Services

Requirements in. A written plan and a named handoff out.

Platinum models the workload, drafts the options memo, and aligns suppliers when a path exists. GPU class and dates are confirmed per request.

Service line

Resource integration

A requirements summary, an options memo, open questions, supplier-alignment notes, and a handoff checklist when access is arranged.

Boundary: Owning a public GPU cloud, or application / MLOps delivery as a default product.

Service line

A scoped capacity ask (node count, topology, interconnect, timing) and access information once a supplier path is confirmed.

Boundary: On-demand provisioning from a Platinum-owned pool, or colo / power contracts unless separately scoped.

Service line

A serving-pattern plan — shared, dedicated, or higher isolation — mapped to latency goals, data handling, and who operates the runtime.

Boundary: A live inference SKU or a public endpoint under the Platinum brand.

Line	You get	Boundary
Resource integration	A requirements summary, an options memo, open questions, supplier-alignment notes, and a handoff checklist when access is arranged.	Owning a public GPU cloud, or application / MLOps delivery as a default product.
GPU capacity	A scoped capacity ask (node count, topology, interconnect, timing) and access information once a supplier path is confirmed.	On-demand provisioning from a Platinum-owned pool, or colo / power contracts unless separately scoped.
Model serving	A serving-pattern plan — shared, dedicated, or higher isolation — mapped to latency goals, data handling, and who operates the runtime.	A live inference SKU or a public endpoint under the Platinum brand.

How the work runs

You receive documents and a named operating party — then Platinum’s engagement ends at handoff.

Requirements summary: What was asked, what is known, and what is still assumed.
Options memo: Two or three delivery shapes, with the questions that still block a quote.
Supplier alignment: Who can host or supply, and in what order they must approve access.
Handoff checklist: Access path, operating party, and follow-up — Platinum stops here.

Fit

Who owns what

Platinum: Requirements modelling, options memo, supplier introduction, and RFQ alignment when a path exists.
Capacity contract: Signed with the hosting or supplying party named in the order — not as a Platinum-owned GPU cloud.
Operating party: Deployment, runtime, any SLA, and incident response, as written in that party’s agreement.
Handoff: Platinum’s engagement ends when access path, operator, and follow-up contact are named in writing.

GPU classes used in planning

These names follow vendor product lines. They label a planning conversation; they are not a stock list.

NVIDIA H100: Hopper-class starting point for large-model training and high-throughput inference.
NVIDIA H200: Hopper-class label when HBM capacity for larger contexts or weights is the constraint.
NVIDIA B200: Blackwell-class label. Availability and interconnect are confirmed per request.
NVIDIA B300 Blackwell Ultra: Planning label for B300-class conversations. HGX B300 architectures may also be referenced. Availability is confirmed per request.
AMD Instinct MI350X: When the conversation is not locked to a single accelerator vendor.
NVIDIA RTX PRO 6000 Blackwell Server Edition: Server Edition only. Distinct from Workstation Edition and Max-Q Edition.