Good question! Here's what I measured today:

Minimum spec — witness only

• CPU: 2 cores is enough. hived's block processing is essentially single-threaded.
• RAM: 8 GB comfortable, 4 GB workable — with chain state on disk. If you put state in RAM
(/dev/shm) you need 30–40 GB, because the state alone is 9–25 GB.
• Disk: NVMe SSD, 64 GB comfortable (128 GB generous). Not an HDD — the state file is
memory-mapped with random I/O.
• Network: always on. No static IP and no port-forwarding needed for a pure witness; it only
makes outbound P2P connections. You only need an open port if you also want to seed.
• OS/clock: Debian/Ubuntu LTS, Docker, and NTP (chrony) — blocks are 3-second slots, so a
drifting clock misses slots.

** Why you'll see far bigger numbers quoted:**
Those are full node / seed specs. Holding complete chain history currently costs 549 GB of block log on a 650 GB disk — that's history, not consensus.

Can you run one from home? On specs, yes — an 8 GB mini-PC with a 1 TB NVMe beats the minimum.

The real constraint is availability, not hardware:

• A witness ranked outside the top 20 gets roughly one slot per hour. Every slot missed while your ISP or mains power is down is recorded publicly in total_missed and visible to voters.
• Initial sync pulls several hundred GB unless you bootstrap from a state copy or snapshot.
• Nothing may suspend or hibernate — laptops and power-saving desktops will miss blocks.
• Add a UPS and a wired connection if you care about your miss count.

Bottom line: 8 GB RAM, 2 cores, 64 GB+ NVMe, running 24/7 on reliable power and internet. 

For reference, a €8–10/month VPS (4 vCPU / 8 GB / 256 GB NVMe) already exceeds that — which is part of why most witnesses rent.

Self-hosting works; you're just trading ~€10/month for an availability risk that you'll probably see in your missed-block count.