eRacks Systems Tech Blog
Open Source Experts Since 1999
Cameras Don’t Sleep: Why Surveillance and Healthcare Workloads Need a Different Kind of NAS
Most NAS conversations start with capacity: how many terabytes, how many drives, what does it cost per gigabyte. For a typical file server or backup target, those are the right questions. For surveillance, broadcast, and healthcare imaging workloads, they are the wrong questions – or at least, far from the only ones.
What unifies video surveillance, broadcast archives, healthcare PACS, and clinical research isn’t capacity. It’s the write pattern. These workloads write continuously, 24 hours a day, 7 days a week, with dozens or hundreds of concurrent streams. Consumer NAS drives, optimized for read-mostly home and small-office use, wear out two to three times faster under that load. Cloud storage solves the wear problem but introduces a different one: bandwidth costs at sustained ingest rates that quickly outpace any savings.
Why Surveillance Breaks Consumer NAS
Consider a modest IP camera deployment: 50 cameras, 4K resolution, H.265 encoding, 4 megabits per second average bitrate per stream. That works out to roughly 1.6 terabytes of new video written to storage every single day. Multiply by 30 days of retention and the active hot storage requirement is around 50 TB at any moment, with new data flowing in continuously.
That write pattern is fundamentally different from what consumer NAS drives are built for. Drives like WD Red Plus and Seagate IronWolf are optimized for the read-heavy workload typical of file sharing, media servers, and personal backup. Push them into 24×7 sequential write duty and the drive’s internal wear leveling, head positioning, and thermal management all get stressed in ways their firmware was not designed to handle. Manufacturers publish workload ratings for a reason: a 180 TB/year rating means exactly that, and surveillance workloads exceed it within months.
The Surveillance-Certified Drive Difference
WD Purple Pro and Seagate SkyHawk AI are different products. They are CMR (conventional magnetic recording) drives, not SMR, which matters because SMR drives perform catastrophically badly under sustained random writes. They carry workload ratings of 550 TB/year (Purple Pro) and 550 TB/year (SkyHawk AI), well above what 24×7 multi-camera deployments produce. Their firmware is tuned for the specific I/O pattern of camera writers: long sequential writes, frequent metadata updates, occasional reads when an operator scrubs back through footage.
Surveillance-certified drives also handle the thermal and vibration environment of a populated chassis differently. A 24-bay or 50-bay NAS with all bays writing simultaneously generates measurable rotational vibration; surveillance-rated drives compensate with internal sensors that consumer drives lack.
Networking Matters as Much as Drives
The other half of the surveillance storage equation is network throughput. Fifty 4K H.265 cameras at 4 Mbps each is 200 Mbps of aggregate ingest. That fits in gigabit Ethernet on paper, but headroom matters – encoded bitrate spikes during high-motion scenes, and a saturated link drops frames. An eRacks Video NAS ships with 25 or 100 gigabit Ethernet as a standard option, leaving plenty of room for both ingest and concurrent VMS playback or operator review without congestion.
Matching the System to the Scale
Honest pricing requires honest scale guidance:
- NAS8 ($4,995): 10 to 20 cameras, 4K, ~30 day retention. Branch office, small retail, smaller campus deployments.
- NAS24 ($8,995): 50 to 100 cameras with multi-week retention. The sweet spot for medium businesses, schools, and mid-sized municipal deployments.
- NAS50 ($14,995): 200+ cameras, broadcast archives, or long-retention security workloads. Large campuses, transit systems, broadcast facilities.
- NAS72 ($24,995): Petabyte-class surveillance backends, broadcast video archives, or city-wide CCTV.
None of these are “from $1,995 supports 1000 cameras” claims. The NAS4 entry tier is a real product for branch offices and small deployments; it is not a 1000-camera VMS backend, and pretending otherwise wastes the customer’s time and ours.
Where Healthcare Fits
The same architecture serves an adjacent vertical with surprising overlap. Healthcare PACS (Picture Archiving and Communication System) imaging produces a similar write profile: continuous, high-bandwidth, multi-source ingest with regulatory retention requirements that often exceed 7 years. Clinical research datasets, EMR (Electronic Medical Record) backends, and DICOM imaging archives all push storage hard in the same way camera systems do.
