#Ai Infrastructure

Your LLM Will Be Dead in 18 Months. Plan for It.

Frontier model lifespans have collapsed to about a year, and your eval suite is the only contract that survives.

Article · 1w ago1

OpenAI's Texas Letter Is Really About a Grid Freeze

Abbott froze data-center grid hookups, and the pledges buying them back will shape where GPU capacity lands.

Article · 2w ago3

NVIDIA's roadmap is a memory story, not a compute story

Read A100 through Rubin by bandwidth, number formats, and rack power, and the right 2026 rental choice falls out.

Article · 2w ago0

llama.cpp Finally Competes With Its Own Wrappers

With Hugging Face backing, the engine under Ollama and LM Studio now ships its own installer and model catalog.

Article · 2w ago4

The Real Math Behind Databricks' 70% AI Coding Savings

The headline number is compounded arithmetic, but the four levers underneath are a playbook worth stealing.

Article · 3w ago2

Anthropic Was the Last Chip-Agnostic Lab. Not Anymore

A $485K job listing and exploratory Samsung talks won't dent Nvidia soon, but the inference economics behind them will.

Article · 3w ago0

xAI Turned Unpermitted Gas Into a Compute Moat

Mississippi's turbine deal shows speed-to-power, not GPU count, now decides who trains frontier models.

Article · 3w ago1

The AI Power-Bill Map Is Really a Market-Design Map

PJM's capacity-auction receipts show why bills spike near data centers, and why 23 states are rewriting the deal.

Article · 3w ago5

GitHub Models Got 29 Days. Plan Your AI Stack Accordingly

Microsoft retired its free inference API with under a month's hard notice, paying customers included. That's the real lesson.

Article · 3w ago0

The LLM Router Was a 2024 Idea

Manifest killed its router after four months, and its post-mortem indicts the whole category.

Article · 4w ago0

Atomarine's Nuclear Data Centers Will Burn Gas First

The YC startup's real bet isn't reactors — it's shipyards, seawater cooling, and skipping the interconnection queue.

Article · 4w ago2

Microsoft doesn't need to beat GPT, just route around it

In-house MAI models are absorbing Copilot's everyday traffic while OpenAI keeps the frontier tier.

Article · 1mo ago1

The Real Cost of 'Just Use vLLM,' According to Netflix

Netflix's Model Runtime team details the version pinning, GIL workarounds, and preemption bugs behind production LLM inference.

Article · 1mo ago2

Google Split Its TPU in Two Because Agents Broke Inference

Cloud Next 2026's bifurcated silicon, flattened networks, and managed sandboxes reveal what agentic workloads actually demand.

Article · 1mo ago2

Hetzner Is Quietly Commoditizing LLM Inference

A free, OpenAI-compatible endpoint from Europe's cheapest host says more about inference margins than any benchmark.

Article · 1mo ago1