Home Assistant Minimum Requirements: What You Actually Need (2026)

Disclosure: Some links in this article are affiliate links. We may earn a commission on qualifying purchases at no extra cost to you.

Home Assistant's minimum requirements start at 2 GB of RAM: a 2 GB Raspberry Pi 4 or 5 with a 32 GB A2-class microSD card, or a virtual machine with 2 GB of RAM and 2 vCPUs. If the same box also runs a broader homelab stack, the host has to be sized for that stack, not for Home Assistant.

Storage and radios decide reliability long before the CPU does. Home Assistant's Recorder writes entity state changes to a database continuously, and that write stream is what wears out a cheap microSD card while the processor sits near idle. A USB extension cable that moves a Zigbee coordinator away from a USB 3 port resolves more "Home Assistant is unstable" complaints than a faster processor ever will.

Two claims circulate that the official documentation does not support. The first is that Home Assistant needs 8 GB of RAM, which the project's own Green appliance contradicts with 4 GB of LPDDR4X on board. The second is that Intel's 6 W N100 figure describes a mini PC's wall draw, which it does not.

What Home Assistant Actually Requires

Home Assistant publishes two floors, one for Raspberry Pi and one for a virtual machine, and both stop at 2 GB of RAM.

There is no official statement that ordinary Home Assistant requires 4 GB or 8 GB. The project's own appliance, Home Assistant Green, ships with 4 GB of LPDDR4X and 32 GB of eMMC, and that is the hardware the vendor considers mainstream. Everything above the 2 GB floor is headroom for other software, not a Home Assistant requirement.

What that means in practice:

• A 2 GB Raspberry Pi 4 meets the floor for a conventional smart home.
• A 4 GB board is the practical sweet spot if the same device also runs the usual add-ons.
• 8 GB or more only becomes rational when Home Assistant shares the host with virtualization, camera processing, frequent ESPHome builds, or a genuinely large Recorder database.

HAOS, Container, and the Installation Methods That Were Retired

Home Assistant OS is the recommended installation for most users. It bundles Home Assistant Core, the Supervisor, the operating system, and the add-on store into one image, which is why it installs cleanly on a Raspberry Pi or an x86 mini PC and then behaves like an appliance.

Home Assistant Container is the supported alternative for anyone who already manages Docker and Linux. It drops the add-on layer, so Zigbee2MQTT, Mosquitto, and Node-RED are containers you configure yourself rather than apps you install from a menu.

Two older methods are gone. Home Assistant Core and Home Assistant Supervised are no longer supported installation paths. Tutorials that walk through a Debian Supervised install are describing a method the project no longer maintains, and building new hardware around it is a dead end. Choose HAOS for the simplest path, or Container if the host is already a container server.

Light, Comfortable, and Heavy: Three Workload Tiers

The tier that matters is the one your full workload lands in, not the one Home Assistant alone would need.

Only the first row describes Home Assistant's actual software requirement. The second is what most engaged users end up running, and the third is where a mini PC replaces a Pi on merit rather than on impulse.

Add-ons do not all scale the same way. Mosquitto and Zigbee2MQTT are light. Node-RED is light to moderate and depends on flow complexity. ESPHome's service is light, but compiling firmware produces short CPU and memory spikes, which is why a low-memory Pi can feel slow during builds while automations stay perfectly responsive.

Raspberry Pi 4 and Raspberry Pi 5

The Pi 4 remains a genuinely capable Home Assistant host. It uses a Broadcom BCM2711 with four 64-bit Arm Cortex-A72 cores at up to 1.8 GHz, LPDDR4-3200 memory in 1, 2, 4 and 8 GB variants, Gigabit Ethernet, Wi-Fi 5, Bluetooth 5.0, two USB 3.0 ports, and a microSD boot path. The 1 GB variant sits below Home Assistant's 2 GB floor and should not be used for a new HAOS install.

The Pi 5 moves to a Broadcom BCM2712 with four Cortex-A76 cores and offers 4, 8 and 16 GB of LPDDR4X. It adds a PCIe connector that accepts an NVMe HAT, and it wants a 5 V/5 A USB-C supply for the full USB peripheral budget, though 5 V/3 A works with reduced current. Sustained load without cooling can push a Pi 5 into thermal throttling, so a case with real airflow or a heatsink is part of the build, not an accessory.