Where healthcare differs is the compliance overlay: HIPAA-aligned architecture means protected health information cannot leave the customer firewall, audit logging must be enabled at the filesystem level, and encryption at rest is non-negotiable. eRacks Healthcare NAS configurations ship with ZFS native encryption (AES-256), full Linux auditd logging, and SIEM forwarding hooks (Splunk, Elastic, Wazuh) configurable at build time. The hardware is the same VNAS; the OS configuration is the differentiator.
What This Looks Like in Practice
Cameras (or external NVR/VMS like Milestone, Genetec, or Frigate) write video files directly to the NAS over NFS, SMB, or iSCSI. The NAS handles all the storage logic: RAID, encryption, replication, snapshots. The VMS handles the recording schedule, operator interface, motion detection, and analytics. Separating these concerns means you can swap VMS software without re-buying storage, and you can scale storage independently of recording capacity.
For smaller deployments, optional VNAS configurations can ship with Frigate or ZoneMinder pre-installed, putting both storage and VMS on one box. For larger deployments, run VMS on separate hardware and use the VNAS as pure write-optimized storage.
Cost Comparison vs. Cloud
For a 50-camera deployment with 30-day retention, storage requirements are roughly 50 TB written per month with 100 to 200 TB of hot storage maintained continuously. AWS S3 Standard storage alone runs about $8,000 per year before egress charges; egress costs spike whenever an operator reviews footage or evidence needs to be exported. An eRacks/NAS24 at $8,995 with surveillance-certified drives is a one-time capital cost that handles the same workload and pays for itself inside year one, then runs for five-plus years with periodic drive replacements.
Configure or Inquire
The Video NAS and Healthcare NAS configurations are both available at eracks.com/products/rackmount-nas-servers/. Click any configuration link and the quote form arrives with vertical-specific notes pre-filled, so you can move quickly to capacity sizing and drive selection rather than starting from a blank form.
Questions about capacity sizing, drive choice, or network topology? Contact us with your camera count, expected retention, and average bitrate; we will spec it.
joe May 28th, 2026
Posted In: NAS Storage, News
Tags: 24x7 storage, Frigate, healthcare IT, HIPAA NAS, IP camera storage, NAS servers, PACS storage, Seagate SkyHawk AI, surveillance, Video NAS, VMS, VNAS, WD Purple Pro, ZFS, ZoneMinder
HAMR vs CMR: 32TB NAS Drives and What They Mean for Your Storage Budget
A quiet but significant update landed in the eRacks configurator this week: 32TB HAMR drives are now available across the full NAS product line. For organizations that measure their storage needs in petabytes, this matters. The 102-bay eRacks/NAS100 can now be configured with 3.264 petabytes of raw capacity in a single 4U chassis – up from 2.6PB with the previous generation of 30TB CMR drives.
That is not a rounding difference. It is an additional 664 terabytes in the same footprint, with no extra rack space, no additional power circuits, and no change to the chassis.
What HAMR Actually Means
Hard drives have been using conventional magnetic recording (CMR) for decades. In CMR, a write head magnetizes small regions of a spinning platter to store data. The physics of that process set a ceiling on how densely bits can be packed – push the magnetic grains too close together and they become thermally unstable, meaning data can corrupt itself over time.
Heat-Assisted Magnetic Recording (HAMR) breaks through that ceiling by using a tiny laser to briefly heat a precise spot on the platter to around 450 degrees Celsius at the moment of writing. At that temperature, the magnetic material becomes temporarily easier to flip, allowing much smaller, more stable grains to be written reliably. Once the spot cools – which happens in nanoseconds – the written data is locked in place more durably than conventional CMR recording allows.
The practical result is higher areal density: more data per square millimeter of platter surface. Seagate’s current 32TB HAMR drives achieve this without increasing the drive’s physical dimensions. The same 3.5-inch form factor, the same power envelope, the same standard SATA interface – just significantly more capacity per bay.
