Monitoring dashboard
A practical set of indicators for quarterly review, now populated with the July 24, 2026 baseline where observable. "Not yet observable" is an explicit instrument state, not missing decoration. Definitions must remain consistent across periods.
| Domain | Indicator | Why it matters | Baseline: July 24, 2026 | Cadence |
|---|---|---|---|---|
| Event and equity | Offering proceeds, primary vs. secondary shares, float, lockups, follow-ons | Separates capital raised from valuation and future supply | SpaceX: 638.9M shares at $135; $86.25B mechanical offer value; about $85.7B company-reported proceeds. Full future-supply schedule remains to be maintained. | Event/quarterly |
| Event and equity | Float-adjusted index weight and top-ten concentration | Measures portfolio transmission rather than headline market cap | SpaceX about 1.34% of Nasdaq-100 at inclusion; NVIDIA 7.93528% of SPY on 2026-07-23 (rounded fund-page display 7.94%); NDX top 10 = 52% at YE2025; S&P 500 top 10 = 36.4% on 2026-05-29. | Monthly |
| Event and equity | Options, securities lending, margin, structured-product exposure | Identifies leverage and forced-flow channels | No complete public aggregate baseline; first review will compile options OI, securities lending and fund filings. | Monthly/quarterly |
| Event and equity | Leveraged and inverse ETFs, options and structured-product rebalancing exposure | Separates flow amplification from ordinary unlevered index ownership | No standardized constituent-level baseline; compile fund assets, leverage targets, option OI and dealer gamma in Q3 2026. | Monthly/event |
| Capital structure | Corporate debt, revolvers, maturities, covenants, interest burden | Shows whether equity risk is migrating to creditors | $25B SpaceX senior unsecured notes priced June 23, 2026; five tranches, 5.35%-6.65%, due 2031-2056. | Quarterly |
| Capital structure | Project debt, SPVs, guarantees, take-or-pay and purchase obligations | Maps off-balance-sheet and infrastructure linkage | Consolidated SPV, guarantee and project-debt exposure not publicly observable at cutoff. | Quarterly |
| Demand quality | Cash from unrelated end customers | Tests whether demand is independent of strategic financing | Not separately disclosed across the three firms. | Quarterly |
| Demand quality | Production conversion, renewal, expansion, churn, workflow volume | Distinguishes pilots from durable adoption | No comparable public production-conversion, renewal and churn baseline. | Quarterly |
| Demand quality | Revenue mix: consumer, API, seat, agent, transaction, outcome, ads | Shows movement up the value-capture ladder | Not publicly disaggregated on a comparable basis. | Quarterly |
| Compute | Accelerator fleet by generation, provider, and location | Measures concentration, refresh risk, and flexibility | No reconciled fleet baseline. Historical: xAI Colossus 100k Hopper GPUs in 2024. 2026 SpaceX filings disclose agreements tied to ~325k NVIDIA GPUs for Anthropic and ~110k for Google; do not sum as a verified non-overlapping productive fleet. | Quarterly |
| Compute | Utilization and productive accelerator hours | Tests whether capacity is economically used | Not publicly observable. | Monthly/quarterly |
| Compute | Cost per successful task and retries | Links technical efficiency to business value | Public API prices exist; matched fully loaded cost baseline not yet compiled. | Monthly |
| Hardware platform | Productive workload share by accelerator, generation, provider and site | Measures NVIDIA, custom-silicon and alternative-stack concentration rather than ordered fleet alone | No comparable productive-fleet baseline. Disclosed anchors: xAI 100,000-Hopper Colossus in 2024; AWS Rainier 500,000+ Trainium2 in 2026; cross-firm fleet shares private. | Quarterly |
| Hardware portability | Porting time, engineering cost, retained performance, security and reliability across CUDA, Neuron, CANN and other stacks | Tests whether nominal multi-platform access creates real bargaining power and continuity | Not publicly observable. Establish matched workload and codebase panel before first portability adjudication. | Semiannual/event |
| Fabrication and packaging | Leading-edge logic, HBM and advanced-packaging qualified capacity, yield, allocation, lead time and geography | Identifies common upstream chokepoints and whether geographic diversification is production-equivalent | TSMC: 15.0M 12-inch-equivalent wafers shipped in 2025; 74% of wafer revenue at 7nm and below; Taiwan remains central while Arizona expansion proceeds. | Quarterly/annual |
