Abstract
This monitoring protocol examines what the public-market emergence of trillion-dollar frontier firms could mean for American finance, hardware supply chains, electricity, enterprise organization, labor, and geopolitical position. It begins with event discipline. SpaceX completed an exceptionally large initial public offering; OpenAI and Anthropic confidentially submitted draft registration statements; and SpaceX signed, but had not closed, an all-stock agreement to acquire Anysphere. The protocol then asks how valuation, index exposure, debt, accelerator supply, semiconductor fabrication, data-center construction, power, workflow ownership, and substitutability could transmit consequences beyond common shareholders.
The thesis rejects a single umbrella category for all forms of importance. It separates hard financial systemic risk, macro-financial amplification, operational criticality, strategic importance, and political salience. Index ownership can spread wealth losses and weaken capital expenditure without creating bank-style contagion. A firm is financially systemic only when distress can materially impair credit, funding, payments, clearing, or major intermediaries. Operational and strategic importance can justify continuity planning, but neither implies that common shareholders should be protected.
High valuation is treated as valuation-dependent acquisition capacity rather than proof of industrial dominance. The pending SpaceX-Cursor agreement is retained as a boundary case, but its full analytical treatment appears once, in Chapter 5; other chapters use cross-references rather than counting the same transaction repeatedly. Energy analysis incorporates sequential agentic compute, latency service levels, local physical ceilings, and rebound demand: efficiency can improve output per unit of energy while total electricity use still rises.
Most enterprises will not build and operate cross-model control planes themselves. The likely architectures are single-vendor consolidation, hyperscaler-managed plurality, independent managed control planes, and self-managed deployments for a minority of governments and unusually capable firms. Hardware is treated as a migrating set of control points rather than a permanent monopoly: NVIDIA combines accelerator, networking, software, and rack-scale advantages; Huawei is building a sovereign alternative; custom silicon can reduce accelerator dependence while deepening hyperscaler dependence; and TSMC, HBM, advanced packaging, data-center commissioning, and power remain shared constraints. Open weights can reduce model-layer dependence while leaving the substrate concentrated or shifting rents to adjacent layers.
The resulting document is Monitoring Protocol v1.2.1 - Final Baseline: a prospective research and adjudication protocol, not a tested theory or calibrated predictive model. It preserves separate qualification tests for financial-system transmission, macro-financial amplification, operational criticality, strategic importance, and resolvability. The final patch fixes verdict precedence, defines how event results roll into hypothesis-level conclusions, separates support-pattern simulation from evidence availability, completes the H1-H7 variable dictionary, relabels the historical exercise accurately, adds source-symmetry and revision ledgers, and clarifies the two-tier company taxonomy and migrating-constraint model. No composite systemic score is used.
| PROTOCOL THESIS The protocol separately tests issuer-specific financial transmission, common hardware-substrate shocks, infrastructure-finance stress, valuation-funded integration, power-latency constraints, hyperscaler intermediation, managed sovereignty, and open-weight rent migration. Each hypothesis can be supported, rejected, mixed, or not adjudicable under its own fixed rule. No cross-hypothesis probability or systemic score is calculated. |
|---|
How to read this document
This document is a frozen baseline protocol, not a results paper. Words such as "may," "could," and "plausible" identify open questions. Observed events populate the baseline; hypotheses remain untested until their trigger and event window occur. Quarterly reviews update the event ledger and evidence state, but do not repeatedly re-test endpoints or select favorable interim results. Formal adjudication follows the precedence, event-level, hypothesis-level, mixed, and missing-data rules in Chapter 12. Any change must be versioned prospectively and cannot rescue a claim after an outcome is known.