For NAS applications running ZFS, this translates directly into larger pools, longer time-to-failure curves on RAIDZ arrays, and more headroom before an expansion shelf becomes necessary.
The Capacity Math
The eRacks NAS lineup runs from 4 bays to 102 bays. Here is what 32TB HAMR drives unlock at a few points in the range:
- NAS12 (12 bays): 384TB raw
- NAS24 (24 bays): 768TB raw
- NAS50 (50 bays): 1.6PB raw
- NAS72 (72 bays): 2.304PB raw
- NAS100 (102 bays): 3.264PB raw
These are raw figures. Usable capacity after RAIDZ2 parity and filesystem overhead will be lower – typically around 60-70% of raw depending on configuration – but the density improvement carries through regardless of the protection scheme you choose.
Where On-Premise Storage Still Wins on Cost
The cost argument for owning your storage rather than renting it has not changed, but the HAMR upgrade sharpens it. As a reference point: 100TB of object storage on Amazon S3 Standard runs roughly $27,600 per year in storage fees alone, before factoring in egress charges when you actually retrieve data.
An eRacks/NAS24 configured with enough capacity to cover that same 100TB – with room to grow – starts at $8,995. That is a one-time capital cost. In year two, cloud egress still costs what it costs. The NAS does not send an invoice.
For organizations in regulated industries – healthcare, finance, legal, government – the calculus has an additional dimension. Data sovereignty means knowing exactly where your data is, who has access to it, and under what legal jurisdiction it sits. Cloud storage agreements involve shared infrastructure, third-party subprocessors, and terms of service that can change. An on-premise NAS running ZFS on hardware you own answers those questions conclusively.
Available Now in the Configurator
The 32TB HAMR option is live in the eRacks online configurator for all NAS models. You can select drive size, drive count, RAID level, operating system (TrueNAS, Ubuntu, Rocky Linux, or Debian), and connectivity options at the time of order. Every system ships assembled and tested from Los Angeles.
eRacks has been building custom rackmount storage since 1999. The NAS line ranges from the 4-bay NAS4 at $1,995 to the 102-bay NAS100 at $29,995. All systems are open-source-friendly, built to order, and designed for data center or on-premise deployment.
Configure your system at eracks.com/products/rackmount-nas-servers/ or contact us to discuss capacity planning for your environment.
joe May 15th, 2026
Posted In: FreeBSD, Linux, NAS Storage, NAS24, News, Storage
Tags: 32TB drives, cloud vs on-premise, data sovereignty, HAMR drives, NAS servers, network attached storage, on-premise storage, ZFS
eRacks Expands NAS Lineup: 11 Models, 4 Bays to 100 Bays, From $1,995
eRacks Open Source Systems has expanded its rackmount NAS server lineup to 11 models, spanning from the 4-bay NAS4 at $1,995 to the 100-bay NAS100 at $29,995. The expansion targets the accelerating cost pressure of cloud storage subscriptions versus on-premise alternatives, with full Linux, ZFS, TrueNAS, and Ceph support across the entire range – and zero per-TB licensing fees.
The math behind on-premise NAS in 2026
Storing 100 terabytes on Amazon S3 costs roughly $27,600 per year in standard-tier fees. The same 100 TB sitting on an eRacks NAS24 – 24 bays, ~480 TB raw capacity – is a one-time $8,995 purchase. Payback is under four months.
Then there are egress fees. A single 100 TB pull from AWS to your office costs around $9,000 just to get your own data back. Cloud storage made sense when the data was small. At terabyte and petabyte scale, the math has flipped.
