Bytes, not boxes.

A component shortage can reduce the number of products a factory completes without reducing the service those products deliver. Understanding that difference changes how a firm measures capacity and chooses a response.

Imagine a storage manufacturer facing a shortage of critical components. Its factories could assemble more drives, but the necessary parts are unavailable. The obvious question is how to restore production. A second question may be just as important: how much storage can each of the drives it can still make deliver?

My paper, Bytes, Not Boxes: A Demand-Unit Contingency in Architecture-Mediated Capacity, develops that distinction. It connects product architecture with supply-chain resilience by asking what the customer actually needs the constrained system to provide. The answer may be a quantity of storage, a number of physical drives, or both.

The unit of commitment changes the decision

A drive is a physical completion. Bytes are the service it carries. When a component is required once per drive, a denser drive can deliver more storage from the same component stock. When component requirements increase with density, the relevant question is whether they increase more slowly than storage capacity.

If they do, the component requirement per unit of storage falls. A service commitment can then be met with fewer physical completions and less of that particular constrained input. That relationship is the paper’s central analytical result: the operational value of a design depends jointly on component scaling and the unit in which demand must be fulfilled.

A customer requiring a fixed number of drives presents a different problem. Extra capacity in one drive may not replace another required device. Physical slots, location, compatibility and redundancy can preserve the need for separate boxes. A design that supports more total storage can still fail the customer’s physical-unit commitment.

A disruption that exposed both measures

The hard-drive industry provides a setting in which physical shipments and storage capacity can be observed separately. The 2011 Thailand floods damaged assembly operations and component suppliers, affecting heads, suspensions, motors and other inputs at different points in the production network.

Seagate’s own Thai factories remained operational, while seven of its ten largest component suppliers reported direct factory damage. The distinction is operationally important: surviving assembly capacity did not ensure that the components needed to use it were available. The bottleneck lay in a supply system, rather than simply in the manufacturer’s own factory.

Seagate’s disclosed June 2011 quantities were approximately 52 million drives averaging 590 GB. December quantities were 47 million drives averaging 653 GB. Multiplying shipments by average capacity gives roughly 30.7 exabytes at both endpoints: materially fewer drives carrying approximately the same total storage.

This six-month decomposition includes ongoing density improvement and changes in the product portfolio. Its contribution is the contrast between the two output measures. Rounded source quantities support “approximately flat” implied storage, rather than an exact offset or an estimate of storage saved by the flood response.

Different evidence answers different questions

A separate contemporaneous Seagate disclosure attributes accelerated transitions to named higher-density products to component disruption. That statement provides evidence about the direction of the company’s response. The shipment arithmetic provides evidence about output. Keeping those roles separate avoids making one source answer a question it cannot resolve.

The paper also retains an opposing account. Hutchinson Technology reported that some manufacturers temporarily lowered average drive capacity to maximise the number of physical drives produced from scarce components. This qualitative observation fits the importance of the demand unit: economising on components per box can matter when physical completions are the priority.

The archival record does not reveal the binding customer commitments of each manufacturer. These responses therefore illuminate the mechanism without assigning every firm to a verified demand category. Their coexistence is valuable: it directs attention to the conditions that can make opposite architectural choices sensible.

Follow the components through the design

Manufacturer specifications make the physical relationship concrete. In a selected Western Digital 22-to-24 TB comparison, both drives use twenty heads and ten disks. In a selected Seagate 20-to-24 TB comparison, heads increase from eighteen to twenty and disks from nine to ten. Storage increases faster than these component counts.

Those selected, later product comparisons establish how nominal component requirements per unit of storage can fall. They do not reconstruct the architectures available in 2011 or estimate an industry-wide relationship. Their role is to demonstrate a physical mechanism using documented designs.

The constraint can also move. A denser design that relieves pressure on one component may require more of another. Manufacturing difficulty, yield losses and technology-specific processing can increase even when physical head or disk counts remain constant. The capacity calculation must follow the whole resource system.

A capacity worksheet that respects the promise

The managerial output is a disciplined comparison of qualified designs. Begin with required storage and required physical units. Add the location, compatibility and redundancy conditions that affect whether service can be substituted across devices. Exclude designs that cannot be qualified and deployed within the shortage horizon.

For each remaining design, calculate the physical completions and the storage service that the available components can support. The most restrictive resource determines system capacity. Then test both customer commitments. Only physically feasible choices proceed to comparisons of margins, switching costs, penalties and qualification risk.

The paper’s worked illustration shows why both checks matter. A higher-service design raises the component’s service ceiling while reducing supported physical completions. It meets the illustrated service-only requirement but fails when the independent box requirement is imposed. The same capacity improvement has different consequences for different promises.

Resilience starts with the right denominator

This is theory developed through archival mechanism evidence. The paper also examines retail listings and finds a strong capacity trend before the flood, ruling out the proposed causal reading of the subsequent mix change. Keeping that result in view helps distinguish a measurable output difference from an established disruption effect.

The broader contribution is a better question for constrained production: what must the surviving resources deliver? Product design, engineering, procurement and customer contracts jointly determine the answer. Measuring both physical completions and usable service reveals opportunities that a single unit count can miss, while protecting against designs that improve a headline capacity figure and still fail the obligation.