Hardware chosen with local AI in mind — not bolted on after the fact.
Running AI models on your own hardware is a fundamentally different exercise than calling a cloud API. It lives or dies on GPU memory, sustained thermal performance, and system memory — the same properties that define whether a system is good at anything else demanding.
Why GPU memory is the real constraint
Local model execution is bounded by how much a model (and its active context) can fit in GPU memory, not just raw compute throughput. This is exactly why VRAM capacity is one of the headline numbers in every ER capability rating, not an afterthought spec.
Privacy and control
A model running on hardware you own processes your data on that hardware — it doesn't leave the machine unless you choose to send it somewhere. That's a meaningfully different privacy posture than a cloud-only workflow, and it's a real reason some workloads belong locally regardless of raw performance.
Hybrid local/cloud is normal, not a compromise
Most real workflows mix local and cloud intelligence — local for latency, privacy, or offline reliability; cloud for scale or specialized capability the local hardware can't match. ER systems are built to be strong local participants in that mix, not to pretend cloud doesn't exist.
Why tiers exist
Local AI capability scales with VRAM and platform class — a Tier I system handles smaller local workloads capably; a Tier III system with more GPU memory handles meaningfully more. Rather than publish unverified model-size promises, every ER product page shows a descriptive capability rating (Capable / Advanced / Extreme / Flagship) grounded in the actual hardware, not a marketing claim.
The ASH A(π) distinction
ASH A(π) is the only system in the launch catalog built around native ASH intelligence from the platform up — 96GB of ECC GPU memory and a workstation platform specified specifically for that role, not a conventional system with AI software added on top. Every other ER system is genuinely AI-capable; ASH A(π) is the one built natively for it.