The lineup at a glance
| Model | Bays | Form Factor | Price (starting) | Best for |
|---|---|---|---|---|
| NAS4 | 4 | 1U or desktop | $1,995 | Branch office, dev team |
| NAS6 | 6 | 1U | $2,995 | Small office, light backup |
| NAS8 | 8 | 2U | $4,995 | SMB primary file server |
| NAS12 | 12 | 2U | $5,995 | SMB with growth headroom |
| NAS16 | 16 | 3U | $6,995 | Mid-tier file + backup |
| NAS24 | 24 | 4U | $8,995 | Mid-enterprise (the bestseller) |
| NAS36 | 36 | 4U | $10,995 | Mid-large workloads, scale-out node |
| NAS50 | 50 | 4U top-load | $14,995 | Media production, surveillance |
| NAS60 | 60 | 4U top-load | $19,995 | High-density archive, large backup |
| NAS72 | 72 | 4U top-load | $24,995 | Broadcast, large-scale archive |
| NAS100 | 100 | 4U top-load | $29,995 | Petabyte-class, Ceph nodes |
Plus a parallel all-flash NAS lineup for performance-tier workloads: FLASH10 ($5,995), FLASH20 ($9,895), FLASH24 ($8,995), FLASH48 ($15,995), and FLASH72 ($19,985) – all-NVMe arrays for database backends, AI training datasets, virtualization storage, and any workload that needs IOPS rather than raw capacity.
Open source the whole way down
Every eRacks NAS ships with full Linux – not a locked appliance OS – and supports your choice of:
- ZFS with ECC RAM for data integrity
- TrueNAS Scale for the friendly web UI experience
- Ceph for clustered scale-out
- MinIO for S3-compatible object storage
- Nextcloud for private cloud file sharing
- OpenMediaVault for the lightweight option
- Proxmox if you want NAS + VMs in one box
No vendor licenses. No per-TB fees. Full root access. You own the OS, you own the data, you own the hardware.
Hardware standards across the line
ECC RAM as standard. Hot-swap drive bays throughout. Redundant power supply options on NAS12 and above. NVMe SSD caching on larger models for accelerated reads. 25 GbE networking on demand for AI training workloads, video production pipelines, and large-scale backup.
The lineup also scales without chassis replacement. A NAS50 shipping with 24 drives today expands to 50 as needs grow – no forklift upgrade required.
When does it pay off?
For most organizations storing more than 5 TB of business data, on-premise NAS is cheaper than cloud subscriptions in year one. For HIPAA-aligned healthcare deployments, law firms protecting privileged data, or any organization with data sovereignty requirements, on-premise is not just cheaper – it is the right architecture.
Custom-built since 1999
eRacks Open Source Systems has designed, built, and shipped custom Linux servers since 1999. Every system is configured to order, burn-in tested before shipping, and supported directly by engineers who built it. No call centers, no upsell scripts, no per-feature licensing.
Get a quote
The full NAS lineup is at eracks.com/products/rackmount-nas-servers. Contact us for a custom quote sized to your specific capacity, performance, and software-stack requirements.
joe April 29th, 2026
Posted In: Backups, Linux, NAS24, NAS50, NAS72, Storage
Tags: backup, Best-Practices, Cloud Storage Server, eRacks/NAS24, eRacks/NAS50, NAS, NAS24, NAS72, near-line storage, Storage, Storage Server
Your AI Server, Your Software Stack: Why eRacks Doesn’t Ship a Custom AI OS
Last week we wrote about the 2026 AI GPU landscape – the hardware story. This week we want to talk about the question that comes right after a buyer picks a GPU: what actually runs on this thing once it’s plugged in?
For a lot of our boutique competitors, the answer is “our proprietary OS, our proprietary management tool, and the AI runtime we picked for you.” That model trades convenience for lock-in. We chose differently. Every eRacks AI server – AILSA, AIDAN, AINSLEY, AISHA – ships with the same software stack: vanilla Ubuntu 26.04 LTS plus a curated set of open-source AI tools, all pre-installed, all standard packages, nothing custom-forked.
What’s actually on the box when it arrives
By default, every AI server ships with:
- Ubuntu 26.04 LTS – standard server install, your team already knows it, supported until 2031.
- Ollama – the easiest local LLM runtime. OpenAI-compatible API. Pull and run any open-weight model with one command (
ollama pull llama4for Llama 4 Scout,ollama pull qwen3,ollama pull deepseek-v3.2, etc.). - Open WebUI – polished self-hosted chat interface. Looks and feels like ChatGPT. Multi-user, works for the whole team out of the box.