| Data-center commissioning | Ordered chips, delivered racks, rack-ready MW, commissioned systems, productive accelerator hours and economically productive MW | Separates chip and gigawatt announcements from operable and revenue-producing infrastructure | OpenAI reported >5 GW Stargate capacity under development; standardized hardware-aware commissioning panel begins Q3 2026. | Monthly/quarterly |
| Power | Announced, contracted, financed, construction, energized MW | Prevents announcement inflation | Large announced pipeline exists; standardized project-status panel begins in Q3 2026. | Monthly/quarterly |
| Power | Firm vs. interruptible load; curtailment performance | Measures reliability burden and flexibility value | Not publicly observable across major sites. | Seasonal/annual |
| Power | Revenue and gross profit per energized MW | Tests value creation against physical scale | Not publicly observable. | Quarterly |
| Power | Interconnection security, deposits, dedicated-facility funding | Shows who bears cancellation and grid costs | FERC show-cause orders issued June 18, 2026; project-level security not standardized. | Event/quarterly |
| Model economics | API and effective inference price by quality tier | Measures commoditization and price pressure | Public schedules available; quality-matched baseline begins Q3 2026. | Monthly |
| Model economics | Serving cost, depreciation, power, gross margin | Tests whether cost falls faster than price | Not publicly observable. | Quarterly |
| Agent economics | Success, escalation, exception, rollback, and incident rates | Measures the true cost of autonomy | Not publicly observable with comparable denominators. | Monthly |
| Agent economics | Task margin and outcome-liability provisions | Tests whether workflow ownership is profitable | Not publicly observable. | Quarterly |
| Enterprise control | Critical workflows by model/cloud/provider | Measures operational concentration | No public cross-enterprise baseline. | Quarterly |
| Enterprise control | Exit tests, recovery time, fallback success, data portability | Measures substitutability and resilience | No public matched baseline; first enterprise panel planned for 2027. | Semiannual |
| Enterprise control | Private inference and open-weight share | Tracks sovereignty and bargaining response | Production share not publicly observable. | Quarterly |
| Security | Material incidents, prompt injection, unauthorized actions, data exposure | Tests trust and operational risk | Case reports exist; no exposure-adjusted denominator. | Continuous/quarterly |
| Security | Independent evaluation and red-team coverage | Shows whether controls are evidenced | Vendor and independent tests exist; coverage denominator incomplete. | Quarterly |
| Labor | AI-exposed hiring, entry-level share, hours, wages, promotions | Detects transition and apprenticeship effects | Public labor series exist; frontier-firm-linked baseline not yet compiled. | Quarterly/annual |
| Labor | Output and quality per paid hour | Measures realized productivity rather than use | No standardized cross-firm workflow baseline. | Quarterly |
| Distribution | Labor share, prices, consumer access, regional effects | Shows who receives the economic gain | Annual public data available with lag; July 2026 baseline not yet compiled. | Annual |
| Geopolitics | Export licenses, restricted hardware flows, domestic substitution | Tracks effectiveness and fragmentation | Advanced-chip controls and revised 2026 China license-review policy active. | Event/quarterly |
| Geopolitics | Allied compute, model, and standards adoption | Measures trusted ecosystem reach | JPMorgan reports installed DC capacity: U.S. 53.7 GW, China 31.9, EU 11.9, Japan/Korea 6.6, India 3.6, Middle East 1.1; installed is not equivalent to AI-ready or utilized. EU/India/UAE/Saudi programs tracked separately. | Quarterly/annual |
| US-China | Secure task cost, deployment speed, reliability by layer | Replaces benchmark-only comparisons | No matched cross-country cost panel; Stanford U.S.-China capability gap reported at 2.7%. | Quarterly |
| Systemic risk | Common-provider exposure among regulated institutions | Identifies correlated operational failure | Not publicly aggregated across regulated institutions. | Annual/stress test |
| Systemic risk | Channel-specific combined stress loss: financial, macro, operational, and strategic outcomes reported separately | Prevents one form of importance from rescuing a failed claim in another channel | No qualifying baseline shock; outcomes remain disaggregated. | Annual |
