Skip to content

Comparison

Build vs. buy: post-quantum security

Most teams frame this as a binary, then discover the real question is which layer they want to own. This is a decision framework, not a sales comparison, and it makes no claim about any other vendor.

Entity
HEOSSI (PTE.) LTD.

The question underneath the question

Building and buying are not the only options, and they are rarely the real choice. A post-quantum programme has two separable layers: the policy layer, which decides who may use which key under which conditions and produces the evidence, and the primitive layer, which implements the algorithms and holds the key material. Owning one and buying the other is usually more sensible than owning both or buying both.

The decision that matters is which layer is a differentiator for your organisation. If cryptographic policy is core to your product, own it. If it is a control you must satisfy, buy it and spend your engineering on the integration and the inventory.

When building in-house is the right call

  • You already operate the key custody layer and the gap is algorithmic. Replacing a working HSM estate to gain a library is usually a loss.
  • Your regulatory boundary forbids a third party from holding or using keys, and you can staff the on-call rotation that comes with owning them.
  • The workload is narrow and stable: one protocol, one key type, one deployment shape, with no inventory problem behind it.
  • You can fund conformance testing as a recurring line item rather than a one-off project, because the test corpus and the standards move.
  • You have an audit function that can accept internally produced evidence, or you are prepared to commission independent assessment.

When adopting a platform is the right call

  • The gap is operational: no cryptographic inventory, no rotation discipline, no approval trail, and no evidence an auditor can re-run.
  • You need several algorithm families and integration paths at once, and maintaining all of them is not a differentiator for your business.
  • You want the conformance evidence to arrive with the product, including the corpus and the version it was run against.
  • You need policy enforcement that survives staff turnover: per-tenant tiers, four-eyes approval, and audited key operations.
  • You are prepared to qualify the deployment rather than assume it, and you have the environment to do that qualification in.

What building actually includes

A build is rarely one library integration. It is a set of components that each need an owner, and the cost sits in the components nobody scoped at the start.

  • Algorithm implementation and its dependency surface, including the advisory feed you now have to track.
  • Key generation, storage and rotation, plus the approval model that decides who may use a key and when.
  • A policy engine that enforces per-tenant or per-workload rules, rather than documenting them.
  • An audit record that survives retention requirements and can be reconstructed years later.
  • A conformance harness that reruns against a pinned corpus whenever the implementation or the standard changes.
  • A migration path for existing keys, ciphertexts and signatures, which is usually discovered late.

Each of those is defensible to build. The problem is that they arrive together, and the programme usually funds the first two and discovers the rest in the second year.

What buying actually includes

A platform is not a completed migration either. What you buy is the primitive and evidence layer, plus a policy surface you still have to configure against your own obligations. The parts that remain yours are the inventory, the integration, the qualification, and the decision about what your organisation will accept.

The costs that do not appear in the quote

Both directions have costs that are easy to leave out of the business case, and they are not symmetrical. Building hides recurring evidence work; buying hides qualification and exit work.

SideHidden costWhy it matters
BuildConformance maintenanceA pass rate is a snapshot. Test corpora, standards revisions and dependency advisories change, so the evidence has to be regenerated and republished on a schedule someone owns.
BuildHSM qualification per modulePKCS#11 is a specification, not a guarantee. Each module, firmware version and deployment shape needs its own qualification, and that work does not transfer.
BuildAudit productionEvidence that satisfies an auditor is a product in itself: provenance, signatures, retention and a way to reproduce a past result.
BuyDeployment qualificationA vendor statement about a module does not cover your configuration. Budget for qualifying the deployment you actually run.
BuyIntegration surfaceThe platform has to meet your identity, network and approval models. That integration is real engineering even when the product is complete.
BuyExit pathAsk what happens to keys and ciphertext if you leave. A migration story that only works inward is a dependency, not a platform.

The comparison, stated plainly

DimensionBuild in-houseAdopt QNSI
NIST conformance evidenceMust produce and maintainPublished (noble 435/435, liboqs 240/240)
HSM integrationCustom PKCS#11 work per vendorBYOHSM connector (PKCS#11 or REST)
Framework mappingManual control mappingInternal posture across ISO/IEC 27001, 42001 and 19790, without a certification claim
Maintenance burdenOngoing algorithm and library upkeepManaged updates and conformance refresh
Time to first PQC workloadDependent on your team and your HSM estateDependent on qualification, not on construction
Cost modelEngineering, HSM capacity, auditSubscription and qualification

QNSI does not claim accredited certification or independent conformity assessment. The ISO/IEC references describe internal framework posture, and ISO/IEC 19790 applies only to specifically identified cryptographic modules or components.

The hybrid that usually wins

Own the inventory and the policy, buy the primitives. The inventory is specific to your estate and nobody else can build it; the primitives are the part where evidence is expensive to produce and cheap to reuse. That split keeps your regulatory boundary intact while avoiding a two-year construction project for code that is not your differentiator.

It also keeps the exit path open. If the policy layer is yours, the inventory is yours, and the cryptographic bill of materials is in a format you control, changing the primitive provider is a migration rather than a rebuild.

Four mistakes that decide it badly

  • Comparing a build estimate with a subscription price. The build estimate usually excludes the evidence work, which is the part that repeats every year.
  • Treating the algorithm as the hard part. The algorithm is published; the custody, rotation and audit around it is the work.
  • Choosing a platform before writing the inventory. Without the inventory you cannot tell which surface the platform actually closes.
  • Letting the decision stand indefinitely. A build that was right for one protocol shape becomes wrong when the estate doubles.

Questions for your own team

These are the questions that decide the answer before any vendor responds. If you cannot answer them, the build-versus-buy debate is premature.

  • Who owns the cryptographic inventory today, and when was it last reconciled with reality?
  • Which keys can this team use without holding them, and which workflows would break if that changed?
  • Who signs the evidence an auditor receives, and can it be reproduced on demand?
  • What is the plan for the next dependency advisory, and who is on call for it?
  • If the provider disappeared tomorrow, how long would the current estate keep operating?

A decision checklist

  1. 01Name the layer you intend to own: policy, primitives, or both. Write it down before comparing vendors.
  2. 02Estimate the recurring evidence cost, not the build cost. Conformance, qualification and audit repeat.
  3. 03Ask which artefacts you can check yourself this month, and which require a sales call.
  4. 04Confirm what happens to keys and ciphertext if you switch providers in three years.
  5. 05Check that the deployment you intend to run has been qualified, not only the module in a reference architecture.
  6. 06Decide the review date for the decision, so a build choice does not become permanent by inertia.

Related guides

This comparison is informational. Actual costs, timelines, and suitability depend on your environment. Contact HEOSSI for a qualified assessment.