- vLLM – production-grade inference engine for high-throughput workloads. PagedAttention, continuous batching, the works.
- PyTorch + CUDA + cuDNN – matched to your installed GPU. Training and custom model work ready to go.
- Docker + NVIDIA Container Toolkit – containerized AI workloads work without setup.
- Standard Linux dev tools – Python 3.12, git, build-essential, tmux, htop, the lot.
That’s the hardware-AI side. On the storage and platform side, you also get the standard eRacks Linux base: ZFS available, SSH hardened, automatic security updates configured, monitoring hooks ready. No surprises, no proprietary agents.
Plug in, open a browser, start chatting
The whole “first 10 minutes” experience looks like this:
- Rack the server, connect ethernet, power on.
- SSH in once with the credentials we ship in the envelope, set your own password.
- Open
http://<server-ip>:3000in any browser on your network. - Open WebUI loads. Click “Sign up” (first user becomes admin), pick a model from the dropdown, start typing.
That’s it. The model list is pre-populated with whatever we sized your GPU for – Llama 4 Scout (17B active, 10M context) on the 48GB tier, Qwen 3 30B / DeepSeek-V3.2 distill on 32GB, Llama 3.1 8B + Mistral on 16GB. You can ollama pull any other model from the Ollama registry or Hugging Face the same day.
Why we chose this stack over building our own
There’s a temptation when you sell hardware to also sell a “platform” – a custom Linux fork, a branded management UI, a vendor-locked update channel. Some of our competitors do this. We don’t, for four reasons:
1. Your team already knows Ubuntu. Every Linux admin in your shop has used Ubuntu. Deploying our box is not a training exercise. Vanilla apt works. Standard systemd. No “did you check the wiki for this version of OurOS” support calls.
2. No vendor lock-in. If we go out of business tomorrow (we’ve been around since 1999, but still), your hardware keeps running on a fully supported open OS. You’re not stranded on an orphaned proprietary stack.
3. Updates are yours to control. When a new version of Ollama drops (which is every couple weeks), or when Meta drops Llama 4.5, or DeepSeek pushes V3.3, you can ollama pull it the same hour it lands. You don’t wait for us to vet it and ship a new firmware bundle.
4. The open-source ecosystem ships faster than any single vendor. Ollama, Open WebUI, vLLM, llama.cpp, the Hugging Face ecosystem – these tools improve weekly. Llama 4, Qwen 3.5, DeepSeek V3.2 all dropped in the last few months and were running on customers’ eRacks boxes within days of release. A vendor stack that re-bundles them is always a release behind. Vanilla Ubuntu lets you ride the open-source release cadence directly.
For teams that want different
That said: if you want a different OS, we’ll ship that too. Customers commonly ask for:
- Debian 12 stable – rock-solid, smaller footprint, similar tooling.
- Rocky Linux 9 / RHEL 9 – for teams standardized on enterprise RHEL.
- Proxmox VE – if you want to virtualize the AI server alongside other workloads.
- Bring your own image – we can boot your custom OS from USB before shipping.
And if you want a different inference stack:
- llama.cpp + llama-server instead of Ollama (more control, smaller dependency footprint)
- Text Generation Inference (Hugging Face’s TGI) for production deployments
- SGLang for advanced structured-output workloads
- LangChain / LlamaIndex stacks for RAG and agents
- JupyterLab + a stack of ML tooling if your buyer is an ML researcher
Tell us what you want at order time and we’ll pre-install it. If you don’t want anything, we’ll ship the bare OS.
The point
The hardware decision (which GPU, how much VRAM, how many drives) is the visible part of buying an AI server. The software decision is the longer-term part – it’s what your team interacts with every day for the next 5-7 years. We think that decision should be yours, on a stack you can fork, audit, replace, and redeploy on commodity hardware if you ever change vendors.