A result may be "not adjudicable with available evidence." Private retention, model mix, utilization, contract, or margin data cannot be inferred from management narratives. Missing data support neither the hypothesis nor the null. The protocol contribution is a clean factual scaffold, disaggregated concepts, a reproducible support-pattern generator, and an executable real-world adjudication design. The empirical program begins after this baseline. Simulation is conditional on complete measurement; evidence availability is tracked separately.
| PROTOCOL STATUS Baseline evidence and decision rules are frozen at July 24, 2026. Version 1.2.1 closes the final implementation gaps: verdict precedence, event-to-hypothesis aggregation, full H1-H7 measurement units, simulation/observability separation, exact holdings traceability, contract-denominator reconciliation, source symmetry, two-tier company taxonomy, migrating constraint regimes, and a complete revision ledger. No primary hypothesis has been adjudicated. No composite systemic score is used. |
|---|
Research propositions and rival explanations
| Research proposition | Affirmative rival or null | Distinguishing measure | Executable rule |
|---|---|---|---|
| H1A. Issuer-specific transmission | A qualifying material AI-system issuer drawdown remains an equity event without material credit, funding, collateral, clearing, or forced-flow effects. | Matched intermediary spreads, funding, collateral, dealer flows, payments and clearing. | Support requires the credit-spread breach plus either funding/collateral or flow amplification. Null requires all three to remain below threshold. |
| H1B. Common hardware-substrate shock | A foundry, HBM, packaging, interconnect, or platform shock impairs operations but is absorbed without material financial spillover. | Multi-issuer operating shock, broad index drawdown, credit/funding effects. | Support requires all three conditions. A common operating shock without credit/funding transmission supports the null. |
| H2. Infrastructure-finance bridge | Cancelled or underused AI infrastructure is redeployed or absorbed without material lender, utility, supplier, municipal, or ratepayer loss. | Principal impairment, spreads, guarantees, rate effects, supplier revenue and redeployment. | Support requires lender/project-credit breach, or both utility/ratepayer and supplier/project-finance breaches. |
| H3. Valuation-funded integration | Stock-financed application ownership destroys neutrality, retention, or economic value. | Retention, neutrality, voluntary owned-model use, margin, ROIC versus WACC, impairment. | Support requires retention, neutrality, no material impairment, and ROIC or operating improvement. Failed, blocked, or abandoned deals remain recorded outcomes. |
| H4. Power-latency constraint | Efficiency and supply outpace demand while latency and service levels remain intact. | Demand versus efficiency, P95/P99 latency, SLA misses, multi-site constraint. | Support requires demand to outrun efficiency plus latency/SLO breach and multi-site evidence. |
| H5. Hyperscaler intermediation | Frontier labs retain direct contracts, data authority, and durable margins despite cloud routing. | Spend share, master-contract ownership, substitutions, channel control and margin. | Support requires at least two of three pre-specified control-plane thresholds. |
| H6. Managed sovereignty response | Portability investments do not improve continuity or cost-adjusted recovery. | Architecture portability, recovery time, fallback success and TCO premium. | Support requires portability, RTO improvement, and TCO premium no greater than 15%. |
| H7. Open-weight rent redistribution | Open weights remain economically marginal or fail to alter concentration and rent location. | Production share, secure task cost, closed-model premium and adjacent-layer rent shift. | Support requires at least three of four thresholds; failure of model share, premium and rent-shift tests supports the null. |
Contents
| 1 | The event: one completed IPO, one signed merger, and two listing scenarios |
|---|---|
| 2 | A transmission taxonomy: financial risk, macro amplification, operational criticality, and strategy |
| 3 | Index inclusion, concentration, and financial transmission |
| 4 | The physical substrate: hardware, data centers, electricity, and capital |
| 5 | From tokens to outcomes: the frontier-lab business model |
| 6 | Enterprise reorganization and the contest for the workflow |
| 7 | Sovereign and open-weight AI as a counterstrategy |
| 8 | The United States, China, and the multipolar sovereignty contest |
| 9 | Productivity, labor, and the distribution problem |
| 10 | Competing explanations, external critiques, and adjudication tests |
| 11 | Four scenarios for 2030 |
| 12 | Executable hypotheses, scenario bands, and the monitoring protocol |
| 13 | Implications for investors, boards, policymakers, and workers |
| 14 | Conclusion |
| Appendix A | Definitions, qualification tests, and analytical identities |
| Appendix B | Baseline monitoring dashboard |
| Appendix C | Scenario-generator parameters and reproducibility specification |
| Appendix D | Source symmetry and complete revision ledger |
| References | Primary sources and research literature |
Method and scope
This is a pre-registration baseline for a theory-building research program, not an investment recommendation or a prediction that any specific rescue, market collapse, merger outcome, or vertical-integration strategy will occur. It distinguishes observed evidence, prospective cases, and conjecture. Public filings establish event facts, capital structure, governance, and disclosed obligations. Company statements establish strategic intent, not the soundness or future success of that strategy. Independent operational evidence is required before a transaction, product, or infrastructure program is used as confirmation of a general mechanism.
