Direct answer: Aradia’s current public pricing page lists three turnkey private AI tiers: DGX Spark at $15,125, DGX Station at $194,093, and DGX B200 at $505,500. They are not three sizes of the same checkout. Spark is a solo or small-team desk-side appliance, Station is a large-memory workgroup system, and B200 is a rack-scale enterprise platform. Each price includes a different deployment target, model configuration, and support shape. The right choice depends on model size, concurrent users, cloud/API spend, facilities, internal engineering, privacy requirements, and uptime—not on the highest specification.
Affiliate disclosure: AI Compute Scout may earn a commission from a marked Aradia referral link if you later make an eligible purchase. This comparison remains independent, and Aradia is not NVIDIA.
Turnkey in one sentence
Aradia Turnkey is hardware plus the first deployment sprint. A bare DGX is a powerful platform, but it is not automatically a working private AI service. Someone still has to choose the model, install and pin the GPU software, configure the inference runtime, set the security boundary, test the workload, and document how the system is handed over.
For a non-specialist buyer, the practical difference is the owner of that work:
| Question | Bare hardware | Aradia turnkey |
|---|---|---|
| Who assembles the first working stack? | Your team or an integrator | Aradia, within the written scope |
| What does delivery mean? | The machine is delivered | A staged and tested configuration is handed over |
| Where does the time cost appear? | Engineering hours, review time, and delay risk | A higher quote that makes part of that work explicit |
| What remains yours? | Everything, including operations | Local facilities, network, usage, and work outside the contract |
Use this model rather than treating the turnkey price as a mysterious hardware markup:
Illustrative initial DIY deployment cost = hardware + engineering hours × your loaded hourly rate + facilities work + delay and incident risk
The public price does not disclose Aradia’s internal labor or margin, so this is a decision framework—not a claim about how its quote is allocated. Add ongoing power, maintenance, support, and replacement costs separately when comparing total cost of ownership. Turnkey is most useful when the organization needs a private service on a defined schedule but does not already have a CUDA/Linux/platform owner. It is less useful when a tested image, cluster, and on-call team already exist.
Does that include consulting?
Treat “consulting” as a scope question, not as an automatic promise. Aradia’s public materials describe a staging checklist, model and runtime configuration, validation reporting, activation, and optional SLA support. Its terms also say that Aradia is not a general corporate IT helpdesk: local LAN/WAN, office facilities, and custom agents or integrations outside the active SLA remain the client’s responsibility. A first-time buyer should ask in writing whether model selection, workload sizing, facility planning, acceptance criteria, operator training, and post-handoff changes are included or billed separately.
For a non-specialist, Turnkey is a strong fit when that written scope covers the first deployment decisions the team cannot make alone. It is not a substitute for an internal owner after handoff.
Current Aradia lineup
The official Aradia pricing manifest says its appliances are turnkey and that every tier includes hardware, staging, model configuration, agent compilation, and benchmarking. On the same page, Aradia publishes three tiers and no additional hardware appliance tier. We therefore treat Spark, Station, and B200 as the complete publicly listed Aradia hardware lineup checked on October 6, 2026. A future product or custom configuration should be added only after a fresh official-page check.
| System | Aradia turnkey price | Compute class | Memory | Intended users | Estimated deployment timeline |
|---|---|---|---|---|---|
| DGX Spark | $15,125 | GB10 Grace Blackwell desk-side appliance | 128 GB coherent unified memory | 1–5 | 30 days |
| DGX Station | $194,093 | GB300 Grace Blackwell Ultra workstation | 748 GB coherent memory | 10–50 | 60 days |
| DGX B200 | $505,500 | 8x Blackwell rack-scale system | 1,440 GB HBM3e + 2–4 TB system RAM | 100–200 | 90 days |
Prices are USD references from the Aradia page checked on October 6, 2026. Aradia’s terms characterize the listed delivery schedules as supply-chain estimates rather than guarantees. Prices and timing can change, and taxes, freight, local facilities, custom work, and optional SLA charges must be verified in a written quote.