| Resolution | Critical-service separability and continuity plan maturity | Shows whether failure can occur without service collapse | No standardized public disclosure. | Annual |
| Public legitimacy | Rate impacts, tax contributions, jobs, community benefits, complaints | Tracks political durability of the build-out | Project-specific data exist; no standardized national baseline. | Annual |
| Event and equity | Stock-financed acquisition value, implied dilution, deal spread, approval and closing status | Measures use of valuation as acquisition currency and separates signed transactions from completed integration | SpaceX-Anysphere: signed $60B all-stock agreement; expected Q3 close; pending approvals. | Event/weekly |
| Vertical integration | Owned-model share inside acquired applications; model-neutral access; user and developer retention | Tests whether distribution converts into durable model demand without destroying application value | Pre-close; no post-close model-mix or retention baseline. | Monthly/quarterly |
| Capital allocation | Segment transfers, cross-subsidy, acquisition costs, goodwill and intangible returns | Reveals conglomerate economics and whether risk is migrating across otherwise distinct businesses | Not publicly disclosed. | Quarterly/annual |
| Event and equity | Monthly concentration range, index-return contribution, free float, lockups, registrations, vesting, insider sales, options and acquisition shares | Treats concentration and supply as time-varying rather than fixed | SpaceX initial NDX weight about 1.34%; future supply and lockups disclosed but require monthly schedule. | Monthly/event |
| Governance | Voting control, controlled-company status, board independence, related-party transactions, succession, restricted awards | Measures founder and key-person risk and public-shareholder influence | SpaceX founder voting control reported at 82.4%; mission/control structures differ at OpenAI and Anthropic. | Quarterly/annual |
| Latency and rebound | Sequential model/tool steps, retries, P50/P95/P99 completion, missed service levels, demand growth versus efficiency | Tests physical and commercial limits of agentic workloads | No matched production telemetry. | Monthly/quarterly |
| Environment and community | Water, carbon intensity, backup generation, land, heat, rate effects, local infrastructure, benefits and complaints | Measures social cost and political durability of data-center growth | Project-level disclosures incomplete and non-comparable. | Quarterly/annual |
| Hyperscaler intermediation | Managed-router share, direct versus intermediary billing, model substitutions, master-contract owner, margin by layer | Identifies the enterprise control point even when many models are used | Azure Foundry and AWS Bedrock routing available; contract and spend shares not public. | Quarterly |
| Model neutrality | Availability, release timing, default share, price, latency, overrides, BYO credentials, context portability, rival-provider terms | Operationalizes whether an acquired application retains meaningful provider choice | Cursor pre-close supports multiple providers; quantitative defaults and shares not public. | Monthly/quarterly |
| Frontier AI safety | Misuse and malfunction incidents, autonomy and access evaluations, permissions, near misses, safety cases, release restrictions | Tracks whether risk management keeps pace with capability and deployment criticality | International and company safety reporting exists; no common exposure-adjusted denominator. | Continuous/quarterly |
| Multipolar sovereignty | EU public compute, IndiaAI access and domestic models, Gulf full-stack capacity, allied fabrication and packaging resilience | Tests whether non-U.S./China strategies create meaningful capability and bargaining power | EU 19 factories/13 antennas; IndiaAI portal and Sarvam program; Stargate UAE 1GW; HUMAIN full-stack program. | Quarterly/annual |
| Competition and remedies | Merger status, interoperability, nondiscrimination, data-use limits, governance separation, divestiture or compliance | Tracks regulatory circuit breakers and post-merger conduct | SpaceX-Anysphere pending; no public remedy or final decision at cutoff. | Event/quarterly |
| Open-weight diffusion | Production inference share, matched secure task cost, deployment time, sector and public-institution adoption, model-layer concentration and price premium | Tests whether open weights create economically meaningful diffusion and model-layer competition | Unknown. Stanford closed-open gap 3.3%; Kimi K3 weights promised July 27 but not available at cutoff. | Quarterly |