We’ve been shipping open-source Linux servers since 1999. Same approach. New use case.
joe April 20th, 2026
Posted In: AI, Deep Learning, Linux, Open Source, Rackmount Servers, servers
Tags: eRacks, Intel Arc, LLM, Rackmount Servers
The April 2026 AI GPU Landscape: From $349 to $16,500, Which VRAM Fits Your Model?
Last updated April 2026. Prices move weekly — keep checking back.
If you’ve been watching the AI GPU market, you already know the usual tension: NVIDIA dominates mindshare and most of the benchmarks, AMD is cheaper per gigabyte of VRAM but software support lags, and Intel keeps quietly shipping cards that punch well above their price tag but nobody talks about them. Meanwhile the actual hardware question most customers ask us is just: “How much VRAM do I need, and what’s the cheapest card that gets me there?”
This post is our answer as of mid-April 2026. We’ve broken the market into seven VRAM tiers, from the $349 low-profile starter card to a $16,500 datacenter accelerator, and matched each tier to the model sizes it actually runs well. All prices are current street prices, not MSRPs. At the end we’ll tie each tier back to one of our AI servers.
What “VRAM fits my model” actually means
As a rule of thumb for local inference:
- Model weight size ≈ parameters × bytes per weight. A 7B-parameter model at 4-bit quantization needs roughly 3.5–4 GB. The same model in full FP16 precision needs ~14 GB.
- Add 2–4 GB of working memory on top for KV cache, context window, and runtime overhead — more if you want long contexts.
- If your model plus overhead doesn’t fit, you’ll spill to system RAM or disk, and your tokens-per-second drops by an order of magnitude.
So the VRAM tier you need is driven by what you want to run, not by marketing tier names. Here’s how the 2026 market actually lines up.
The seven tiers
| Tier | VRAM | Price range | Models it runs comfortably | Example cards |
|---|---|---|---|---|
| Low-Profile (2U) | 8–16 GB | $320–$450 | 3B–8B quantized, embeddings, small classifiers | RTX 5060 LP, Intel Arc Pro B50, nVidia RTX A1000/A2000 LP |
| Entry | 16 GB | $480–$1,500 | 7B–13B full, 30B quantized | RTX 4060 Ti, RTX 5070 Ti, RTX 5080, AMD RX 9060 XT 16GB, AMD RX 9070 |
| Workstation | 20 GB single-slot | $1,280–$2,500 | 13B full, 34B quantized; quiet, ECC, space-efficient | nVidia RTX A4000 Ada (single-slot), AMD Radeon Pro W7800 32GB |
| Prosumer | 24–32 GB | $2,000–$3,740 | 34B full, 70B quantized | RTX 3090 Ti refurb, AMD RX 7900 XTX, RTX 5090 (availability-dependent) |
| Server | 48 GB | $1,299–$8,800 | 70B full, early 100B class | Intel Arc Pro B60 Dual 48GB, RTX 6000 Ada, NVIDIA L40S (passive), AMD Radeon Pro W7900 |
| Flagship | 96 GB | ~$9,680 | 70B full comfortably, 120B quantized, long-context everything | RTX PRO 6000 Blackwell 96GB ECC |
| Datacenter | 192 GB HBM3 | $15k+ (by quote) | Serious training + 405B-class inference | AMD Instinct MI300X |
The two surprise cards of 2026
If you only remember two things from this post, remember these:
Intel Arc Pro B50 ($399). A 16 GB low-profile card for under $400 didn’t exist twelve months ago. This card ships with both a standard and a low-profile bracket in a dual-slot form factor, slides into a 2U chassis without drama, and gets you enough VRAM for 7B-class models, embedding pipelines, and small classification workloads. As a starter card for a team dipping into local AI, nothing NVIDIA sells competes on $/GB at this form factor.
Intel Arc Pro B60 Dual 48GB ($1,299). This one is genuinely wild. Intel’s Project Battlematrix puts two Arc Pro B60 GPUs on a single PCIe card with 48 GB total VRAM — at roughly a fifth the price of an NVIDIA RTX 6000 Ada ($7,150) or a quarter the price of an L40S ($8,800). The software stack isn’t as mature as CUDA and your specific workload may or may not run well on Intel’s Battlematrix Linux drivers, but if your model runs, you’re getting 48 GB of VRAM for $1,299. For inference-bound 70B-quantized workloads where you don’t need peak training throughput, this is the best $/VRAM-GB in the market right now by a wide margin.
