Civilisation matérielle is Braudel's name for the slow-moving, often unconscious infrastructure of daily life — what people eat, how they heat their homes, what they wear, how they travel, what they trade locally, what money they use — distinguished from both the market economy above it and the capitalism above that. Developed across his three-volume Civilization and Capitalism, 15th–18th Century (1967–1979), the framework insists that economic life has a vast substrate of customary practice that operates below the level of price formation and contractual exchange. Applied to AI, the concept demands that we see the data centers, the silicon supply chains, the cooling water, the electrical grids, the content-moderation labor, the training datasets — the material base without which no inference happens.
Braudel's three-level model inverted the usual hierarchy of economic analysis. At the bottom, material civilization — the ordinary, repetitive, locally embedded practices of provisioning and production. Above it, the market economy — transparent exchanges, visible prices, public trade. Above that, capitalism — opaque, concentrated, long-distance, monopolistic. The three layers coexist; each operates by different rules; the top two depend on the bottom.
The AI discourse talks almost exclusively about the top layer: platforms, valuations, investment flows, acquisition dynamics. It rarely talks about the material base. Yet the material base determines nearly everything: which countries can host frontier models, what the energy costs of inference are, where the cobalt and rare earths come from, which populations do the annotation labor, what water tables are being drawn down. A structural analysis of AI that ignores civilisation matérielle has missed the foundation on which the whole apparatus rests.
The framework also illuminates what changes slowly. The material infrastructure of AI — hyperscale data centers, transatlantic cables, chip fabrication facilities — is physical, capital-intensive, and geographically anchored. It changes at construction timescales (years), not software timescales (months). This creates a specific kind of path dependence: the physical base accumulated between 2020 and 2027 will shape the AI landscape for decades.
The political implications are sharper than the philosophical ones. The material infrastructure of AI is concentrated in specific jurisdictions (United States, China, select European hubs), owned by specific corporations, and powered by specific energy systems. The distribution problem lives in this concentration, and no rhetorical framework about 'democratization' survives honest contact with the infrastructure map.
Braudel's three-volume Civilization and Capitalism (1967, 1979) was his culminating work, a planetary economic history structured around the material/market/capitalism distinction. The first volume, The Structures of Everyday Life, is the locus classicus for the concept.
The forgotten base. Economic life rests on a substrate of everyday provisioning that formal economics routinely ignores.
The three-level hierarchy. Material civilization, market economy, and capitalism are distinct domains with distinct rules; they coexist rather than succeed each other.
AI has a material civilization. The chips, the power, the water, the labor, the geographic anchors — these constitute the infrastructure without which no inference exists.
Physical timescales constrain digital ones. The material base of AI changes at construction speeds, creating a slower structural layer beneath the rapid software conjuncture.
Whether the three-level distinction holds cleanly in the digital age — where data centers are owned by the same firms that operate the platforms — is contestable. The cleaner reading is that the levels are analytically distinct even when operationally integrated: the capitalist layer extracts from material civilization regardless of ownership overlap.