Industry coalition statements are treated as evidence of policy preference, commercial alignment, and strategic narrative. A broad signatory list can demonstrate that firms across chips, cloud, models, applications, cybersecurity, and capital share an interest in an open ecosystem. It cannot establish that the coalition's claims about cost, safety, competition, sovereignty, or national leadership have been realized.
The unit of analysis is the system formed by frontier models, hyperscalers, accelerator and equipment suppliers, foundries, memory and packaging providers, electricity and data-center infrastructure, capital markets, enterprise workflows, and the state. SpaceX remains an extreme boundary and mechanism-revealing case, not a representative blueprint. The protocol distinguishes primary frontier firms from material AI-system issuers. NVIDIA is analyzed symmetrically as both a hardware supplier and a material AI-system issuer. Generalization requires comparable behavior across firms, explicit common-shock controls, and causal mechanisms that survive failed transactions, neutral applications, custom-silicon substitution, and hyperscaler-controlled distribution.
Related work and benchmarking
Version 1.2 positioned this protocol against four external frameworks rather than presenting its method as sui generis. The IMF identifies vendor concentration, common models and data, herding, market fragility, and cyber risk as capital-market concerns, based partly on outreach to 27 industry stakeholders; its unit of analysis is mainly the provider category rather than named cross-layer firms. JPMorgan Chase evaluates U.S.-China AI competition across seven dimensions and explicitly rejects a single metric. The Cloud Security Alliance maps concentration and supply-chain-security risks across hardware, memory, packaging, frameworks, model distribution, and cloud. When We Crash provides a non-peer-reviewed structural analog for scenario generation, sensitivity analysis, and historical testing. This protocol uses only a historical applicability and rule-coverage audit; it does not claim backtest performance (IMF 2024; JPMorgan Chase Center for Geopolitics 2026; Cloud Security Alliance 2026; When We Crash 2026).
| Work | Primary contribution | Method or evidence | How this protocol differs |
|---|---|---|---|
| IMF GFSR, Ch. 3 (2024) | AI in capital markets; vendor concentration, herding, market fragility and cyber risk. | Market analysis plus detailed responses from 27 stakeholders. | Names firms and follows capital, hardware, infrastructure and workflow links, while retaining narrower financial-system tests. |
| JPMorgan Chase Center for Geopolitics (2026) | Seven-dimensional U.S.-China assessment: policy, hardware, models/software, energy, finance, socioeconomics, military/security. | Systemic country comparison using quantitative and qualitative indicators. | Adapts the dimensions to firms and regions without totals or a composite rank. |
| Cloud Security Alliance (2026) | Concentration and security risk across the AI development stack. | Research note drawing on public and secondary estimates plus incident evidence. | Uses its figures as reported estimates, separates market concentration from verified exposure, and adds portfolio and continuity tests. |
| When We Crash (2026 working paper) | Scenario generator, sensitivity ranges, selected-episode dashboard audit. | Factor-copula Monte Carlo and preliminary backtests; not peer reviewed. | Uses hypothesis-specific generators and fixed decision states; does not copy its probabilities or aggregate H1-H7. |
Table M.2. External benchmarks clarify the protocol's scope, method, and limits.
