models for 3.2x faster LFM2.5 inference
- Chinese AI firms build their own data centers in Inner Mongolia
- Waymo designs a custom robo-taxi chip to stay ahead of Tesla
- SpacetimeDB's launch benchmarks are misleading, review argues
- GitHub says capacity failures caused its second August outage
- SondeHub: a joke balloon tracker caught up in war and the military
- Inco AI's DFlash 2 lifts LLM decoding output 16-25%
- GrapheneOS says Google violates GPLv2 with Google Drive source delivery
- machine0 launches CLI-driven CPU and GPU VMs for AI agents
- FreeToken serves 753B-parameter GLM-5.2 on one workstation GPU
- DDR5 memory prices climb near 500% in a year on AI datacenter demand
- Cerebras unveils CS-4, claiming inference up to 30x faster than GPUs
- Speko launches a router for voice AI models
- Siemens and Reinhausen build 800 VDC transformer for AI racks
- OpenAI signs 20-year Ohio data center lease with Nvidia backing up to $105 billion
- DuckDB previews v2.0 with server mode, new SQL parser, 40x speedups
- Constraint-aware GPU scheduler beats FIFO by 33 points
- 1872 launches robotic factory for automated steel fabrication
- Engineer rebuts RISC-V critic with developing-world cost math
- AWS rations CPU cycles as agentic AI strains server capacity
- RISC-V's ISA design gets a detailed technical critique
- Go, Kotlin, and Erlang: three ways to switch tasks safely
- DuckDB adds async I/O, up to 3.7x faster on S3
- OpenAI previews Ultrafast tier for GPT-5.6 Sol, up to 14X faster
- Kog aims to speed up LLM inference on existing GPUs
- systemd-journald writes 55KB+ to disk per log line
- Oxide ships Kubernetes integrations shaped by customer needs
- Bluesky ships Jetstream v2 with Network Replay and new SDKs
- AWS's Strands Robots syncs robot data through Hugging Face Storage Buckets
- Nebius plans over 1 GW of new GPU capacity a year from 2027
- Hazy Research says AI agents are retiring CUDA kernel DSLs
- OasisKV boosts LLM inference throughput with lookahead KV cache prefetching
- Data-Centric Parallel method promises up to 2.88x faster long-sequence training
- Amazon's Texas data center plant permitted to emit 33 million tons of CO2
- Snowflake pushes Postgres CDC into Iceberg with data mirroring
- AT Protocol lays out how its decentralized backend scales
- Shopify replaces Redis with MySQL for inventory reservations
- RFC 10023 defines a _for-sale DNS record for domain sales
- Pinterest traces Ray training crashes to zombie memory cgroups
- Nvidia and Amazon pour billions into AI power infrastructure
- Brevis compresses model checkpoints by treating compression as program synthesis
- A developer turns his CMF Phone 1 into a production server
- Samsung, SK Hynix and Micron reportedly sell out 2027 memory capacity
- pgrust v0.2 claims 300x Postgres speed on analytics
- Databricks shares how it cuts AI coding costs at scale
- Cloudflare ships Kitesurf, a browser built for AI agents, not Chromium
- Naïve raises $28.5M Series A to let AI agents run a company
- Inside vLLM: anatomy of a high-throughput inference engine
- Baseten joins Hugging Face Inference Providers
- AMD acquires Taalas to etch AI models directly into silicon
- NVIDIA's Vera whitepaper overstates its lead over x86 rivals
- Deno ships celld, a self-hosted alternative to Cloudflare Durable Objects
- Dev runs TinyStories LLM on a $10 ESP32 microcontroller
- Anthropic signs $10B deal with AI cloud startup Volta
- OpenAI details GPT-Live, its full-duplex voice architecture
- Cloudflare quantizes Kimi and GLM caches for 41% faster inference
- "200 Milliseconds" traces one HTTP request from click to render
- Topology-aware routing cuts LLM KV cache transfer latency up to 18x
- NixOS-DGX-Spark brings Nix and NixOS to NVIDIA's DGX Spark
- The Catalyst team finds it built a general JAX-to-LLVM compiler by accident
- AMD MI355X beats Nvidia B300 on cost per GPU running Kimi K3
- A home NAS builder computes real drive failure odds from Backblaze data
- Nvidia Vera CPU: 88 custom Olympus cores challenge Intel, AMD
- Cloud infrastructure spending hits $143bn a quarter, fastest growth in eight years
- OpenAI, Anthropic, Google seal AI sessions behind encrypted state
- MCP goes stateless in its largest update since launch
- GPU idle time, not model size, is becoming AI's real cost problem
- Ai2 launches OlmoEarth Platform for planetary-scale satellite inference
- OpenAI's Michigan data center draws thousands of electricians