For Home Assistant itself, 4 GB on either board is comfortable. The 8 GB and 16 GB Pi 5 variants make more sense for compilation, multiple local services, or a deliberately future-proof node than for automations. If you are buying a Pi 5 specifically for HAOS, add the cooling and the SSD, then compare the total against an N100 mini PC, because a compact fanless alternative now sits in the same price class as a fully kitted Pi.

Home Assistant Green and the Yellow Reference Point

Home Assistant Green is the clearest evidence against the "you need a mini PC" claim. It runs a Rockchip RK3566 with four Cortex-A55 cores at 1.8 GHz, 4 GB of LPDDR4X, 32 GB of eMMC, Gigabit Ethernet, and two USB 2.0 ports behind a 12 V/1 A supply. Home Assistant rates it at 1.7 W idle and 3 W under load, which is lower than almost any x86 host can reach.

Green is built to behave like an appliance, and its microSD slot is for recovery only while live data sits on eMMC. Community installations well past 90 devices still describe the 4 GB platform as ample for normal Home Assistant work.

Home Assistant Yellow is useful mainly as a historical reference. It is built on a Raspberry Pi Compute Module 4, and Home Assistant recommends at least 2 GB of RAM plus 16 GB of eMMC, with CM4 Lite configurations requiring external NVMe storage. Production of the standard Yellow hardware has ended, so it belongs in the "how low the official floor can go" conversation rather than in a current buying list.

N100 and N150 Mini PCs

The Intel N100 is the community's default answer for a Home Assistant host with headroom, and the N150 is its mild refresh.

The two are the same performance tier for Home Assistant. Chassis quality, NIC choice, memory configuration, storage layout, and cooling matter more than the N100-to-N150 badge.

The 6 W figure is where most articles go wrong. It is Intel's processor base power, not the mini PC's wall consumption. RAM, SSD, NICs, voltage regulators, USB peripherals, and firmware settings all draw on top of it, so a small N100 box consumes meaningfully more at the wall than the processor line item suggests.

Memory deserves a footnote too. Intel specifies 16 GB as the official maximum for both parts on a single channel, while vendors such as GMKtec advertise 32 GB on some N100 systems. Treat the higher figure as a vendor-validated platform claim rather than an Intel specification.

Representative N100 and N150 systems for Home Assistant:

• Beelink Mini S12 Pro: N100, 16 GB DDR4-3200, a 500 GB M.2 2280 SSD, Gigabit Ethernet, Wi-Fi 6 and Bluetooth 5.2, with active cooling and dual HDMI. Check the Beelink Mini S12 Pro on Amazon
• GMKtec NucBox G3: N100 with one DDR4-3200 SO-DIMM, sold in 8 and 16 GB configurations, an M.2 2280 NVMe PCIe 3.0 slot plus an M.2 2242 SATA slot, and 2.5 GbE. The second M.2 slot and the faster NIC are what separate it from a generic N100 box. Check the GMKtec NucBox G3 on Amazon
• GEEKOM Mini Air12: available in N100 and N150 configurations with DDR5-4800 SO-DIMM memory, an M.2 2280 PCIe 3.0 x4 SSD, Gigabit Ethernet, Wi-Fi 6 and Bluetooth 5.2. A straightforward HAOS or container host. Check the GEEKOM Mini Air12 on Amazon
• Minisforum UN150P: N150 with a single DDR4-3200 SO-DIMM capped at 16 GB, M.2 2280 SATA or PCIe storage, a 2.5-inch SATA bay, and 2.5 GbE. The SATA bay is a rare feature at this size. Check the Minisforum UN150P on Amazon
• ASUS NUC 14 Essential (NUC14MNK): N150 in a barebones kit with one DDR5-4800 SO-DIMM up to 16 GB, M.2 2280 PCIe 3.0 and M.2 2242 SATA support, Wi-Fi 6E, Bluetooth 5.3, 2.5 GbE, and a 65 W supply. You supply RAM and storage and get the NUC ecosystem. Check the ASUS NUC 14 Essential on Amazon

The GEEKOM GT13 Max does not belong on this list. It is a Core Ultra 9 185H system with 16 cores and 22 threads, DDR5 support up to 64 GB, dual 2.5 GbE, Wi-Fi 7, and a 120 W adapter. It is a fine workstation or virtualization host, and it is unrelated to what Home Assistant needs. If you are weighing that class of machine, start from the Proxmox host guide rather than from Home Assistant.