The AMD side
AMD’s RDNA 4 generation (RX 9060 XT, RX 9070, RX 9070 XT) turns out to be genuinely competitive for consumer-grade AI inference once you’re running on a framework that’s ROCm-aware — llama.cpp, Ollama, and vLLM all work. Performance-per-dollar on 16GB RDNA 4 cards is very close to the NVIDIA 50-series and sometimes ahead. For customers who don’t need CUDA and want to avoid NVIDIA’s pricing, this is a real path.
On the workstation side, AMD’s Radeon Pro W7800 (32 GB) and W7900 (48 GB) are direct replacements for NVIDIA’s RTX A5000/A6000 at roughly half the price, with ECC memory and workstation driver support. If you’re building a quiet single-user AI workstation, the W-series deserves a serious look.
At the top end, the AMD Instinct MI300X with 192 GB of HBM3 is the only single card that holds an entire 405B-class model in VRAM without any quantization tricks. It’s quote-only, it’s expensive, and the software story is still improving — but for the handful of customers for whom “does it fit” is more important than any other consideration, it’s currently the only game in town below $30k.
Which eRacks AI server for which tier?
We built our AI rackmount server line around this same VRAM-first thinking. Each model defaults to a different VRAM tier out of the box, and you can upgrade within the tier or jump tiers at configuration time:
- eRacks/AILSA — 2U, from $5,995. Default tier: Low-Profile. The “affordable starter” for teams trying local AI for the first time. Upgrade chassis to 3U-GPU or 4U-GPU if you want to move up to full-height cards later.
- eRacks/AIDAN — 2U full-height (up to 3 GPUs mounted sideways), from $9,995. Default tier: Entry 16GB. For 7B–13B models full-precision.
- eRacks/AINSLEY — 4U, from $14,995. Default tier: Prosumer 24–32GB. For 34B full or 70B quantized, with room for up to 4 full-height GPUs.
- eRacks/AISHA — 4U 8-GPU, from $19,995. Default tier: Workstation. Scales to the Server and Flagship tiers with up to 8 full-height GPUs — including the Intel Arc Pro B60 Dual for 48GB-per-card pricing unavailable anywhere else.
All four run Ubuntu Linux LTS Server out of the box, come with ECC-capable DDR5 RAM up to 512 GB, and ship with assembly, burn-in, and a 3-year warranty.
A note about prices
Our internal component costs tracked above — and therefore the baseline configuration prices you see on each product page — are mid-April 2026. The two forces moving them right now are (1) the AI-driven DDR5 memory supply crunch, which has roughly doubled ECC server RAM pricing since Q3 2025, and (2) the NAND flash shortage pushing SSD prices up. We’ll keep this post synced with our configurator. If you see a number here that doesn’t match what the configurator shows, trust the configurator — it’s the system of record.
Questions we haven’t answered yet
This post is the overview. Over the next few weeks we’ll be publishing deeper dives on:
- Why we just bumped our RAM prices 3x — an honest look at the 2026 memory market
- Arc Pro B60 Dual vs RTX 6000 Ada — real-world benchmarks on Llama 3 70B quantized
- The eRacks AI server lineup in depth — AILSA, AIDAN, AINSLEY, AISHA side-by-side
Got a specific model you want to run and aren’t sure which tier fits? Drop us a line and we’ll build the configuration for you.
joe April 15th, 2026
Posted In: AI, Deep Learning, LLM, Local AI, New products, Open Source, Rackmount Servers, servers, Technology
Tags: AI, AMD Radeon, Blackwell, Deep Learning, eRacks, GPU, Inference, Intel Arc, Llama, LLM, Local AI, Machine Learning, Open Source, Rackmount Servers, RDNA 4, VRAM