Implications for investors, boards, policymakers, and workers

The same system produces different duties for capital providers, enterprise operators, frontier firms, utilities, regulators, governments, educators, and workers.

The thesis does not imply one universal action such as buying infrastructure, avoiding mega-cap equities, or building a local model. It implies a diligence framework. Each stakeholder should identify the dependency it is creating, the balance sheet that bears the downside, the evidence required to justify scale, and the fallback available if capability, price, power, or policy changes.

For public-equity and private-market investors

Investors should resist the comforting idea that every "picks-and-shovels" asset is safer than the application or model layer. Utilities, data centers, cooling providers, chip suppliers, and power producers can benefit from structural demand, but they can also be exposed to customer concentration, technological obsolescence, take-or-pay risk, rate disputes, or overbuilding. The safest-looking layer can become the most leveraged layer.

Underwrite independent demand: Separate customer cash from strategic-investor funding, prepayments, barter, and related ecosystem transactions.

Underwrite acquisition currency: Compare stock-financed deal value with market capitalization, implied dilution, target cash flow, customer retention, model neutrality, integration cost, and expected return on the acquired workflow.

Normalize capital intensity: Treat accelerator refresh, depreciation, power, networking, and facilities as recurring economic costs rather than temporary growth spending.

Map contract duration to asset duration: A 20-year infrastructure asset backed by a short or cancellable AI commitment has a maturity mismatch.

Track value per compute: Revenue and gross profit per energized MW, successful task, and dollar of invested capital are more informative than tokens or users alone.

Track rent migration: Open-weight adoption can compress model pricing while increasing demand and margin at the chip, cloud, control-plane, data, or application layer. Measure spending and gross margin by layer rather than calling the whole stack "open."

Stress the index channel: Model drawdowns, lockup expiries, follow-on issuance, and rebalancing rather than assuming passive demand is permanent.

Map future share supply and control: Track lockups and waivers, registration rights, options, vesting, restricted awards, acquisition shares, follow-ons, dual-class votes, insider sales, and key-person dependence.

Price governance and succession: Understand voting rights, controlled-company exemptions, board independence, mission structures, related-party arrangements, founder dependence, succession, and the ability to separate or sell critical assets.

Distinguish the technology from the security: A transformative technology can coexist with overvalued equity and poor project returns.

Interpret impairment correctly: Goodwill and intangible write-downs usually recognize a deterioration in expected value; they do not by themselves create the underlying economic loss.

For boards, chief executives, and chief financial officers

Boards should treat production AI as an operating-model and concentration decision, not merely an IT procurement. The most important questions concern workflow authority, liability, resilience, and economics. A board that approves enterprise-wide agents without knowing their permissions, failure modes, fallback, and fully loaded process return is making an unpriced operational bet.

Create a critical-workflow inventory: Identify which processes use which models, clouds, tools, data, and human escalation paths.

Set dependency limits by layer: Define acceptable concentration separately for cloud/control plane, model, application, data, identity, and compute.

Own evaluation: Retain customer-specific test sets, outcome labels, red-team results, and acceptance thresholds.

Contract for portability: Secure access to logs, configuration, prompts, tools, memory, and process data required to migrate.

Protect model choice: When an application is owned by a model or infrastructure provider, contract for transparent defaults, access to alternatives, data-use limits, and the ability to preserve workflow history during migration.

Choose an operating model explicitly: Decide whether single-vendor consolidation, hyperscaler-managed plurality, an independent managed control plane, or self-management best fits the organization's capability and risk.

Treat open weights as an option, not a label: Compare license rights, provenance, portability, deployment burden, security evidence, and secure task cost with proprietary alternatives at equivalent quality and service levels.

Measure realized economics: Require process-level cash, quality, cycle-time, risk, and labor metrics after integration and supervision costs.

Govern action rights: Use least privilege, transaction limits, separation of duties, approvals, and revocation for agents.

Fund the human system: Preserve expertise, apprenticeship, exception operations, and accountable owners rather than assuming automation removes them.

For frontier labs and platform companies

Frontier firms that seek public-market scale will need a disclosure and governance model suited to infrastructure-like and conglomerate dependence. Investors and regulators will ask about model capability, capacity commitments, stock-funded acquisitions, related-party and intersegment economics, customer concentration, power availability, depreciation, application neutrality, safety incidents, and outcome liability. The companies can reduce the risk of punitive regulation by making economics, substitutability, and continuity visible before a crisis.