| Rent migration | AI spending and gross-margin share by model, chip, cloud, control plane, data, and application layer | Tests whether model-layer openness decentralizes the stack or transfers value to a concentrated substrate | Not publicly observable by layer. | Quarterly/annual |
| Openness profile | Weights, architecture, code, license rights, training-data provenance, evaluation transparency, reproducibility, and portability | Prevents an "open" label from collapsing distinct rights, artifacts, and operating capabilities | Model-family differences documented in Chapter 7; no single binary "open" baseline. | Release/event |
| Open-weight safety | Misuse and malfunction incidents, vulnerability discovery, independent red-team coverage, patch time, concentration outages, and operating burden versus closed comparators | Tests distributed defense against irreversible capability and maintenance risk | International and company safety reporting exists; no common exposure-adjusted denominator. | Continuous/quarterly |
| Model-derived training | License and output-use terms, provenance, distillation or extraction disputes, legal rulings, technical restrictions, and remedies | Tracks the balance between investment protection, competition, follow-on innovation, and misappropriation | Anthropic alleges large-scale illicit distillation; coalition urges targeted remedies; no adjudicated ruling. | Event/quarterly |
| Adjudication | H1A-H7 state, endpoint and missing evidence | Makes the decision engine executable and prevents repeated-look cherry-picking | All hypotheses not triggered or awaiting future data at baseline; no formal verdict. | Quarterly/endpoint |
| Simulation | Hypothesis-specific conditional support-pattern bands | Shows assumption sensitivity under a triggered event and complete measurement; does not estimate observability or verdict probabilities | Base and 5th-95th support bands fixed in Table 12.4; no null/missing simulation and no cross-hypothesis aggregation. | Versioned amendment |
| Hardware concentration | Common dependency share and exposure-adjusted security incidents | Tests whether one platform or distribution layer creates correlated operational loss | CSA-reported figures entered as source-qualified estimates; organization-specific exposure not yet observable. | Quarterly/event |
| NVIDIA issuer risk | Index weight, customer concentration, spreads, options and common-substrate exposure | Applies the same issuer and transmission tests to the largest public hardware firm | SPY daily workbook: 7.93528% at 2026-07-23; fund page rounds to 7.94%. FY2026 direct customers 22% and 14%; indirect concentration disclosed but estimated. | Monthly/quarterly |
| TSMC diversification | Announced, built, qualified and volume-producing U.S. fabs and packaging | Separates geographic diversification from completed substitution | $265B total Arizona investment announced; commissioned contribution and timeline tracked separately. | Quarterly/annual |
| Evidence integrity | Source-symmetry ledger completeness | Shows whether load-bearing sources are represented with supportive and disconfirming evidence | Appendix D baseline completed; every future quarterly review must append changes and exclusions. | Quarterly/versioned |
Table B.1. Frontier-firm monitoring dashboard.
Quarterly review questions
Before scoring: confirm that the trigger, threshold, comparison group and data source were fixed before the event. Record any unavailable measure as "not observable," and publish versioned amendments prospectively.
What changed in the thesis? State the three strongest confirming observations and three strongest disconfirming observations.
Which announced projects became real? Move each capacity item through the status ladder and record cancellations.
Where did risk migrate? Identify whether exposure moved from equity to debt, utilities, suppliers, governments, or customers.
Did economics improve? Review task margin, revenue per MW, utilization, cash conversion, and verified enterprise ROI.
Where did openness move concentration: Compare open-weight production share and model prices with chip, cloud, control-plane, data, and application spending, margins, and switching power.
Did substitutability improve? Review model diversity, exit tests, control-plane portability, and recovery performance.
Who benefited and who paid? Review wages, hiring, prices, rate impacts, tax incentives, and regional effects.
Which scenario gained probability? Use signposts from Chapter 11 without assigning false precision.
APPENDIX C