Why does the material
travel to the machines?
In most systems, entropy is made in one place and cryptography runs somewhere else entirely. That gap costs throughput, adds exposure, and turns one job into three procurements. Closing it is the cheapest upgrade in the rack.
A factory with the warehouse on the other side of town.
Stage 01 established that entropy is a raw material and that it needs inspecting. Now look at where it goes next. In a typical build the entropy source is one component, the cryptographic work happens on a general-purpose processor, and the two are joined by whatever the system happens to have.
- Component AEntropy source
- SharedSystem memory
- Component BCPU or crypto card
- OutputKeys and sessions
The material crosses a bus, sits in memory shared with everything else on the machine, and competes for cycles with the workload it exists to protect.
- Cost 01
Contention
Cryptography competes with the application for the same processor. Under load, the security work is what gets squeezed, which is precisely backwards.
- Cost 02
Exposure
Raw material and intermediate values move through memory shared with everything else on the machine. Every hop is somewhere to be watched.
- Cost 03
Integration overhead
You buy an entropy device, an acceleration device, and an integration programme to make the two agree. Three vendors, three qualifications, three sets of evidence.
Post-quantum cryptography made the machines heavier.
The new standards work, and they are also bigger. Post-quantum keys and signatures are substantially larger than the elliptic-curve material they replace, handshakes carry more bytes, and hybrid modes mean doing both the old work and the new work for a while. Every connection costs more than it used to.
Entropy demand goes up at the same time. More sessions, more reseeding, more key material per transaction. A source that was adequate when it trickled is a bottleneck when the system needs it in volume.
Every box in the estate gets opened for post-quantum migration. What goes into the slot while it is open is the decision that matters.
This is the refresh cycle, and it decides what the hardware looks like for the next product generation. Putting a heavier cryptographic load onto the same general-purpose processor, next to an entropy source that cannot keep up, is a design that will need revisiting within one cycle.
Put the machines next to the material.
Generate the entropy, verify it, and run the cryptography in the same place, on the same card, in the same slot. It is the reasoning that puts the stamping press next to the steel.
Here is the same capability counted two ways. These are counts of things you have to manage, not benchmarks.
- Slots and connectors usedBoard real estate is finite, especially in an appliance
- 2
- Vendors to qualify and manageEach one is a contract, a roadmap risk and a support path
- 2
- Driver and firmware stacksEvery stack is a maintenance and vulnerability surface
- 2
- Integration programmesEngineering time before a single unit ships
- 3
- Times entropy crosses a shared boundaryBetween where it is made and where it is consumed
- 2
- Certification evidence packagesAssembled, maintained and re-run at every product refresh
- 2
Two components, three programmes. The entropy source and the accelerator were designed by different companies for different reasons, and somebody on your team owns the seam between them.
Generate, verify and execute, in the same slot.
The same form factor and the same host interface, with more of the work finished before anything leaves the device.
- 01
Generate
Quantum entropy from photonic silicon, at rates that do not throttle the system above it — up to 1 Gbit/s on the high-performance platform, and 50–100 Mbit/s from the compact embedded line.
- 02
Verify
The physical source is measured against its model while the system runs, so quality is evidence rather than assumption. That is Stage 01, and it sits on the same board.
- 03
Execute
Post-quantum offloading, root-of-trust functions and DRBG seeding run on dedicated hardware instead of borrowing cycles from the application.
- 04
Deliver
Finished keys and completed operations leave the card. The raw material never has to make the trip.
Same slot, more value.
The operations themselves, and how they reach a host, are on cryptographic acceleration. Which card, in which slot, is on products.
Cryptographic agility stops being a hardware problem.
Post-quantum standards are settling rather than settled. Parameter sets get revised, hybrid modes come and go, and national schemes differ on what they will accept. If your algorithms are cast into fixed silicon or spread across a software estate, every change is a project.
Algorithms fixed in place
A standards revision means a board respin, a new qualification cycle, or a fleet-wide software migration under time pressure.
The cost of change is high enough that teams delay, which is how an estate ends up running algorithms nobody would choose today.
Algorithms in programmable hardware
The machines on the line can be reconfigured. A new scheme is a firmware update to a card that is already qualified and already deployed.
The entropy foundation underneath does not change, because good randomness is a requirement of every algorithm that will ever replace the current one.
Worth being precise: post-quantum algorithms and strong entropy solve different parts of the same cryptographic architecture. A quantum entropy source does not make a system post-quantum secure, and a post-quantum algorithm seeded from a weak source is not secure either. They are the material and the machine. You need both, in the same place, able to change independently.
How many separate components does it take to turn your entropy into a key, and who owns the seams between them?
Ask an architect to draw the path from source to key on a whiteboard. Count the boxes and count the boundaries. Every one is cost, latency and exposure that a single card removes.
The line is efficient now. The last question is where it physically sits, and who else is on the floor.