NVIDIA H100
80 GB HBM3
- Scale
- 8 – 1,024+ GPUs
- Regions
- Asia core, selected global
- Network
- InfiniBand or 400G RoCE for multi-node
- Term
- 1 – 36 months
- Tenancy
- Dedicated or shared
- Management
- Managed or self-managed
On requestRequest this class
AI compute procurement · Asia-first, globally connected
Your gateway to AI compute in Asia.
Source GPU capacity across Asia and selected global markets, with support from procurement to deployment.
Sourcing mapSilver River · Asia-firstNot live capacity
Regions we help you source in
Drag to rotate · click a region to bring it to the front
- Multi-supplier sourcing, one point of contact
- Comparable technical and commercial proposals
- Assisted procurement: our team coordinates each step
Pixel globe: the regions we help you source in. Supply points with confirmed capacity are shared inside the workspace, per request.
01 — Built for AI workloads
traininginference
One procurement path, every stage of the model lifecycle.
Pre-training, post-training, serving or something custom: the hardware classes overlap, the sourcing questions differ. Start from the workload and we handle the rest.
Agents, RAG and evaluation pipelines are workloads that run on the same classes of hardware — pick the closest fit and describe the rest.
02 — Compute options
Request by class, confirm by supplier.
Availability confirmed on request.These are the resource classes we source most often. They describe what you can ask for, not what is in stock today.
NVIDIA H200
141 GB HBM3e
- Scale
- 8 – 512+ GPUs
- Regions
- Asia core, selected global
- Network
- InfiniBand or 400G RoCE for multi-node
- Term
- 3 – 36 months
- Tenancy
- Typically dedicated
- Management
- Managed or self-managed
On requestRequest this class
Blackwell-based systems
B200 180 GB HBM3e · GB200 NVL72 racks
- Scale
- Nodes to full racks
- Regions
- Selected sites; lead times vary
- Network
- NVLink intra-rack, InfiniBand inter-rack
- Term
- 12 – 36 months
- Tenancy
- Dedicated
- Management
- Usually managed
On requestRequest this class
Custom clusters
Specify (L40S, MI300X, mixed fleets)
- Scale
- Any
- Regions
- Any sourcing region
- Network
- Specify
- Term
- Specify
- Tenancy
- Specify
- Management
- Specify
On requestRequest this class
- ▪ GPU model & memory
- ▪ GPU count / cluster size
- ▪ Deployment region
- ▪ Network & interconnect
- ▪ Term
- ▪ Dedicated / Shared
- ▪ Managed / Self-managed
03 — Network
Asia at the core. Selected markets beyond.
Explore the networkSilver River helps you source in the regions below. The map shows sourcing focus, not live capacity: confirmed supply points are shared inside the workspace, per request.
Asia core sourcing regions
- Singapore SG
- Johor MY
- Kuala Lumpur MY
- Jakarta ID
- Bangkok TH
- Hong Kong HK
- Taipei TW
- Tokyo JP
- Osaka JP
- Seoul KR
- Mumbai IN
- Chennai IN
- Sydney AU
Selected global regions
- US West US
- US Central US
- US East US
- London GB
- Amsterdam NL
- Frankfurt DE
Supply-point statuses are shown only in the demo workspace and are illustrative.
04 — How it works
Four steps, one point of contact.
- 01
Share your requirements
A short, structured form captures workload, scale, regions, timing and constraints. Advanced details are optional.
Output: a request with a reference ID and a named Silver River contact.
- 02
Review matched options
We invite qualified suppliers to quote and normalise their proposals so the technical and commercial terms are comparable.
Output: side-by-side options with explicit inclusions, exclusions and validity.
- 03
Confirm terms and capacity
Choose an option and ask to proceed. We confirm capacity with the supplier and coordinate contract terms.
Output: confirmed capacity, agreed terms, an order record.
- 04
Deploy with coordinated support
Provisioning milestones, access hand-over and issues are tracked in one place with the supplier and our team.
Output: a deployment timeline and a single thread for follow-up.
