Frontier Value Capture · inaugural baseline
When intelligence creates more value than tokens can capture.
Frontier-model economics will move beyond metered tokens and seats where AI creates economic value far larger than inference cost. The strongest opportunity is for frontier labs and domain incumbents to co-commercialize governed workflows and share downstream economics while the incumbent contributes scarce data, tools, validation, workflow control, and distribution.
The value-capture ladder
This is a classification of disclosed commercial structure, not a score. Unknown economics stay unknown. The ladder lets Strategic Signal test whether the market moves from access and metered inputs toward downstream participation and jointly commercialized products.
| Stage | Structure | Definition |
|---|---|---|
| Stage 0 | access only | Governed retrieval, data access, workflow access, or action execution with no separately disclosed AI-provider monetization structure. |
| Stage 1 | input metered | API, token, seat, training subscription, or license economics tied primarily to model/input access rather than downstream customer value. |
| Stage 2 | workflow product | A priced or co-developed AI workflow/service is commercialized, but disclosed economics are not tied to downstream revenue, milestones, royalties, savings, or outcomes. |
| Stage 3 | downstream participation | Disclosed economics participate in downstream commercialization or use through revenue share, sublicensing, transaction share, milestone, royalty, savings share, per-success fees, or another non-token mechanism. |
| Stage 4 | shared product economics | A frontier model provider and domain incumbent jointly commercialize a product or service and both participate in disclosed downstream commercial revenue. Profit sharing is not required. |
Opening anchors
Three observations establish the initial economic spectrum. Only GPT-Synopsys is classified here as shared-product economics between a frontier-model provider and a domain incumbent.
| Observation | Stage | Why it matters |
|---|---|---|
| OpenAI × Synopsys / GPT-Synopsys | Stage 4 | OpenAI pays Synopsys a training subscription and the parties share downstream GPT-Synopsys revenue; exact economics are undisclosed. |
| Harell Data governed training marketplace | Stage 3 | Data owners receive a revenue share on training jobs executed against their datasets; this is an ecosystem precedent, not a frontier-lab partnership. |
| OneMedNet real-world-data license | Stage 3 | The arrangement includes recurring sublicensing revenue on downstream licenses; this is an ecosystem precedent, not a frontier-lab partnership. |
Primary longitudinal forecast · SS-FVC-2026-001
Non-token value capture becomes a repeatable frontier-lab business model rather than an isolated GPT-Synopsys structure.
Mechanism. When model inference cost is small relative to measurable customer surplus, frontier labs have an incentive to exchange model capability and compute for participation in downstream economics instead of charging only for tokens or seats.
Confirmation. By 2027-10-02, at least three new qualifying frontier-lab arrangements across at least two industries disclose Stage 3 or Stage 4 economics, and at least one reaches Stage 4.
Invalidation. After a documented horizon-end search, fewer than three qualifying arrangements meet Stage 3 or Stage 4, or all disclosed frontier-lab arrangements remain Stage 0-2.
Confidence. moderate — GPT-Synopsys is a strong Stage 4 anchor and the broader registry already contains multiple Stage 3 ecosystem precedents, but the sample contains only one clear frontier-lab shared-product structure.
Industry predictions
Industries are the forecast unit. Company lists are predetermined candidate sets used to discipline future monitoring; appearing in a candidate set is not a prediction that the company will sign a frontier-model deal.
| Industry | Forecast | Candidate set | Confirmation test |
|---|---|---|---|
| Semiconductor design, engineering simulation, and digital twinsSS-FVC-IND-2026-001 · moderate-high confidence | This is the strongest near-term candidate for additional Stage 3-4 structures because economic outcomes are high value and unusually measurable through deterministic design, simulation, verification, and manufacturing constraints. | Cadence Design Systems, Synopsys / Ansys, Siemens Digital Industries Software, Dassault SystemesCandidate set, not predicted counterparties. | At least one new Stage 3-4 frontier-lab arrangement beyond the existing OpenAI-Synopsys deal is announced in this industry during the horizon. |
| Drug discovery, biotechnology, and clinical-data scienceSS-FVC-IND-2026-002 · moderate confidence | Milestone, royalty, training-run, or shared-service economics should emerge because successful model outputs can influence high-value research decisions and the domain already contains separately monetized data rights. | Tempus AI, Recursion Pharmaceuticals, Schrodinger, IQVIA, Thermo Fisher ScientificCandidate set, not predicted counterparties. | At least one new frontier-lab arrangement in this industry discloses Stage 3-4 economics during the horizon. |
| Cybersecurity and managed defenseSS-FVC-IND-2026-003 · moderate confidence | Outcome-linked structures should become viable where frontier models are embedded in live detection, validation, remediation, and re-testing loops, although liability will slow the move from workflow pricing to shared economics. | Palo Alto Networks, CrowdStrike, Fortinet, Cloudflare, Microsoft SecurityCandidate set, not predicted counterparties. | At least one frontier-lab cyber arrangement discloses Stage 3-4 economics tied to managed defense, remediation, incident outcomes, or downstream service revenue. |
| Financial information, capital-markets infrastructure, legal, and professional workflowsSS-FVC-IND-2026-004 · moderate confidence | Frontier labs should move from generic model access toward co-commercialized specialist workflow products, but regulated accountability and incumbent control make Stage 2 more likely before Stage 3-4 economics. | Bloomberg, LSEG, Nasdaq, S&P Global, FactSet, Thomson Reuters, RELX / LexisNexisCandidate set, not predicted counterparties. | At least two new frontier-lab specialist workflow arrangements are announced, with at least one disclosing Stage 3-4 economics. |
| Industrial operations, energy, equipment, and process optimizationSS-FVC-IND-2026-005 · low-moderate confidence | The value-capture gap is structurally large because small improvements in uptime, yield, maintenance, or process design can be economically material, but fragmented systems and safety constraints will make observable deals slower to emerge. | Siemens, Honeywell, Rockwell Automation, Caterpillar, SLB, Schneider ElectricCandidate set, not predicted counterparties. | At least one new Stage 3-4 frontier-lab arrangement tied to industrial yield, maintenance, process optimization, or energy operations is announced during the horizon. |
| Healthcare operations and clinical workflowSS-FVC-IND-2026-006 · low-moderate confidence | Frontier-model value capture should advance first through governed workflow products and later through outcome-linked economics because the economic surplus is large but reimbursement, safety, privacy, and clinical accountability make direct outcome sharing harder. | Epic, Oracle Health, Cognizant TriZetto, Optum, athenahealthCandidate set, not predicted counterparties. | At least two new frontier-lab healthcare workflow arrangements are announced and at least one moves beyond access or seat economics into Stage 3-4. |
Secondary watch
Commerce and payments: Shopify, PayPal, Amazon, Instacart, Stripe. Transaction-share economics are structurally plausible and the registry already shows governed commerce and payment rails, but per-task economic surplus may be lower than engineering, life sciences, cyber, and industrial domains.
Longitudinal discipline
Reassess with each monthly market outlook and whenever a newly approved canonical record reaches Stage 3 or Stage 4. The issued wording, candidate sets, horizon, confirmation tests, and invalidation tests remain preserved. Later evidence is appended as dated observations or published in a successor version.
Baseline: 47 canonical records = 18 strict data-rights records + 29 capability-only records; 42 are human-reviewed and 5 legacy strict records remain provisional.