For a deeper comparison of the low-power x86 tier, the N100 mini PC buying guide covers the field.

Refurbished Thin Clients

Corporate surplus is the cheapest way into a quiet, low-draw Home Assistant host, and it is the tier commercial comparison pages tend to omit entirely.

• HP t630: an AMD GX-420GI with four cores at roughly 2.0 to 2.2 GHz, two DDR4 SO-DIMM slots in dual channel, and platform support up to 32 GB. HP's published energy figures land at roughly 11 to 13 W at idle, about 1.2 W in sleep, and 0.2 W off. Its age makes it a poor fit for modern camera AI.
• HP t640: an AMD Ryzen Embedded R1505G with two cores and four threads at 2.4 GHz and up to about 3.3 GHz, dual DDR4 channels, M.2 storage, and Gigabit Ethernet. Capable enough for HAOS plus a few services, and notably efficient as a headless host.
• Dell Wyse 5070: Intel Gemini Lake, usually a Pentium Silver J5005 with four cores. Two SO-DIMM slots, passive cooling, and official memory support of 8 GB. Its M.2 slot takes SATA drives, so do not assume NVMe compatibility.
• Fujitsu FUTRO S740: a Celeron J4105 with four cores, typically 4 GB of memory and an 8 GB documented ceiling. One of the most efficient basic hosts available second-hand, particularly in European surplus markets.

Older thin-client CPUs are still fine for ordinary Home Assistant. They are the wrong foundation for Frigate enrichments that need newer instruction support, which is the one workload that moves you onto an N100-class system. If you want the cheapest path, the budget mini PC picks cover this end of the market.

NAS and Shared Hosts

Running Home Assistant on a NAS that is already on 24/7 is often sensible, and buying a NAS just for Home Assistant is not.

A Synology DS224+ makes the point. It ships with 2 GB of DDR4 that expands to 6 GB officially, two SATA bays, a Celeron J4125, and dual Gigabit Ethernet. Stock memory technically equals the HAOS virtual machine minimum, but leaves no useful margin once DSM itself is running. Synology lists it at 14.69 W during access and 4.41 W with drives hibernating, so a NAS consolidates power draw rather than eliminating it.

The larger issue is the failure domain. When Home Assistant lives on the NAS, a NAS update, a drive rebuild, or routine maintenance is also Home Assistant downtime. USB radio passthrough into a VM adds another dependency that a bare-metal HAOS install does not have. If storage and virtualization are the plan, the NAS-specific picks cover the trade-offs in full.

Storage Decides Reliability More Than CPU or RAM

Recorder continuously writes entity state changes to a database, and SQLite remains the default and recommended engine. That continuous write stream is the single biggest reason a Home Assistant install becomes unreliable, and it is usually blamed on the wrong component.

Reliability ordering for this workload, broadly:

• Ordinary microSD card: the supported floor, and the first part to wear out.
• High-endurance A2 microSD: less risk, same architecture.
• eMMC or a decent SATA SSD: the practical baseline for a host expected to run for years.
• Quality SSD or NVMe: capacity and headroom are the reasons to go further, not speed.

Home Assistant's minimum storage is 32 GB on both the Pi and the virtual machine image, and a realistic long-lived installation uses 64 to 128 GB or more. Larger drives are driven by Recorder history retention, rapidly changing entities, add-on data, backups, and camera recordings rather than by Home Assistant itself.

One practical feature matters here. HAOS can relocate its data disk to external storage, and Home Assistant documents that faster storage can improve system performance. On a Raspberry Pi 4 this is the safer route than some full USB-boot arrangements. The mistake to avoid is assuming NVMe's sequential speed will speed up automations. The upgrade that pays off is moving off a plain microSD card, not buying the fastest drive on the shelf. The setup guide covers the HAOS data-disk move step by step.