Bare hardware versus Aradia turnkey
The comparison is straightforward only for Spark. NVIDIA’s U.S. Marketplace listed the 128 GB / 4 TB bare DGX Spark at $6,950 on October 6, 2026, while marking it out of stock. Against Aradia’s $15,125 turnkey listing, the visible difference is $8,175—approximately 118% of the listed bare price, with the turnkey total approximately 2.18× that price. This is a dated listing comparison, not a purchasable bare-hardware quote while stock is unavailable or a disclosed service-cost breakdown. The same Marketplace page had shown $4,699 on September 11 and no displayed price on September 29.
For DGX Station, NVIDIA’s public page routes buyers to a specialist rather than publishing one universal bare price. For DGX B200, NVIDIA’s enterprise channel also requires a configured quote. Their bare prices and percentage premiums are therefore not calculable without matching OEM configurations, support, installation, freight, and facility scope. A fully installed OEM quote might already include much of what a turnkey package covers; a chassis-only quote will not.
The conceptual equation for every tier is:
hardware + staging + model deployment + inference stack + security configuration + benchmarking + deployment support + vendor margin = base turnkey price
Do not present that equation as Aradia’s internal accounting. It is a checklist for procurement. The optional monthly SLA is a separate recurring cost and should be added separately to a TCO comparison.
Tier 1: DGX Spark
Spark targets a solo operator or small team whose model fits within 128 GB coherent memory. Aradia lists INT4-AutoRound quantization, configured vLLM continuous batching, hardened Linux and Docker, an air-gapped default with zero open inbound ports, a 14-day re-staging guarantee when configuration drift is detected, and an optional $1,500/month SLA. The listed estimated deployment timeline is 30 days and the target is 1–5 concurrent users.
This tier makes sense when an organization wants a local private system but does not want to assemble the Arm64, CUDA, container, model, and security stack itself. It is less compelling for a developer who already has a tested image, a low-utilization workload, or a strong internal platform team. Read the bare DGX Spark versus Aradia turnkey analysis for the line-item comparison.
Tier 2: DGX Station
Station is a workgroup platform. Aradia lists a GB300 reference configuration with 748 GB coherent memory, AWQ-INT4 quantization, configured vLLM continuous batching, hardened Linux and Docker, air-gapped default, priority SLA support and configuration, and a 14-day re-staging guarantee when configuration drift is detected. The optional SLA is $5,000/month, with 12 hours/month listed, and the estimated deployment timeline is 60 days.
This tier is for teams whose bottleneck is memory and concurrent local access rather than a single developer proof of concept. It requires serious power, cooling, networking, and Arm64 dependency planning. A conventional workstation, cloud GPU, or bare Station plus an experienced integrator may be a better value when the workload is experimental or custom. See the DGX Station versus Aradia turnkey guide.
Tier 3: DGX B200
B200 is rack-scale infrastructure for sustained enterprise workloads. Aradia lists eight Blackwell GPUs, 1,440 GB HBM3e, FP8/INT4 hybrid quantization, multi-GPU vLLM continuous batching over NVLink, hardened Linux and Docker, an air-gapped default, 48-hour staging burn-in and benchmarking, and a 14-day re-staging guarantee when configuration drift is detected. The target is 100–200 concurrent users, the estimated deployment timeline is 90 days, and the optional SLA is $12,500/month with 30 hours/month listed.
Aradia pairs the label “air-gapped default” with “zero open inbound ports.” Those controls are not automatically equivalent to a physical air gap, and a multi-user appliance still needs an approved local access path. Require a network diagram covering local access, outbound traffic, updates, telemetry, and the client-initiated support bridge before relying on the label.
The B200 choice is justified only when the organization can use its scale and operate the facility. The rack, power, cooling, network fabric, monitoring, and incident response are part of the purchase. If the model fits on Station or cloud capacity is bursty, B200 may be needless capital. The DGX B200 versus Aradia turnkey guide covers the rack-specific questions.