Disclose the capacity ladder: Announced, contracted, financed, under construction, energized, utilized, and economically productive capacity.

Clarify revenue quality: Separate consumer, API, enterprise seat, agent, transaction, advertising, and outcome revenue, including related-party activity.

Disclose acquisition integration: Report consideration and dilution, retention, model mix, product neutrality, intercompany pricing, cross-subsidy, integration costs, and the operating return attributed to acquired applications.

Define the partner boundary: State which workflows the platform will own and where software vendors, integrators, and service providers remain primary.

Build resolution architecture: Separate critical services and data sufficiently to preserve continuity during restructuring or ownership change.

Price liability explicitly: Use bounded action scopes, service levels, insurance, and contract design rather than hiding delivery risk inside gross margin.

Support portability: Common interfaces and exportable logs can increase trust and market size even when they reduce short-run lock-in.

State model-derived training terms precisely: Distinguish evaluation, adaptation, and legitimate follow-on development from unauthorized extraction through clear licenses, provenance, technical controls, and proportionate remedies.

Report failure: Publish meaningful incident, reliability, and evaluation information rather than capability metrics alone.

Publish safety cases proportionate to access: Report misuse, malfunction, autonomy, cyber, biological, and control evaluations together with permissions, containment, incident response, and release decisions.

For utilities, grid operators, and state regulators

Large AI loads can finance valuable infrastructure and improve regional economics. They can also create stranded costs and reliability pressure. The correct response is neither an automatic subsidy nor a blanket moratorium. It is contract and queue design that rewards credible, flexible projects while assigning cost to the party that creates it.

Use milestone-based queue rights: Require deposits, site control, equipment progress, and escalating financial security.

Protect ordinary customers: Large loads should fund dedicated facilities, exit risk, and incremental capacity unless broader benefits are demonstrated.

Price environmental and community effects: Measure water, emissions, backup generation, land, heat, transmission, local infrastructure, tax benefits, and community commitments instead of treating them only as permitting risk.

Value flexibility honestly: Compensate technically and contractually verifiable curtailment, geographic shifting, and demand response.

Plan portfolios, not single solutions: Use transmission, generation, storage, efficiency, firm fuel, and flexible demand together.

Coordinate across jurisdictions: Prevent duplicated reservations and ensure that simultaneous large-load plans are physically consistent.

Publish standardized status: Distinguish inquiries, studies, agreements, construction, energization, and actual load.

Test failure cases: Model customer bankruptcy, downsizing, or relocation before approving socialized investment.

For financial and operational regulators

The Financial Stability Board's work on AI concentration and responsible adoption points toward a cross-sector approach rather than a new capital regime based on firm valuation (FSB 2026). Authorities should map both financial exposures and critical-service dependencies. The objective is to detect when market concentration is becoming leveraged or operationally irreplaceable.

Collect exposure maps: Bank and nonbank credit, derivatives, collateral, project vehicles, guarantees, insurer holdings, and pension concentration.

Identify critical third parties: Models, clouds, data centers, identity services, and agent platforms whose failure would affect many regulated institutions.

Review vertical conduct: Monitor tying, bundling, default placement, discriminatory model access, data combination, intercompany pricing, and whether acquisitions make critical services harder to separate or resolve.

Run channel-specific combined stress tests: Pair equity drawdowns with credit and funding exposures, operational outages, cyber incidents, power constraints, and project cancellations, then report which core function is impaired.

Require recovery and exit: Regulated institutions should demonstrate migration or fallback for critical AI services.

Prepare resolution playbooks: Protect service continuity and customer data while preserving ordinary loss allocation.

Coordinate mandates: Financial, energy, cyber, competition, labor, and national-security agencies should share dependency information.

Avoid incumbent entrenchment: Scale obligations to criticality and systemic linkage so compliance does not become an entry barrier unrelated to risk.

