Marginal Cost of Digital Reproduction
For digital goods, the cost of an additional copy is effectively zero — storage and bandwidth are de minimis. This creates downward price pressure and a sharp asymmetry between high fixed creation costs and near-zero reuse costs, driving business models like bundling, ads, subscriptions, paywalls, and freemium, and shaping how knowledge commons handle attribution and licensing.
The marginal cost of a good is the cost of producing one additional unit. For purely digital goods — software binaries, audio files, ebooks, research papers, knowledge base entries — that incremental cost is dominated by a few bytes of storage and a brief burst of bandwidth, both measured in fractions of a cent. Carl Shapiro and Hal Varian captured the dynamic in their 1999 book Information Rules with the formulation that information goods are "costly to produce but cheap to reproduce," because production is dominated by Fixed vs Variable Cost while subsequent copies cost almost nothing. Two consequences follow. First, in a textbook competitive market price trends toward marginal cost, so unrestricted digital goods face persistent downward pressure toward zero — the same pressure that reshaped recorded music after MP3s and that drives much of the modern piracy and open access discourse. Second, the gap between high fixed creation cost and near-zero reuse cost creates an extreme cost asymmetry: writing a song, training a model, or compiling an encyclopedia can cost millions, while the 1,000,001st download is essentially free. Jeremy Rifkin in The Zero Marginal Cost Society (2014) argued that this asymmetry, extended by the Internet of Things and renewable energy, would gradually erode conventional capitalist pricing across more sectors; critics counter that fixed costs, scarce attention, and infrastructure keep prices well above zero in practice. Sellers respond with business models that recover fixed costs without charging per copy: Bundling (Economics) many goods into a single price (cable packages, Office suites), advertising-supported distribution (search, social networks), flat-rate subscription services (Spotify, Netflix), Freemium tiers that give the base product away and charge for premium features, hard paywalls on journalism and research, and price discrimination through versioning — selling a deliberately degraded edition cheaper than a full one. For a knowledge commons, the same arithmetic cuts two ways. Redistributing an entry — copying, mirroring, translating, embedding it in an AI training set — is nearly free, which makes broad reach trivially achievable. Maintaining attribution, provenance, and license discipline, however, remains genuinely costly: it requires metadata pipelines, human review, and social norms that do not scale down with reproduction cost. The economic gravity of zero marginal cost tends to strip attribution off as content propagates, which is why Creative Commons licensing, copyleft, and machine-readable provenance are recurring engineering problems for any open knowledge project.