Which system fits which workload?
| Decision question | Spark | Station | B200 |
|---|---|---|---|
| Typical model envelope | Small-to-large quantized local models | Larger models and shared memory workloads | Multi-GPU models, high-throughput inference, training |
| Users | 1–5 | 10–50 | 100–200 |
| Facilities | Desk-side, still plan power and cooling | High-power workstation room | Data-center rack, 200–240 V and high-capacity cooling |
| Primary value | Shorter path to a private local service | Workgroup memory and shared access | Enterprise throughput and interconnect |
| Main risk | Paying for deployment when DIY is easy | Underestimating power and software compatibility | Buying rack capacity before demand is proven |
These user counts and service descriptions are Aradia’s planning references, not independent capacity benchmarks. Test the exact model, context, quantization, and concurrency before signing.
What turnkey does—and does not—remove
Turnkey can remove the first integration sprint: agreed model configuration and quantization, runtime configuration, agent compilation, container setup, hardening, burn-in, benchmark capture, and a staged handoff. Aradia’s deployment overview describes a staging studio, validation report, client-controlled activation, model updates, security patches, and performance diagnostics when the appropriate support arrangement is active. Its terms say support begins only through a client-initiated outbound WireGuard connection and does not include passive background monitoring.
It does not remove every operating responsibility. Aradia’s terms state that local LAN/WAN, office facilities, and how the appliance is used remain the client’s responsibility. Custom agents and new integrations outside included SLA hours may be billed separately. A private system can support a data-locality strategy but does not automatically satisfy HIPAA, GDPR, SOC 2, or other regulations.
When none of the Aradia tiers is the right answer
Choose bare hardware or an integrator when the organization already has experienced CUDA/Linux engineers, a validated image pipeline, an existing GPU cluster, and a clear owner for patches and incidents. Choose cloud GPU rental for bursty workloads, rapid experiments, larger temporary capacity, or uncertain demand. Choose a cloud API when shipping a feature matters more than operating hardware. Choose a smaller non-Aradia workstation when the model fits and utilization is low.
The strongest buying signal for turnkey is not “we like the hardware.” It is “we need a private service soon and do not want to build and staff the deployment path.” The strongest signal for bare procurement is “we already operate this class of infrastructure.”
Procurement checklist
For any tier, request a written scope covering:
- Exact hardware, memory, storage, network, OS, driver, CUDA, and container versions.
- Models, quantization, context length, agent workflows, and license provenance.
- Benchmark dataset, latency/throughput targets, thermal logs, burn-in, and rollback.
- Identity, secrets, backups, network boundaries, and support-bridge procedure.
- Warranty, freight, staging report, re-staging terms, SLA hours, response targets, and exclusions.
- Engineering ownership after handoff and the rate for custom changes.
For high-cost Station and B200 purchases, include facility readiness, cloud/API baseline spend, staffing cost, availability target, and a five-year replacement plan. Do not use Aradia’s public payback examples without recalculating them with your own utilization, energy, tax, and support assumptions.
Verdict
Spark, Station, and B200 are three different operating models, not simply three price points. Spark is a local private AI entry tier; Station is a shared large-memory node; B200 is a rack-scale AI factory component. Aradia’s value is the deployment path around the hardware. That value is real only if it replaces work, delay, and operational risk your team would otherwise pay to absorb.
Use Aradia’s current pricing page to verify the quote, then choose the smallest tier that meets the workload and service requirements. If your team can deploy and operate the stack, bare hardware or cloud capacity may be the more honest recommendation. The marked link is a referral link, not a reason to buy a larger system than you need.
Sources and verification note
Aradia’s lineup, prices, timelines, tier-specific specifications, staging scope, SLA options, and re-staging terms were rechecked against its pricing manifest, deployment overview, and terms on October 6, 2026. NVIDIA’s U.S. Marketplace listing listed the 128 GB / 4 TB DGX Spark at $6,950 and out of stock that day; $4,699 is a historical September 11 observation. NVIDIA Station and B200 public pages were rechecked for product positioning and the availability of a universal list price. Prices, configurations, service scope, and availability can change; request a current written quote.