For competition authorities and corporate-governance regulators

Cross-layer mergers should be evaluated as platform and vertical conduct problems, not only as overlaps between products. Under Section 7 of the Clayton Act, the central U.S. merger question is whether the transaction may substantially lessen competition or tend to create a monopoly. The 2023 U.S. Merger Guidelines direct agencies to examine competition between, on, and to displace multi-sided platforms and whether a merger can limit access to products, services, or routes to market. The relevant evidence for a compute-model-application combination includes defaults, ranking, price, latency, data access, interoperability, bundling, transfer pricing, and the ability of users and rival models to multi-home (U.S. Department of Justice and Federal Trade Commission 2023).

Do not presume either prohibition or clearance: Possible outcomes include unconditional approval, delay, nondiscrimination obligations, continued rival-model access, routing auditability, data-use separation, interoperability, governance separation, structural separability, divestiture, or prohibition. Remedy intensity should follow evidence of foreclosure and the feasibility of monitoring.

Review founder and related-party control: Require clear disclosure of voting power, board exemptions, succession, affiliated transactions, segment transfers, and conflicts that can move resources or risk across the group.

Preserve service separability: A merger should not make a critical application, model, communication system, or data asset impossible to transfer or continue during restructuring.

Measure post-merger conduct: Track model availability, defaults, pricing, release timing, routing overrides, data combination, customer retention, partner access, and the independent economics of each layer.

Map concentration beyond the model layer: Open-weight availability does not establish a competitive stack. Track chips, cloud, identity, control planes, data access, application distribution, and the location of margin and bargaining power.

For national economic and security policy

The United States should pursue a portfolio objective: abundant and reliable power, advanced semiconductor capacity, frontier research, capable open weights, shared datasets and evaluation infrastructure, secure cloud and networking, talent, competitive capital markets, trusted governance, and diffusion into ordinary firms. No single champion can substitute for the ecosystem. Protecting an incumbent from competition can weaken the very innovation and trust that create strategic advantage, while releasing capability without measured safeguards can create irreversible externalities.

Accelerate credible infrastructure: Modernize interconnection, permitting, transmission, equipment supply, and generation while protecting reliability and communities.

Build a plural open ecosystem: Support interoperability, open research, capable open-weight options, compute access, shared datasets and evaluations, cloud competition, and application-layer entry while measuring operational and safety outcomes.

Target strategic subsidies: Tie public support to measurable capacity, resilience, knowledge spillovers, and domestic or allied supply-chain value.

Invest in measurement: Build public statistics for compute, energy, AI prices, adoption, productivity, labor transitions, and incidents.

Coordinate with allies: Treat trusted compute, chips, standards, and research as a shared network rather than a purely national stockpile.

Engage multipolar sovereignty strategies: Build interoperable relationships with European public compute, IndiaAI, Gulf sovereign infrastructure, and allied manufacturing rather than treating every non-U.S. initiative as alignment with China.

Calibrate controls: Protect genuinely strategic capabilities while monitoring substitution, allied costs, and market fragmentation.

Calibrate model-derived training rules: Use targeted legal and commercial remedies for misappropriation while preserving legitimate evaluation, distillation, interoperability, and follow-on innovation where rights and safeguards permit.

Maintain legitimacy: Ensure that infrastructure benefits, environmental costs, data rights, and productivity gains are distributed visibly enough to sustain public consent.

For workers, educators, and professional institutions

The practical response to AI is not to compete with a model at every routine task. It is to develop judgment, domain accountability, system understanding, and the ability to supervise and improve automated work. Yet individuals cannot solve a structural transition alone. Employers and institutions must preserve pathways through which novices become experts and workers share in the value of the tools they help train and operate.

Teach verification and systems thinking: Workers need to understand evidence, model limits, tool permissions, downstream effects, and when to escalate.

Preserve deliberate practice: Replace disappearing junior tasks with simulations, rotations, critique, and supervised decisions.

Build portable skills: Data literacy, process design, security, evaluation, and domain expertise transfer better than mastery of one vendor interface.

Negotiate data and monitoring rights: Clarify how worker interactions, performance, and corrections are used to train or manage systems.

Share gains: Use wages, bonuses, reduced hours, employee ownership, and advancement to connect productivity with worker benefit.

Track differential exposure: Direct transition support toward occupations, regions, and demographic groups facing concentrated change.

A sequence for action

Table 13.1. A common action sequence across stakeholders.