The questions tenants ask first
Tenant FAQ.
Tenant FAQ.
Money, hardware, security, contracts.
Twenty-plus questions, organized by topic. Updated October 2026. If your question is not here, email ARO at contact@aropps.com or fill the reservation form. Real person reply within two business days.
Pricing & capacity
How reservations and pricing work.
How is capacity priced?
Per deal. Pricing depends on GPU (B300 or H200), server vendor, term length, site, and the size and shape of your reservation. We do not publish hourly rates because two tenants reserving the same rack count for the same term can land at meaningfully different prices once storage, networking, and support are specified. Prepay discounts are available on multi-year terms. The reservation form gives us enough to come back with a sized number within two business days.
What contract length is required?
Reservations start at 12 months. Most tenants reserve 24 to 36 months or longer; multi-year terms earn better unit economics, and prepaying part of the term earns a further discount. ARO is not built for spot or short-burst usage; if you need 200 GPUs for an afternoon, an on-demand marketplace is a better fit.
What capacity sizes can I reserve?
From one rack up. There is no fixed GPU ceiling: a reservation scales with the rack space available at the site, and large reservations span multiple sites. Servers ship with eight GPUs each, so a single rack holds a defined number of servers set by the power and cooling at that site. At Micro Edge Data Hubs, a deployment can start at a single 8-GPU server. We size to your workload during the reservation conversation.
Do reservations come with capacity guarantees?
Yes. ARO reserves the servers in your reservation up front; they are not shared in a multi-tenant pool and are not reclaimed or relocated mid-term. Performance is predictable, wait time is zero, and the capacity is dedicated to you for the term.
What is the cancellation policy?
Reservations are commitments. Early termination terms are negotiated in the definitive agreement and typically require either a defined break fee or substitution of a comparable-value workload. We do not lock you into something untenable; we do require commitment proportional to the capacity we hold.
Can I get on a wait list for upcoming capacity?
Yes. If capacity at a specific site, with a specific GPU, or at a specific compliance posture is not yet available, the reservation conversation can include priority placement when it comes online. Reach out and we will keep you in the queue.
Hardware
What runs on ARO capacity.
What GPUs does ARO offer?
The NVIDIA B300 (288 GB HBM3e per GPU) leads. The NVIDIA H200 (141 GB HBM3e) is available where its memory profile or price fits better. ARO also supports RTX, H-series, and B-series GPUs and keeps pace with new NVIDIA generations through its hardware partners. Final GPU and server recommendation depends on your workload and is set in the Letter of Intent.
Can I run 70B-parameter models on a single GPU?
Yes, and much larger ones across a server. A single B300 carries 288 GB of HBM3e, enough for 70B-class inference on one GPU with room for long contexts and production concurrency. Eight B300s in one server put more than 2 TB of GPU memory behind a single NVLink domain for frontier-scale models.
What network performance can I expect?
100 GbE tenant backhaul with redundant fiber paths from every site. NVLink connects the GPUs inside each server; multi-node clusters use a high-speed Ethernet or InfiniBand fabric between servers. Direct connect and private VLAN handoff are available. We document the network design and latency profile in your reservation.
What software stack runs on the servers?
Production stack supports TensorRT-LLM, vLLM, SGLang, NVIDIA Triton Inference Server, NVIDIA MONAI for healthcare, Kubernetes orchestration, and standard containerized inference workflows. Custom stacks are accommodated where the workload requires.
Is the capacity single-tenant or multi-tenant?
Single-tenant by default. You receive dedicated servers and racks, not shared ones. Performance, security, and noisy-neighbor risk are bounded by your reservation. If a smaller reservation would land on shared infrastructure, we say so explicitly before you sign.
What about training versus inference?
ARO capacity is sized for production inference first, but dedicated B300 and H200 clusters with a high-speed fabric between servers handle fine-tuning and multi-node training as well. If you are training a foundation model from scratch on tens of thousands of GPUs, a hyperscale training cluster is the right tool. If you are serving models in production, fine-tuning, or running RAG and agent workloads at scale, we are a strong fit.
Security & compliance
The compliance and security posture.
Is ARO SOC 2 compliant?
