20 use cases · 6 domains · colocation and GPU cloud
Situations you recognise. Answers on one screen.
Each use case starts from a moment an operator knows: a UPS on battery, a GPU that stalls a job, a tenant riding a fault, a hall that cannot be sold. Each one names the SynapseDC modules that answer it.
20Use cases
6Domains
2Business models in the first release
13Modules they draw on
01SeeEvery alarm scoped to the site, the building and the estate in one model.
02UnderstandWho is affected, what it costs, and the ranked cause with evidence.
03ActThe next step ordered by money and customers at risk, approval-gated.
04PlanSellable capacity, the next hall and the records that keep the loop honest.
This is not a separate product. It is what alarm scope, the estate model, the money model and Opstral AI do together whenever something fails. It does not care whether the alarm came from a UPS, a CDU, a fibre trunk or a GPU: the fault is scoped to the site, the services and customers behind it are read from the physical paths, the money is charged to one issue, and the recommended step waits for approval when it is disruptive.
Output: incident with services, customers and money attachedOutput: ranked cause with evidenceOutput: approval-gated action, recordedWorks across: power · cooling · network · GPU · access
01ScopeSite, edge, dependency or downstream. Downstream never counts on its own.
02ImpactServices, customers and premium customers read from the busway, the trunk or the job.
03MoneyLost so far, credits by tier, the next cliff. One issue per service.
04ActionOrdered by money and customers at risk. Disruptive steps wait for a person.
Facility 4 use cases
Power, cooling, environment and the physical last mile.
FacilityUC-F1Colocation · NeoCloud
A UPS goes on battery and nobody knows who is affected
The BMS raises a point alarm. Which cabinets, which tenants and whose SLA are worked out by hand.
The alarm is scoped to the site, an incident opens with every service on that busway, the customers and premium customers are listed, and the money per hour and the next credit cliff are on the incident.
Alarm scoped to the site, incident opened automatically
Services, customers and premium customers listed from the power path
< 60 salarm to incident with impact54services on the busway listed$2.6k/hexposure shown on the incident
FacilityUC-F2Colocation · NeoCloud
How long do we ride through if the utility fails now?
UPS minutes, generator readiness and fuel live in three places and are checked by walking the plant room.
One card per site: UPS minutes at today's load, generator kW and start readiness, fuel hours, and the checks behind each figure. The what-if takes any feed, UPS, side or PDU away and shows what goes dark.
Ride-through from live UPS, DG and fuel readings
What-if per feed, UPS, side, PDU, cooling unit and trunk
14 minUPS ride-through at today's load38 hfuel at current draw7SPOF classes checked per site
FacilityUC-F3Colocation · NeoCloud
A cooling unit fails and the floor has no spare
The CRAH alarm is visible; whether the floor still has N+1 and which racks will heat first is not.
Heat balance per floor against remaining capacity, the racks served by the failed unit, inlet temperatures trending, and liquid loops with their CDUs, so the operator knows whether to move load or wait for the fix.
Heat balance and spare per floor
Racks and services served by the failed unit
Inlet trend and environment zones against safe bands
Cords, panels and trunks were documented at install and never again. A fault proves the record wrong.
The physical last mile is derived per site: feed → UPS → busway → PDU → outlet → PSU, and port → patch cord → panel → trunk → ODF → cross-connect. Both are drawn on the device and in the 3D datacenter layout.
every PSUmapped to its outletevery portmapped to its trunk1 CRon every record change
GPU cloud 4 use cases
Jobs, GPUs, rails and the building's effect on both.
GPU cloudUC-G1NeoCloud
One failed GPU stops a 152-GPU training job
An Xid error on one GPU stalls the whole allocation. The GPU team sees the error; nobody sees the job, the customer or the cost.
The job with the biggest blast radius is on the Command Center and GPU Fleet: the customer, GPU time lost per hour, minutes since the last checkpoint and work to redo, with a Critical incident opened on the node.
Per-GPU telemetry joined to the job and the customer
152GPUs stopped by one failure$625/hGPU time lost52 minsince the last checkpoint
GPU cloudUC-G2NeoCloud
A job is 24 % slower and the cause is in the plant room
CDU flow drops, GPUs throttle, the job slows. The facility team sees the CDU; the platform team sees a slow job.
Thermal slowdown is traced from the job to the GPUs to the rack to the CDU loop. The slowest job and its cause sit together, with the hardware-slowdown flags and the step-time share lost.
24 %step time lost62 %CDU flow at the loop144GPUs slowed
GPU cloudUC-G3NeoCloud
Where should the next 256-GPU job go?
Placement is a guess between free GPUs, rack headroom, cooling state and power already paid for.
Pods are ranked for the job: free healthy GPUs, rack power headroom, cooling state (CDU for liquid), contracted headroom, kW already paid under the block or floor, and the marginal power cost per GPU-hour.
8–256GPUs per placement query$0.19power per sold GPU-hourper podfit and headroom
GPU cloudUC-G4NeoCloud
GPUs are allocated, drawing power and doing nothing
Idle allocations are billed and powered, but invisible between the scheduler and the facility.
Idle jobs are listed with their GPUs, the power they draw and what that costs per day, next to the inference pods that could be capped in the peak tariff window.
