About Substrate Co
Measurement as a Discipline, Not a Sales Tool
We started Substrate Co because engineering teams deserve benchmarking that is not shaped by who paid for it. Every figure we produce comes with a method note and stated caveats.
← Back to HomeOur Story
How Substrate Co Came Together
Substrate Co was founded in Kuala Lumpur by a group of engineers who had spent years evaluating AI infrastructure inside large organisations. The recurring frustration was the same: vendor-supplied benchmarks, synthetic suites, and spec sheets that did not translate to production workloads.
The decision to form an independent practice came from a simple observation — teams were making expensive hardware decisions based on figures that were not designed to reflect their actual tasks. The gap between published performance and measured performance in context is often significant, and it compounds over time when procurement cycles are long.
We established our office at Megan Avenue II in Kuala Lumpur and began working with engineering leads across the region. Our engagements are deliberately scoped and documented, so clients understand precisely what was measured, how, and under what conditions.
We do not sell hardware, hold reseller arrangements, or accept placement fees from infrastructure providers. Our income comes from the services on this site and nothing else. That structure is not incidental — it is the foundation of what we offer.
Mission
To give AI engineering teams a reliable measurement layer for hardware decisions — independent of commercial pressures and written in a form they can read, repeat, and challenge.
What We Stand For
- Caveats stated openly in every report
- Methods published alongside results
- No single winner declared without your input
- Workloads defined by you, not by us
Where We Work
Our primary base is Kuala Lumpur. We work with teams across Malaysia and the wider Southeast Asian region, and can conduct remote engagements where direct hardware access can be arranged securely.
The Team
Who You Work With
Wei Liang Ong
Principal, Benchmarking Methods
Spent eight years evaluating compute infrastructure at a large Malaysian financial institution before moving to independent practice. Focuses on method design and reproducibility.
Priya Rajendran
Lead, Workload Analysis
Background in ML systems engineering with a focus on inference performance. Responsible for benchmark execution and data quality across client engagements.
Amirul Hakim
Advisory & Client Relations
Manages the advisory retainer programme and client scoping. Translates engineering findings into language useful for procurement and leadership conversations.
How We Work
Our Operating Standards
Every engagement follows the same set of practices. These are not aspirational statements — they are the conditions under which we agree to take on work.
Written Method Before Execution
We document the benchmark plan — including workload definitions, warm-up procedures, and hardware configurations — before running a single test. You review and approve it first.
Confidentiality by Default
Workload details, architecture decisions, and results datasets are treated as confidential. We do not share client data with third parties or reference client names without written agreement.
Caveats as First-Class Content
Every result is accompanied by a method note explaining what conditions applied, what was not tested, and where the figures may not hold outside the measured scope.
Reproducible Procedures
Benchmark templates are written in a format your own team can execute. We aim for you to be able to re-run the same procedure independently, without needing us present.
No Vendor Relationships
We hold no reseller, referral, or co-marketing arrangements with hardware vendors, cloud providers, or software companies. Our advice is not shaped by who we might refer you to.
Data Minimisation
We collect only the information needed for the engagement. Access credentials, system details, and benchmark artefacts are handled with defined retention limits and disposed of securely after project close.
Independent Benchmarking for AI Infrastructure Decisions in Malaysia
Engineering teams selecting AI accelerators — whether for training large models or running inference at scale — face a measurement problem. Published throughput figures from hardware vendors reflect ideal conditions and selected workloads. What matters in practice is how the hardware performs on your specific tasks, with your data shapes, your precision requirements, and your concurrency patterns.
Substrate Co fills this gap as an independent practice. We help teams define what to measure, run the measurements correctly, and read the results with appropriate scepticism. The deliverables — benchmark plans, results datasets, method notes — are yours to keep and re-use as your evaluation needs evolve.
Based in Kuala Lumpur and working across the Southeast Asian region, we serve engineering leads, infrastructure architects, and technology decision-makers who need a data layer that holds up to scrutiny. Transparent scope, documented assumptions, and a clear separation from vendor interests are the conditions under which we operate.
See How We Can Help Your Team
Talk to us about your current evaluation or benchmarking question. A short scoping call costs nothing.
Contact Us