ARO does not hold its own SOC 2 report today. The data centers and managed services partners that operate the facilities behind ARO deployments maintain SOC 2 audit reports, which we make available to tenants under NDA. ARO's own SOC 2 Type I is on the roadmap, with Type II to follow after the required observation period.
Can ARO sign a HIPAA Business Associate Agreement?
Not today. The architecture is designed for HIPAA workloads (single-tenant, AES-256 at rest, TLS 1.3 in transit, audit logging), and a Business Associate Agreement for healthcare tenants is on our roadmap. If your workload needs a BAA, tell us in the reservation form and we will walk you through what we can commit to now and when.
What does single-tenant isolation mean in practice?
Dedicated GPU nodes physically reserved for your workload, private VLAN handoff at the demarc, no shared neighbors at the host level, encrypted storage, and access controlled by your IAM policies. Tenant separation is at the node level, not the workload level.
What encryption do you support?
AES-256 at rest for storage, TLS 1.3 in transit for network. Optional confidential-computing add-ons (TEE-protected GPU workloads) where the workload requires. Key management can be tenant-managed via your KMS or ARO-managed depending on the deal.
Will I have audit logs?
Yes. Audit logs are retained per the contract, can be forwarded to your SIEM, and are available for compliance review. Healthcare and regulated tenants typically receive immutable audit trails as part of the BAA or compliance addendum.
What about data residency and geographic constraints?
All ARO capacity is in the United States. Tenants with residency requirements can specify the sites in scope, and your data stays where you put it. International expansion follows tenant demand.
SLA & operations
Uptime, support, and what happens when things go wrong.
What SLA do you commit to?
Tier I sites target 99.671% availability, Tier II 99.741%, Tier III 99.982%. Each reservation is sized to a tier appropriate to the workload. The SLA is codified in the definitive agreement, not a marketing claim.
How do I reach support?
Through a real person, and only one. ARO is your single point of contact for every server at every site. Calls route to the ARO NOC, which monitors 24/7 and is backed by our data center and managed services partners for hands-on response. Tenants have a named account contact during business hours and an on-call contact off-hours. No support tickets, no chatbots, no support queue.
What happens if my workload is degraded?
ARO operations detects most degradation before tenants do. If you observe a problem, escalation is direct to the on-call engineer. Documented incident response, defined response times by SLA tier, and post-incident reports for tenants on request.
Will my workload be migrated if a site fails?
For tenants with workload portability requirements, ARO supports failover to redundant capacity at another ARO site. The failover model and recovery time depend on tier and reservation profile; documented in the agreement.
What hardware vendors stand behind the deployments?
Servers come from Dell, Supermicro, or MSI with manufacturer warranty and support. NVIDIA GPUs are under standard manufacturer warranty. Motivair cooling is under its own warranty. ARO is the single point of contact; we manage every vendor escalation so you never have to.
Commercial & contracts
How agreements work.
What does the contract process look like?
Initial conversation, sized reservation proposal in two business days, Letter of Intent when there is a fit, definitive agreement in 30 to 60 days. Equipment ordering and rack allocation follow the definitive agreement, and delivery windows are written into it.
Can I negotiate terms?
Yes. Standard terms exist as a starting point; meaningful negotiation is normal for enterprise reservations. Material commercial points (SLA tier, payment terms, exit conditions, custom compliance language) are negotiated based on your specific requirements.
What payment terms apply?
Net 30 from invoice is standard. Prepaying part of a multi-year term earns a discounted rate, negotiated per deal. Custom payment structures (milestone-based, ramp-aligned) are available for larger reservations.
Can I assign the contract to a successor?
Yes, with ARO consent (which is not unreasonably withheld). Tenant M&A is normal; the contract is structured to support successor entity assignment without renegotiation of material terms.
What if I need to expand mid-term?
Capacity expansion mid-term is supported subject to rack availability at your site or at another ARO site. Expansion can be added under the existing agreement; pricing reflects the marginal capacity added, not a renegotiation of the original.
Question not on the list?
Ask ARO directly.
If your specific question is not above, ask it on the reservation form or send it directly. A real person at ARO answers within two business days. Real person, no chatbot, no support ticket.
- 2 business day responseReal person, every inquiry.
- Letter of Intent within 5 daysSized reservation if there's a fit.
- Direct phone access(303) 803-7355 reaches Doug.