6 hidle allocation found in the demoper tenantenergy and cost4finding types against contracts
Network & edge 2 use cases
Backhaul, fabric and the alarms that belong to a site.
Network & edgeUC-N1Colocation · NeoCloud
A backhaul link drops: is it our fault or a symptom?
Remote PoPs raise alarms when a datacenter link fails. The NOC chases symptoms instead of the cause.
Every alarm carries a scope: site, edge (counted at both ends), dependency or downstream. Downstream symptoms never count on their own; they appear under the causing incident. Edge redundancy reads redundant, lost or isolated.
Scope on every alarm, downstream never double-counted
Edge redundancy per site
Ranked root-cause hypotheses with evidence
A completed change cited by the RCA marked as the cause
Four fabric cables down and a host lost both uplinks
InfiniBand and front-end Ethernet faults land as port alarms with no view of which node lost its rails.
Links read by plane (InfiniBand, Ethernet front-end, OOB), rails per node with ports and errors, the host that lost both uplinks named, and the transceiver or fibre trunk in the RCA with DOM readings as evidence.
3planes: IB · Ethernet · OOBper noderail healthDOMreadings in the RCA
Commercial 4 use cases
Tenants, SLAs, credits and the power bill.
CommercialUC-C1Colocation · NeoCloud
Which issue is costing the most right now?
Revenue at risk and SLA credits are reconstructed after the event, and never agree between teams.
Each impacted service is charged to one issue, its most severe cause. Lost so far, credits by tier, the next cliff and what is at stake reconcile across issues, customers, datacenters and sites. A cost-of-delay curve orders the work.
1issue per impacted service3 tiersof SLA credit tracked24 hforecast per issue
CommercialUC-C2Colocation
Which tenants are riding a critical fault right now?
A fault is live and the account team hears from the tenant first.
Exposed services are shown while the fault is live: tenants on the affected path, their SLA, what a four-hour outage would cost in credits, and the notice cadence due to each.
4 houtage scenario priced15/30/60minute notice cadenceper tenantwhy at risk
CommercialUC-C3Colocation · NeoCloud
Free racks are not sellable racks
Space says yes, power says no, cooling says maybe. Sales promises what the hall cannot deliver.
Space, power, cooling, ports and GPUs per site and hall, net of planned devices, contracted power, approved plans and idle allocations. Net available is what can be sold today.
6capacity dimensions4hold classes netted offper hallspace and power
CommercialUC-C4Colocation · NeoCloud
We pay for power we don't use
Take-or-pay floors, demand ratchets and idle colocation blocks are found when the invoice arrives.
The bill is priced against the tariff and the contract: energy, demand charge on the floor or peak, idle kW, and findings when a block is nearly used, a new peak sets the ratchet, or draw rises with no change in progress.
4contract finding typesper siteidle kW priced$/GPU-hpower against price
Security & access 3 use cases
Logical, OT and physical security as one case.
Security & accessUC-S1Colocation · NeoCloud
A badge-in at 02:10 and an ACL change on the WAN edge
The door system and the network logs never meet. Two events, two tools, no case.
Physical and logical detections are correlated: the badge read, the unapproved config change on the edge router outside any change window, and the user, in one case on NIST 800-61 phases with ATT&CK techniques.
1 casefrom two domains800-61phases on every caseATT&CKEnterprise + ICS
Security & accessUC-S2Colocation · NeoCloud
A UPS network card still has default credentials
Plant controllers sit on the OT network with vendor defaults and Modbus open to anything that can reach them.
OT zones are rated on IEC 62443 (target vs achieved level), default credentials and a Modbus write on a UPS card raise a case, and hardening checks per device family show the drift.
62443zones and levelsSSVCvulnerability decisionsper familyhardening checks
Security & accessUC-S3Colocation
Who entered the cage, and were they escorted?
The access audit is a quarterly spreadsheet; tailgating and held-open doors are found by chance.
Twenty-four hours of entries and badge reads per site, exceptions (tailgating, forced or held doors, anti-passback, unescorted visitors, overstays), who is on site now, and the quarterly rights review.
8exception types24 hlog on every sitequarterlyrights review
Plan & build 3 use cases
Capacity, plans, records and commissioning.
Plan & buildUC-B1Colocation · NeoCloud
What does a new 2 MW hall cost, and when can it go live?
A new hall is planned in a spreadsheet, priced from memory and approved by email.
A template sizes the hall, benchmarks price it, lead times schedule it, rules check it, the approval path follows the CapEx band, and the plan pack carries drawings, BoM and the business case to sign-off.
Refresh candidates are found by walking the halls or by vendor notice.
Lifecycle per device: install date, warranty, support contract, vendor end of sale and support. A refresh plan selects every eligible device at a site, splits them over waves and sources the replacements.
4lifecycle statesper modelend of sale and supportwavesfor every refresh
Plan & buildUC-B3Colocation · NeoCloud
Install a rack of GPUs without losing the record
The build team installs, the record is updated later, and the live estate starts from a record nobody trusts.
The install is planned work under a change request: U reserved, device installed as a record, ports connected, cords to outlets, and the work order, kit and commissioning gates tracked to handover.
