On-Prem & Cloud Advisory · Kuala Lumpur
Where should your AI compute live?
Lumen Array helps organisations in Malaysia think through the hardware-versus-cloud decision calmly — weighing data sovereignty, operational cost, and workload patterns without pushing either path.
What We Offer
Three ways to work with us
Each engagement is scoped to a different stage of the placement decision — from a first-look briefing through to an ongoing advisory relationship.
Placement Options Briefing
A structured session that lays out the practical trade-offs of keeping AI compute on-premises versus drawing it from cloud providers. We help you frame the decision around your own constraints — not ours. Delivered as one session with a summary, trade-off sheet, and open questions.
- Single session format
- Trade-off sheet included
- Suited to first-call decisions
Hybrid Design Workshop
A two-session workshop that helps your platform team sketch a sensible hybrid approach — deciding which workloads sit where and why, with attention to data movement and operational effort. Guidance stays vendor-neutral. Delivered with diagrams, a hybrid blueprint, and a follow-up note.
- Two sessions with diagrams
- Hybrid blueprint document
- Best for platform teams
Placement Strategy Advisory
An ongoing advisory engagement that supports your organisation as it commits to a placement strategy across teams and quarters — documenting the reasoning and revisiting it as needs change. We act as an independent advisor across a three-month engagement with periodic reviews.
- Three-month engagement
- Strategy document & revision support
- For multi-workload organisations
Why Lumen Array
Guidance shaped around your situation
We don't sell cloud credits or hardware. That means our advice has nothing to gain from steering you in any particular direction.
No commercial ties
Our advisors hold no reseller arrangements with cloud vendors or hardware suppliers. The analysis you receive reflects your workload patterns, not our revenue targets.
Data sovereignty awareness
Malaysian organisations face specific considerations around data residency and regulatory posture. We factor these into every placement discussion from the outset.
Workload-level thinking
Rather than treating AI infrastructure as one block, we look at each workload type separately — training versus inference, batch versus real-time — and match them to suitable placement.
Written deliverables
Every engagement produces a document you keep — a trade-off sheet, a hybrid blueprint, or a strategy document — so the reasoning is on record for your team to revisit.
Fits your team's pace
Engagements are scoped to where you are in the decision process. A first-look briefing takes a single session; a longer advisory runs at a rhythm that suits your planning cycle.
Both paths respected
On-prem and cloud each have genuine strengths depending on context. We present both clearly so your organisation can make a well-reasoned choice and stand behind it.
Infrastructure Context
Where NVIDIA fits — and where the real decisions begin
NVIDIA's GPU lineup has become the default compute substrate for AI workloads across the region. Understanding what it can do — and where placement decisions still matter — is central to what we do at Lumen Array.
Hardware layer
NVIDIA GPU Platforms
NVIDIA's H100 and H200 accelerators have set the pace for large-scale AI training, while the A100 and L40S remain practical options for inference workloads where cost and density matter. The DGX platform bundles these into turnkey systems designed for on-premises deployment, and the HGX architecture underpins most hyperscaler AI compute available in the region today.
CUDA remains the dominant programming model, and the NVIDIA AI Enterprise software stack — including TensorRT, Triton Inference Server, and NIM microservices — significantly reduces the time between a trained model and a production deployment, whether that sits on-premises or in a cloud tenancy.
Training workloads
H100 / H200 clusters, NVLink fabric, high-bandwidth storage
Inference workloads
L40S, A10, TensorRT-LLM, low-latency serving with Triton
On-prem systems
DGX H100, DGX A100, BasePOD reference designs
Cloud availability
HGX-based instances via AWS, GCP, Azure, and regional providers
Advisory layer
How Lumen Array Applies This
Knowing the hardware options is a starting point, not an answer. The question that matters for most organisations is whether to own and operate GPU infrastructure on-site, draw on cloud-hosted equivalents on demand, or split workloads across both. That decision turns on factors specific to your organisation — not on the spec sheet.
Our advisors work through utilisation patterns, data residency requirements, team capability, and total cost of ownership to give you a grounded placement recommendation. We cover the NVIDIA ecosystem in depth because it is where most AI compute decisions actually land — but we treat it as a means to your outcome, not the outcome itself.
Workload profiling across training, fine-tuning, and inference stages
TCO modelling for owned DGX systems versus cloud GPU instances
NVIDIA AI Enterprise stack guidance — TensorRT, Triton, NIM deployment paths
Data residency and regulatory fit for Malaysia-based deployments
Hybrid architecture options that split workloads by sensitivity and frequency
H100
Primary training accelerator covered in our briefings
3
Placement tiers assessed: on-prem, cloud, hybrid
TCO
Total cost modelling included in every engagement
MY
Malaysia data residency and PDPA considerations built in
Start the conversation
Not sure which engagement fits your stage?
Tell us briefly where you are in the decision — on-prem, cloud, or somewhere between — and we'll suggest the most appropriate starting point.
Common Questions
Things people ask before starting
Straightforward answers to the questions that come up most often.
What does "vendor-neutral" actually mean in practice?
It means we have no commercial relationship with any cloud provider, hardware vendor, or software reseller. We are not compensated based on what you choose. Our fee is fixed to the engagement, so our analysis is not affected by which direction you go.
Which engagement should a first-time client choose?
If your organisation has not yet formed a clear view on placement, the Placement Options Briefing is the natural entry point. It is one session, produces a trade-off sheet, and costs RM 530. Many clients use it to align internal stakeholders before moving to the workshop or advisory.
Do these sessions require us to share sensitive data?
No. Our discussions focus on workload characteristics, operational constraints, and decision criteria — not on your underlying datasets or source code. We work with descriptions and architecture diagrams, and we are happy to sign a confidentiality agreement before any session begins.
How long does the Hybrid Design Workshop take to complete?
It runs across two sessions, typically spaced one to two weeks apart to allow your team to review the preliminary diagrams between them. The total elapsed calendar time is usually two to three weeks, and the final deliverable — the hybrid blueprint — is shared within five working days of the second session.
Is the Placement Strategy Advisory a retainer arrangement?
It is a fixed three-month engagement at RM 2,710, not an open-ended retainer. The scope includes a strategy document, periodic reviews, and revision support within that period. At the end of three months you can choose to continue under a new engagement or use the document independently.
Can sessions be conducted remotely?
Yes. All three engagements can be delivered fully online via video call and shared documents. For clients in the Klang Valley, in-person sessions at our Bangsar office are also available if that suits your team better.
Find Us
Our Kuala Lumpur Office
Level 7, Menara UOA Bangsar, Kuala Lumpur
Contact
Start the conversation
Describe your situation briefly and we will respond within one working day to suggest a suitable next step.
Contact Details
Phone
+60 3-2287 6041Address
Level 7, Menara UOA Bangsar
5 Jalan Bangsar Utama 1
59000 Kuala Lumpur, Malaysia
Working Hours
Monday – Friday: 9:00 AM – 6:00 PM
Saturday: 10:00 AM – 1:00 PM
Sunday & Public Holidays: Closed