Zigbee, Thread, and Z-Wave Radios

Radio hardware and its placement cause more Home Assistant reliability problems than any CPU choice, because USB 3.0 ports and devices generate interference in the 2.4 GHz band that Zigbee uses. Home Assistant specifically explains why a USB extension cable ships with its own coordinator, and the same physics applies to every other stick.

• Home Assistant Connect ZBT-1 (formerly SkyConnect): a Silicon Labs EFR32MG21 running Zigbee 3.0, Thread-capable through firmware, up to +20 dBm output, and roughly 5 V/150 mA. Home Assistant recommends its own Zigbee firmware for standard use. Check the Home Assistant Connect ZBT-1 on Amazon
• SONOFF Zigbee 3.0 USB Dongle Plus-E: the same EFR32MG21 silicon with up to +20 dBm transmit power and a shielded aluminum enclosure. A common Zigbee2MQTT coordinator. Check the SONOFF ZBDongle-E on Amazon
• Zooz 800 Series Z-Wave Long Range USB Stick (ZST39 LR): Z-Wave 800 with Long Range, S2 security, and SmartStart, on the US 908.42 MHz band, using a USB 1.1-class interface that is more than sufficient for Z-Wave traffic. Zooz recommends a USB extension cable in Home Assistant setups. Check the Zooz ZST39 LR on Amazon
• Aeotec Z-Stick 10 Pro: a combined Z-Wave 800 (EFR32ZG23) and Zigbee 3.0 (EFR32MG21) controller with Long Range support where regional rules allow, useful when minimizing USB devices matters. Check the Aeotec Z-Stick 10 Pro on Amazon
• Nortek HUSBZB-1: a legacy stick that carried both Zigbee and Z-Wave in one USB device. Existing installations can keep running, but new buyers should choose current Zigbee and Z-Wave 800 hardware instead. Check the Nortek HUSBZB-1 on Amazon

The rule that fixes most problems costs nothing: put the coordinator on a USB extension cable, away from the host chassis, USB 3 storage, and Wi-Fi access points.

Frigate and Cameras: The Workload That Changes the Spec

Frigate is where "Home Assistant minimums" stop being about Home Assistant. Video decoding, object detection, and optional features such as semantic search and license-plate recognition are a different class of workload, and folding camera requirements into a Home Assistant spec is how people end up over-buying.

Frigate strongly favors hardware-accelerated video decoding and supports Intel OpenVINO for inference on compatible Intel integrated graphics. That is why an N100 mini PC becomes the right answer for a camera-heavy system: its UHD graphics can handle decode and detection without a separate accelerator. Some Frigate features carry their own requirements that plain Home Assistant never touches, including roughly 8 GB of RAM and AVX/AVX2-class hardware for certain semantic-search deployments.

The old shorthand that every Frigate host needs a Google Coral USB Accelerator is out of date for modern Intel hardware. The Coral remains a valid accelerator at up to 4 TOPS and roughly 2 TOPS per watt, and it still makes sense in existing or low-power setups, but it is not part of Home Assistant's requirements and it is no longer the automatic starting point for a camera build. If the shared host also runs media services, the way you size a Plex server follows the same workload-first logic.

Frequently Asked Questions

Conclusion

Home Assistant's real minimum is 2 GB of RAM and 32 GB of storage, and the project's own Green appliance proves how little the software needs by shipping 4 GB of LPDDR4X and drawing around 1.7 W idle. Everything above that is a decision about what else the host runs. The 4 GB tier covers a large conventional smart home with the usual add-ons. The 8 to 16 GB tier belongs to Frigate, virtualization, and heavy container stacks, and that is the only tier where an N100 mini PC beats a Raspberry Pi on merit.

Spend the budget in the right order. Move Home Assistant onto SSD or eMMC storage before upgrading the CPU, put Zigbee and Z-Wave coordinators on extension cables before blaming performance, and choose the host for the workload you are actually running rather than for a number in a spec sheet. If you are assembling a first server, the homelab starter guide walks through the full stack, and the Beelink and GEEKOM brand guides break down the N100-class systems that fit this workload.

For most people the answer is a Raspberry Pi 4 at 4 GB, Home Assistant Green, or an N100 mini PC with an SSD. If you want the headroom pick, the Beelink Mini S12 Pro on Amazon covers HAOS, containers, and light camera duty from one quiet box.