Three terms decide almost everything about who can compete where in this industry, and what they are allowed to earn for it. Learn these three and the rest of the value chain organises itself.
A reference design — Nvidia calls its own versions MGX and HGX — is a complete, prescriptive blueprint: chassis dimensions, power-delivery topology, cooling-loop layout and an approved component list, handed to a manufacturing partner to build almost exactly as specified. Nvidia's own documentation describes these as "prescriptive design patterns to ensure optimal performance when deployed in a cluster environment." The manufacturer's job shrinks to precision execution — sourcing, assembling, cabling and testing a design someone else has already engineered. Custom engineering is the opposite: a hyperscaler or systems house designs its own rack, cooling loop and network topology from first principles, owns the resulting intellectual property, and captures whatever performance edge that design delivers. It is a far smaller, far harder door — in practice, one only Nvidia itself, a handful of Taiwanese ODMs building to Nvidia's own next-generation specification, and the hyperscalers' in-house hardware teams (Google's TPU pods, Meta's own rack designs) have fully walked through.
A generic ISO-grade quality certification says a factory runs a disciplined, repeatable process. It says nothing about whether a specific server design will actually keep 72 GPUs alive under sustained real-world load. That second, narrower gate is Nvidia's own NVQual stress-testing regime — covering "GPU stability under sustained compute and memory workloads, thermal performance and throttling behaviour under stress, power delivery and system stability, and PCIe link integrity and bandwidth" — which a system must pass before it can carry the NVIDIA-Certified Systems label. Passing it once is not enough: each new architecture generation (Hopper, then Blackwell, now Vera Rubin) requires its own re-certification cycle, on a roughly annual cadence set entirely by Nvidia's own product roadmap.
A server is not sold once it is built correctly; it is sold once its builder has actually been allocated the chips to put inside it. Nvidia's own supply is the scarce input in this industry, not manufacturing capacity, and Nvidia is reported to be tightening its control over who gets to build with it: industry reporting describes the company "centralising" AI server assembly around a shrinking circle of select manufacturers, and raising the share of components marked "customisation not allowed" in its own reference designs from roughly 5-7% in earlier generations to 20-22% in newer ones — with its next-generation Vera Rubin racks reported to eliminate "ODM-designed" component line items entirely. Being NVQual-certified is necessary to compete for allocation; it is not sufficient to win it.
| Term | What it actually certifies | Who typically holds it | What it does NOT tell you |
|---|---|---|---|
| Reference design compliance | The build follows Nvidia's own prescriptive chassis/power/cooling specification | Nearly every serious India-listed AI-server builder in this report | Whether the builder has actually secured chip supply, or added any differentiated engineering of its own |
| NVIDIA-Certified Systems (NVQual) | Thermal, power, signal-integrity and mechanical stress-testing of one specific system design | A smaller, named set of Nvidia partners — confirmed for Netweb in this report's own research; see its individual company report | Whether that certification, once won, guarantees ongoing chip allocation at scale |
| Allocation | Actual, contracted chip supply against a specific order | An even smaller, largely undisclosed set — Nvidia does not publish an allocation list | Whether an allocation this quarter implies one next quarter, as Nvidia's own component-customisation policy keeps tightening |
Source: Dart Consultants, from nvidia.com and docs.nvidia.com (NVIDIA-Certified Systems and NVQual programme descriptions), and Digitimes reporting on Nvidia's ODM-centralisation strategy (13 Nov 2025 and 15 May 2026) — see Notes for full citation list.
If a company's revenue comes from following Nvidia's reference design as written, assume the margin is thin and getting thinner, no matter how large the order book looks. If a company's revenue comes from something layered on top of or upstream of that design — chip-level IP, advanced packaging, or a genuinely scarce piece of engineering the reference design does not specify — assume that is where this report's attention should go, and where the moat, if any, actually sits.
