Custom AI notes from the shop floor.
Practical writing on AI hardware, open-weight models, evaluation and the systems we build, from the engineers who build them.
From the bench.
Private AI Infrastructure for 10, 25, 50 and 100 Concurrent Users
What on-prem AI hardware you need for 10, 25, 50 and 100 concurrent users, with GPU, VRAM, memory, networking and power ranges for each tier.
Building an Enterprise AI Stack with vLLM, Ollama and Open WebUI
A technical guide to combining vLLM, Ollama and Open WebUI into one on-prem LLM stack, with GPU, networking and governance trade-offs for enterprise buyers.
How Much GPU Memory Do You Need to Run a 70B Parameter LLM
Work out how much GPU memory a 70B parameter LLM needs on-prem, with math for weights, KV cache, quantization, concurrency and headroom.
Where Mac Studios Fit and Fail for On-Prem LLMs
A technical look at using Apple Mac Studio for on-prem LLM work, including unified memory wins, throughput ceilings, and when a GPU server is the better buy.
How to Deploy Llama, Qwen and Mistral Models on Private AI Hardware
A practical walkthrough for deploying Llama, Qwen and Mistral open-weight models on your own AI hardware, from weights and quantization to serving.
Best GPU Server Configurations for Running Open-Weight LLMs Locally
Compare GPU server configurations for running open-weight LLMs locally, from single-GPU workstations to 8-GPU nodes, with VRAM, interconnect and power guidance.
Custom AI Hardware for Financial Services Data Residency and Access
How banks, asset managers and insurers use custom AI hardware to keep LLM workloads under their own data residency, encryption and access control rules.
How to Build a Private AI Server for Your Business
A step by step guide to building an on-prem AI server for your business, from workload sizing and GPU choice to networking, storage, software and rollout.
Implement Enterprise SSO Using SAML, OAuth and OpenID Connect
How to wire SAML, OAuth 2.0 and OpenID Connect into an on-prem LLM stack so logins, agents and APIs honor your existing identity provider.
On-Prem AI Hardware Statistics Worth Knowing in 2026
A reference of on-prem AI hardware statistics covering GPU memory, power draw, cooling, model sizes and adoption trends shaping buying decisions in 2026.
GPU Alternatives to NVIDIA for On-Prem LLMs
A buyer's tour of AMD, Intel, AWS, Google, Groq, Cerebras, SambaNova, Tenstorrent and Huawei silicon for running open-weight LLMs on your own hardware.
Sizing GPU Capacity by Team Size and Active Use
How to size on-prem GPU volume by seat count, concurrency and active use patterns, so your AI hardware matches real team load without overbuying.
What Breaks in the First 90 Days of On-Premises AI
The real failure modes in the first 90 days of on-premises AI, what each one costs to fix, and the scoping decisions that stop them happening.
How Many GPUs Will I Need for Custom AI
Work out how many GPUs your custom AI workload actually needs, from model size and concurrency to VRAM headroom, without over-buying hardware you will not use.
Buy or Lease GPU Servers for On-Premises AI
How to decide between buying, leasing or financing GPU servers for on-premises AI, with the tax, refresh cycle and utilization trade-offs that move the number.
Spec your system.
Tell us the models you want to run, how many people will use them and where the hardware should live. An engineer replies with a first configuration and the questions that decide the quote.