Revision record and selection correction
External review identified a directional omission in the prior baseline: it discussed SpaceX scale and acquisition capacity without presenting the company's 2025 loss, third-party valuation countermarks, actual index weight, an independent estimate of pre-existing claims on proceeds, or the inaugural bond issuance. The v1.0 baseline restored those observations before the first quarterly adjudication. Recording the correction prevents later versions from silently changing the baseline after outcomes are known.
Protocol v1.1 added the hardware-substrate amendment. Version 1.2 corrected TSMC's announced U.S. investment baseline, replaced the 100,000-GPU figure as a current SpaceXAI anchor, added the disclosed Anthropic and Google compute agreements, split H1 into issuer-specific and common-substrate tests, admitted NVIDIA to the issuer test, and added the six-state logic, support-pattern simulation, and related-work benchmark.
Version 1.2.1 is a bounded final patch. It corrects adjudication precedence, defines event-level and hypothesis-level aggregation, separates simulation from evidence availability, completes missing H1-H7 units and denominators, relabels the historical exercise, adds exact NVIDIA holdings traceability and contract-ratio caveats, requires source symmetry, adopts a two-tier company taxonomy, reconciles hardware and power through migrating constraint regimes, and records every material factual and methodological change in Appendix D. The baseline is frozen after this release; later changes require a prospective versioned amendment.
| Evidence class | Permitted use | Non-permitted inference |
|---|---|---|
| Public filings and transaction notices | Offering status, terms, share classes, governance, obligations, and disclosed risk factors | A signed deal or stated rationale is not evidence of successful integration or industry-wide inevitability. |
| Company disclosures | Management intent, product direction, reported scale, and claimed strategic logic | Self-description does not validate returns, customer value, neutrality, safety, or causal effect. |
| Industry and coalition advocacy | Commercial alignment, policy preference, ecosystem strategy, and the claims participants want policymakers to accept | The number or prominence of signatories does not validate lower cost, stronger security, decentralization, or realized diffusion. |
| Regulators, agencies, and index providers | Rules, system exposure, infrastructure status, and policy response | A dated aggregate statistic is not a stable structural parameter; source, date, and methodology must travel with the number. |
| Independent empirical and customer evidence | Productivity, retention, reliability, labor, security, and process economics | Early or context-specific results require replication and matched comparisons. |
| Comparative cases and counterfactuals | Testing whether a mechanism appears beyond SpaceX and against neutral or failed alternatives | Case analogy cannot substitute for identification; selection and endogeneity remain. |
| Scenario analysis | Defining coherent outcomes and observable signposts | Scenarios do not validate a thesis unless classification rules and disconfirming observations are specified in advance. |
Table M.1. Evidence hierarchy and limitations.
Analytical boundaries
The protocol evaluates channel-specific transmission and strategic incentives. It does not predict a specific offering price, assume announced infrastructure is completed, or aggregate heterogeneous forms of importance into one score. Financial transmission, macro amplification, operational criticality, strategic importance, and resolvability remain disaggregated; the remedy for vagueness is a threshold within each channel, not a composite index. Concentration statistics are dated observations rather than constants. The SpaceX-Cursor agreement remains pending at the evidence cutoff and cannot count as successful integration until it closes and produces independently observable post-close results. Chapter 5 contains the one canonical treatment; all other mentions are event facts or cross-references and create no additional evidentiary weight.
Limits of inference and case selection
SpaceX combines founder control, launch, communications, defense, compute, a frontier model, and a pending application acquisition in a structure that is unusually difficult to replicate. The case is valuable because it exposes mechanisms at their most concentrated, but it creates a selection risk: a theory built around the most extreme firm can mistake anomaly for equilibrium.
Evidence is classified as observed, prospective, or conjectural. Observed evidence records completed events and measured outcomes; prospective evidence records signed transactions, filings, or contracts with unknown effects; conjecture identifies mechanisms for later comparison. Outcomes cannot be counted as support by relabeling the channel, scenario, or perspective.