| Frontier firm |
A company operating near the leading edge of a strategically important technology and capable of financing or coordinating infrastructure at unusually large scale. |
| Frontier platform |
A frontier firm whose models, networks, distribution, interfaces, or services become a shared operating layer for other organizations. |
| Hard financial systemic risk |
Distress that materially impairs credit, funding, payments, clearing, or major financial intermediaries. |
| Macro-financial amplification |
Transmission from valuation or investment shocks into consumption, collateral, capital expenditure, suppliers, utilities, employment, or regional public finance without necessarily impairing core financial functions. |
| Operational criticality |
Reliance on a service whose interruption cannot be absorbed or replaced within the required recovery time and loss tolerance. |
| Strategic importance |
Relevance to defense, communications, science, industrial capacity, public administration, or national autonomy. |
| Resolvability |
The ability to continue or transfer critical services while ordinary losses, ownership change, and restructuring occur. |
| Boundary case |
An extreme case used to reveal a mechanism; it is not representative evidence unless the mechanism survives comparison with ordinary, failed, and countervailing cases. |
| Valuation-dependent acquisition capacity |
The ability to use highly valued equity as consideration; a higher relative valuation can reduce percentage dilution but does not eliminate dilution, overpayment, or integration risk. |
| Model neutrality |
Meaningful, auditable choice among model providers across availability, release timing, price, latency, routing override, credentials, data use, and portable workflow context. |
| Hyperscaler intermediation |
A market structure in which cloud, identity, data, security, routing, procurement, and billing abstract multiple model providers behind one managed enterprise control plane. |
| Managed control plane |
Routing, policy, evaluation, observability, failover, cost, and governance functions operated by a provider while specified decision rights and evidence remain with the customer. |
| Sovereign-capable deployment |
A workload-specific architecture in which the organization has legal authority, audit evidence, portable context, a tested alternative within its recovery objective, and the ability to sustain critical operation through a defined disruption. |
| Open-weight model |
A model whose trained parameters are available under a license, even if training code, data, or process are not fully open. |
| Managed digital labor |
An ongoing service in which a provider operates AI agents and human exception processes to execute defined work. |
| Outcome-as-a-service |
A commercial model priced against completed transactions, savings, service levels, or business results rather than tokens or seats. |
| Secure latency-compliant cost per successful task |
Fully loaded compute, energy, integration, supervision, security, compliance, expected-error, and delay cost divided by verified outcomes delivered within the required service level. |
| Founder and key-person control risk |
Dependence created when voting power, technical authority, strategic relationships, succession, or affiliated transactions are concentrated in one person or a small insider group. |
| Openness stack |
The separate dimensions of weights, architecture, code, license rights, data provenance, evaluation transparency, reproducibility, and deployment portability; no single dimension establishes full openness. |
| Open ecosystem, concentrated substrate |
A market structure in which models and applications are plural or portable while chips, fabrication, cloud, identity, control planes, or data-center infrastructure remain concentrated. |
| Model-derived training and distillation |
Use of model outputs to evaluate, adapt, or train another system; its legal and competitive treatment depends on license, contract, provenance, extraction method, and applicable law. |
| Primary frontier firm |
A firm whose principal strategic identity is a frontier model, frontier application, or integrated frontier platform. Baseline cases: SpaceX/xAI, OpenAI, and Anthropic. |
| Material index weight |
At least 1.0% float-adjusted weight in the S&P 500 or Nasdaq-100 on the trigger date. |
| Major infrastructure project |
A data-center, generation, transmission or equipment program with at least 1 GW of planned load/capacity or at least $10B of committed capital. |
| Matched comparison |
A pre-specified control selected on observable size, sector, beta, leverage, contract structure, region, workload or other variables named in the relevant hypothesis before the event window. |
| Verified outcome |
A completed task that meets pre-specified correctness, safety and latency requirements. Reliability and latency are included in the outcome definition rather than multiplied again. |
| Not observable |
A formal protocol state used when required treated and comparison data are unavailable. It supports neither the hypothesis nor the null. |
| AI factory |
A commissioned production system combining accelerators, host processors, memory, advanced packaging, interconnect, network, storage, software, power, cooling, facility, and operators. Ordered chips or announced MW are not an AI factory. |
| Hardware portability |
The demonstrated ability to move a defined workload between accelerator ecosystems within a stated time, cost, performance-loss, security, and recovery threshold. |
| Economically productive MW |
Energized capacity that is rack-ready, commissioned, software-usable, utilized, and delivering successful latency-compliant work after power, depreciation, networking, maintenance, and operating cost. |
| Adjudication state |
One of six protocol outcomes: not triggered; triggered/window open; support met; null met; mixed/indeterminate; not adjudicable. |
| Common hardware-substrate shock |
A foundry, memory, packaging, interconnect, platform-security, power or other shared-input event that affects multiple material issuers and therefore cannot be treated as an idiosyncratic firm shock. |
| Conditional support-pattern frequency |
The share of simulated draws satisfying a support Boolean under disclosed priors and dependence assumptions, conditional on a qualifying event and complete measurement. It is not a verdict probability, calibrated forecast, null estimate, or observability estimate. |
| Material AI-system issuer |
A public issuer with material index weight and direct control of a critical AI-stack layer whose distress could transmit an AI-related shock. NVIDIA qualifies at the baseline; other issuers enter only through frozen inclusion criteria. |
| Event-level verdict |
The six-state result for one qualifying event at its fixed endpoint. It is not automatically a conclusion about the general hypothesis. |
| Hypothesis-level verdict |
A replicated conclusion formed only under the minimum-event, diversity, support-share, null-share and missing-data rules in Table 12.1A. |