Value Capture
The share of the economic surplus a firm — or a stack position — manages to keep, as distinct from the surplus it generates. AI vendors plainly create vast aggregate value (faster code, faster customer support, faster materials science); the question this paper exists to answer is who keeps the resulting margin. Value creation and value capture are routinely conflated; in the AI stack they pull apart sharply, with whole categories of firm building products that are useful and unprofitable at the same time.
The creation–capture wedge
The wedge is the central diagnostic of the bear-case-synthesis and the circular-ai-economy. Application-layer firms create real consumer surplus while running at deeply inverted margins (perplexity-burn-ratio at ~164% of revenue on compute); foundation labs create civilisation-scale tools while spending ~50% of revenue on inference and ~75% on training; hyperscalers create the substrate while booking hyperscaler-capex-trajectory of ~1.7T planned against revenue lines that may be VC pass-through. Surplus is being generated everywhere along the stack — and the wedge between created and captured value sits inside each layer, not just between them. lee-01ai-pivot’s B2B re-orientation is essentially an admission that capture in Chinese B2C is structurally hostile to the producer, regardless of the value being created. choudary-ecosystem-teardown makes the same observation in pre-AI vocabulary: as transaction costs fall, capture concentrates in archetype positions (Aggregator, Infrastructure, Integrator switchboard, IP-rich Capability), not in the firms doing the most work.
Where capture lands — the paper’s spatial answer
H2_u-curve-of-value is the paper’s working spatial answer to the where question: capture concentrates at the top of the stack (workflow + distribution ownership — distribution-moat, middle-layer-defensibility) and at the very bottom (silicon, energy, frontier weights — ai-factory-huang), with the generic middle squeezed. deloitte-ai-dossier-eri supplies a non-software instantiation: in ER&I the unit of capture is the multi-agent operational orchestrator (vertical-ai-orchestration), not the underlying LLM. Each of these claims is a claim about capture, not creation — generic middle firms still create value; they just cannot hold it. The labor-side mirror (capital-labor-divergence, ai-young-worker-hiring-slowdown) extends the same logic: surplus created by AI-augmented knowledge work is captured by capital owners of L4–L6 platforms, not by the workers whose tasks are being automated.
Related
- H2_u-curve-of-value — the spatial answer to where capture lands across the stack.
- distribution-moat — the mechanism behind top-of-U capture.
- middle-layer-defensibility — what survives the squeeze; why “where you sit” ≠ “what you capture”.
- circular-ai-economy — the bear-case threat: capture at the bottom may be illusory.
- capital-labor-divergence — how created surplus distributes between capital and labor.
- choudary-ecosystem-teardown — pre-AI structural precursor; capture as archetype-position, not effort.
- ecosystem-business-archetypes — the four archetypes mapped onto capture mechanisms.
- vertical-ai-orchestration — non-software capture unit (industrial orchestrator).
- deloitte-ai-dossier-eri — vertical instantiation; capture at the orchestration layer.
- lee-01ai-pivot — strategic re-pricing of capture (B2C → B2B) in the Chinese stack.
- bear-case-synthesis — the cluster that pressures every layer’s capture claim.
- H5_ai-as-operational-not-product — operator-side capture: for builders outside the VC-megafund track, the structural opportunity is to be the buyer who captures the surplus on their own P&L rather than the seller competing to extract it.