We do not promise fixed response or delivery times. Each request shows its own next step and who owns it.
05 — Why Silver River
Procurement that understands the hardware.
01
Asia-focused sourcing
Supplier relationships across Singapore, Malaysia, Japan, Korea, Taiwan, Hong Kong, India and Australia, plus selected US and European sites.
02
Supplier qualification
Suppliers are reviewed for authorisation, operations and security posture before they are invited to quote. Approval and per-listing verification are tracked separately.
03
Comparable proposals
Every quote is normalised to the same fields: unit, term, currency, inclusions, exclusions, deposits, SLA and validity. Missing terms stay visibly missing.
04
Cross-border coordination
Contracting entities, currencies, data residency and import constraints differ by market. We work through them with you and the supplier.
05
Custom deployment requirements
Interconnect, storage, orchestration, isolation and compliance requirements are captured up front and carried into supplier conversations.
06
A clear point of contact
One Silver River contact owns the request from submission to hand-over, with every message and document kept on the case.
06 — In between
Buyers on one bank, suppliers on the other. Silver River runs between them.
One request in, comparable offers out, and a single thread from quote to hand-over. Neither side has to find, vet or chase the other; the process sits with us.
BuyersSilver RiverSuppliers
Buyers
Describe the workload once and receive options normalised to the same terms, with one Silver River contact. Supplier names and fees stay out of view until contracted.
Silver River
Qualifies suppliers, invites quotes, checks capacity claims, normalises terms and coordinates delivery. Every message and document stays on the case.
Suppliers
See redacted briefs for the requests they are invited to, quote on their own terms and keep listings verified. Buyer identity is released per invitation.
07 — Workspace
One case, three views.
Illustrative dataThe same request as the buyer, as Silver River and as a supplier see it. Each side gets exactly what it should: comparable options, the whole case, or a redacted brief.
Buyers see comparable options and one contact. Supplier names and Silver River's fees are not shown until contracted.
Silver River / Kestrel Dynamics (demo)
Buyer
REQ-2026-0012 · Request
Fine-tuning · 64 × H100 class · Singapore or Tokyo
Your contact: M. Tan, Silver River · Next step: review two options
- Submitted
- Clarifying
- Sourcing
- Offers ready
- Negotiating
- Contracted
- Provisioning
- Active
Option A · Singapore
Proceed requested- GPUs
- 64 × H100 80 GB
- Unit price
- USD 2.35 / GPU-hour
- Term
- 6 months
- Includes
- InfiniBand NDR, 2 PB NVMe, 24×7 support
- SLA
- 99.5% node availability
- Valid until
- 2026-11-15
- Capacity
- Verified capacity
Option B · Tokyo
Available- GPUs
- 64 × H100 80 GB
- Unit price
- USD 2.18 / GPU-hour
- Term
- 12 months
- Includes
- RoCE 400G, 1 PB NVMe
- SLA
- Not specified
- Valid until
- 2026-11-02
- Capacity
- Confirmed on request
Documents
2 files- Option-A-storage-spec.pdfShared by Silver River
- REQ-2026-0012-summary.pdfYour copy
Files are private to your organization unless Silver River shares them with you explicitly.
Messages
Buyer ↔ Silver River- Silver River · M. TanTue 14:02
Two options are ready. Option A has verified capacity from 1 Nov; Option B needs a 12-month term to hold its price.
- Kestrel Dynamics · R. OkaforTue 15:40
Does Option A's storage include snapshots? We need about 200 TB of checkpoint retention.
- Silver River · M. TanWed 09:15
Checking with the supplier. Their storage spec sheet is now in the case documents.
Supplier names and Silver River's fee treatment are not shown to buyers until contracted. Missing terms stay “Not specified”.
Open the demo workspace
Fictional organizations and figures. The live workspace uses the same screens.
08 — Updates
Announcements, partnerships and milestones.
All updatesNo announcements yet.
Company news, partnership and MOU announcements, deployment milestones, case studies and market notes will appear here once published.
Tell us what you need to run.
A request takes a few minutes. You keep a copy in your workspace and we take it from there.