Topic: Infrastructure
232 stories · page 1 of 5
What AI runs on: chips, data centres, cloud capacity, power and networking. Nvidia, AMD, TSMC, the large cloud providers and the companies building compute.
Inference engines belong here as well, along with anything else that moves the cost of a single model call.
- WeaveGrid and Veir say AI power demand could modernize the US grid
- Solid-state transformers could ease the grid transformer shortage
- Moxnox 1.0 released: hobby OS with NT syscalls, POSIX and btrfs
- Micron drops 4F² DRAM after two decades, sticks with 6F² as rivals push ahead
- Ukraine says Starlink and Starshield speeds are restored after throttling
- DreamHost resets some email and FTP passwords during File Manager and Webmail incident
- WD says AI inference leaves a hidden storage bill, in a sponsored piece
- Hugging Face libraries now run natively on AWS Trainium via TorchNeuron
- Cerebras explains how Kimi K3 (~3T parameters) maps onto 59 WSE-3T wafers
- rqlite creator traces twelve years of distributed SQLite
- Google announces Agent Compute Engine: one microVM per AI agent
- Nvidia's Huang unveils DSX Pod, a modular data center ready in 14 days
- DuckDB 2.0.0 LTS adds Quack server protocol, async I/O and triggers
- DeepSeek open-sources DeepInfer inference framework for V4.1-Flash
- OpenBSD 8.0 released with a 512G amd64 kernel address space
- Arch Linux ARM runs on ASUS ProArt P16, shows 116 GiB to CUDA
- Ukraine says sudden Starlink speed limits disrupted key drone missions
- Kubernetes through a skeptic's eyes: control loops and Services carry the value
- US nuclear uprates could add 3 to 8 GW, a fraction of AI demand
- Ricursive Intelligence founders pitch AI-designed chips to ease the compute crisis
- Oxide's Voxel emulates a full Oxide rack on a single machine
- APNIC: fishing and anchors cause most undersea cable faults
- Fervo Energy's Utah geothermal plant starts selling power to the grid
- Anthropic moves its monorepo off git to in-house Scion in five months
- Microsoft announces Maia 300 AI chip, rolling out in 2027 to cut reliance on Nvidia
- Cheaper AI tokens are driving more demand, Jensen Huang's best-case scenario
- Unikernels are practical again because AI ports the libraries, one post argues
- DuckDB 2.0 alpha reads S3 Parquet 2x to 3x faster in laptop test
- Ukrainian drones knock out two of Yandex's five data centers
- Amazon says it will stop using NDAs in data center deals with local governments
- Ai2 replaces its GPU priority scheduler with time budgets and fair-share
- TokenRouter claims 2.01-64.15x higher decoding throughput for token-level LLM routing
- DuckDB's DuckLake extension reads and writes an open SQL and Parquet lakehouse format
- TidesDB ships TideSQL, a storage engine plugin for MySQL 9.7 and 26.7
- openTPU, an open-source AI accelerator built by AI, runs ten models on an FPGA card
- Page tables can eat tens of GiB of RAM: kernel and database cases
- Go garbage collector pause hits 40ms when its metadata is swapped out
- Bad PDF redaction exposes Google's Nebraska data center water and power use
- Post urges default hard budget caps on APIs as AWS, Google Cloud add limits
- FTL presents a cloud OS that runs a userspace OS in each container
- Cloudflare opens Artifacts beta and a contest to build the next Git platform
- Startups put small batteries in e-bike cabinets and stoves to help the grid
- Sponsored MIT Technology Review Insights report says enterprise AI needs process redesign first
- Cloudflare opens OHTTP Gateway closed beta, renames Privacy Gateway
- Amazon pledges over $1B to data center communities, critics push back
- turbopuffer v3 makes vector ANN a secondary index
- Olmo-core 3 released: open MoE training stack benchmarked past 1 trillion parameters
- Google launches Suncatcher TPU satellite, says Starship needs about 1,800 launches
- Cloudflare K2 enters public beta as serverless event streaming on R2
- HPE-sponsored piece says enterprises should treat AI capacity as an asset