What $25,000 gets you in production.
from $25,000one-time, hardware separate
A working installation with the operator's view built in. Below is what you will watch after handover: throughput, latency, nodes and an audit log of every request. Every simulated number is labelled demo telemetry.
- 99.95%
- uptime SLA on managed ops
- 5 min
- incident response
- 24/7
- Limassol and Auckland
Tokens per second
Demo telemetry- Active requests
- 53
- Queue depth
- 1
- Prefix-cache hit rate
- 66%
Demo telemetry. Sum of six sample nodes.
GPU utilisation
Demo telemetryEvery GPU is yours, and so is the data about it.
Demo telemetry. Last 48 s, 0 to 100 %. Hover to scrub all six, select a node to filter the audit log.
p95 time to first token
Demo telemetryInfiano held 1.1–1.4 s time to first token across 40 GPUs.
Table view
Demo telemetry. The shaded band is the range measured at the Infiano hackathon. The line is simulated.
Cost meter
Customer reportedSame models, a fraction of the bill. Nocodo LTD went from about $10,000 to about $2,000 a month.
Nocodo LTD, per month
Infiano hackathon, 240 submissions scored
Reported by the customers, not simulated. Each pair is scaled to its own larger bar: about 80 % and 56 % lower.
Nodes
Demo telemetrySelect a node to filter the audit log.
Demo telemetry. Model and driver updates are handled by managed ops.
Audit log
Demo telemetryEvery request is logged inside your network, with who asked, which model answered and whether a role rule allowed it.
- 09:14:08NODE-06svc-analyticsllama-4-scoutTokens 265TTFT 1.35 s200 OK
- 09:14:08NODE-01svc-support-botgpt-oss-120bTokens 343TTFT 1.23 s200 OK
- 09:14:07NODE-01usr-legal-04gpt-oss-120bTokens 0TTFT 0.02 s403 DENY
- 09:14:06NODE-03usr-legal-04qwen3-32bTokens 103TTFT 1.21 s200 OK
- 09:14:05NODE-05svc-ocr-batchgemma-3-27bTokens 0TTFT 0.02 s403 DENY
- 09:14:04NODE-02svc-support-botmistral-small-3.1Tokens 198TTFT 1.16 s200 OK
- 09:14:03NODE-06svc-support-botllama-4-scoutTokens 648TTFT 1.33 s200 OK
- 09:14:02NODE-05svc-analyticsgemma-3-27bTokens 753TTFT 1.21 s200 OK
- 09:14:01NODE-05svc-searchgemma-3-27bTokens 277TTFT 1.36 s200 OK
- 09:14:00NODE-05svc-ocr-batchgemma-3-27bTokens 0TTFT 0.02 s403 DENY
- 09:13:59NODE-01svc-ocr-batchgpt-oss-120bTokens 606TTFT 1.40 s200 OK
- 09:13:59NODE-05usr-legal-04gemma-3-27bTokens 338TTFT 1.27 s200 OK
Demo telemetry. RBAC and audit logging ship with every production deployment. 403 means a role rule refused the request.
What $25,000 includes.
A fixed fee for the installation, one time. Hardware is separate and sourced at cost.
- Model selection and GPU sizing.
- Quantization, batching, caching.
- Monitoring, RBAC, audit logging.Visible above: GPU utilisation, tokens per second, p95 latency, the node list and the audit log.
- OpenAI-compatible API inside your network.Visible above: the audit log with the principal, the model and the allow or deny result of every request.
- Handover and documentation.
What we need from you.
The same prerequisites start the clock on a pilot, so a pilot first removes most of them.
Hardware at cost.
We size the hardware and source it at cost, with no markup. Ballpark figures, before sizing; smaller models run on far less.
Compare against your cloud API spend- One H100 80GB
- $25–30k
- a 4×H100 node
- $180–240k
One H100 route in total: the $25,000 deployment plus about $25–30k of hardware, so roughly $50–55k. A 4×H100 node adds $180–240k instead.
Handover, then your choice.
Handover and documentation. After handover your team can run the installation, or we can.
Scope your production deployment.
Tell us the model, the data and the GPUs you have. An engineer from Limassol or Auckland replies.