The deployment math nobody's writing about.

In November 2025 the Tropical Forest Forever Facility launched at COP30 Belém with $6.7 billion in sponsor commitments toward a $125 billion target. In January 2025 the Manulife Forest Climate Fund closed at $480 million. In April 2026 BTG Pactual Timberland Investment Group raised $1.24 billion for its Latin America Reforestation Strategy and added a $370 million first close on a new $1.5 billion core strategy a month later. Microsoft Carbon Removal contracted 45 million tonnes of CDR in 2025 alone, twice 2024 and nine times 2023. Frontier Climate's advance market commitment consortium crossed $1 billion. Nuveen launched a $3 billion private farmland REIT. The Symbiosis Coalition issued a 20-million-tonne removal RFP in April 2026 through Space Intelligence. The World Bank priced a $120 million Spekboom Restoration Outcome Bond the same month, with Nuveen, MetLife Investment Management and AllianceBernstein in the investor syndicate and Amazon as long-term carbon offtaker.

Combined, more than $50 billion has been formally committed to nature deals in the past 18 months.

The deployment ceiling, under current manual workflows, is somewhere between 8 and 12 deals per year per analyst. That is not a number a vendor invented. It is the number you get when you ask a senior analyst at any farmland fund, DFI, or carbon principal buyer in a candid conversation how many real-asset nature deals they actually close per year. Twenty-four to thirty-two weeks of due diligence per deal. Four to seven specialist firms in parallel — a GIS consultancy, a Brazilian or Indonesian legal counsel, a sustainability advisory like Pollination or Carbon Direct, a carbon ratings provider like Sylvera or BeZero, a physical-risk data provider like Jupiter Intelligence, often a TNFD-aligned biodiversity baseline firm. $400,000 to $700,000 in external consulting per $50-million-class transaction. Memo half-life of about twelve months before the parameters drift far enough to require a full re-baseline.

Capital is deploying at roughly 1.5% of the speed it was raised. The bottleneck is not the money. It is not regulation. It is not data — there has never been more open geospatial data, more methodology guidance, more disclosure frameworks. The bottleneck is the underwriting workflow itself. Specifically, it is the workflow's inability to compress the diligence cycle while preserving institutional-grade defensibility. Eight specialist verticals must each clear an underwriter, an investment committee, an external auditor, a customer-side legal review, and a regulatory disclosure layer. Today that requires five months of co-ordinating consultants. Tomorrow it requires a different architecture entirely.

This article is about that architecture.

What's actually durable

Look at the past twenty-four months of regulatory motion in nature finance. The Omnibus I package in February 2026 cut Corporate Sustainability Reporting Directive scope from roughly 50,000 to fewer than 7,500 covered entities and pushed the Corporate Sustainability Due Diligence Directive application to 2029. The EU Deforestation Regulation was codified as Regulation 2025/2650 in December 2025 but its application date has been pushed twice — now 30 December 2026 for large and medium operators, 30 June 2027 for small. The EU Green Claims Directive was withdrawn in June 2025. The EU Forest Monitoring Law was rejected by the European Parliament in September 2025. On 17 April 2026 the US interagency (OCC Bulletin 2026-13, Federal Reserve SR 26-2, FDIC FIL) replaced the fourteen-year-old SR 11-7 framework. The Cambridge over-crediting paper in Nature Communications, also April 2026, wiped over $1 billion in market value from first-generation REDD+ credits in ninety days. The EU AI Act binding date for high-risk classifications under Article 6 may slip to December 2027 under the Digital Omnibus proposal. The Tropical Forest Forever Facility's MRV protocol design, led by FAO, is still in draft.

In ten years almost every named regulation in that paragraph will have been revised, withdrawn, expanded, or replaced. Most of them more than once.

Now look at what does not change.

A Brazilian Cerrado parcel of 25,000 hectares, registered under CAR with INCRA federal title and a SICAR boundary polygon, is a durable object. Its land-cover history from 1985 onward, captured in MapBiomas and validated against AlphaEarth Foundations annual embeddings since 2017, is durable. Its IBAMA federal embargo status is volatile in the short term but durable in long-term archive. The ownership chain of Verde Cerrado Participações S.A. through three transfers since 1987, captured in Junta Comercial filings, is durable. The animal-movement permits (GTAs) flowing through INDEA-MT for the past 36 months are durable as data, even though the regulatory reporting requirements built on them change. Sentinel-2 imagery archived back to 2015 is durable. Planet Labs daily commercial imagery available through Earth Engine is durable. The methodology of VM0047 v1.0, as approved by Verra in May 2023, is durable as a specification even though new credits will eventually issue under VM0047 v2.0 or successor methodologies.

This is the principle. The deal underneath, the land underneath, the entity underneath, the supply-chain underneath, these are durable. The regulation, the framework version, the disclosure format, the threshold, these are volatile.

A nature-deal underwriting infrastructure designed for the regulation has to be re-cut every time the regulation moves. A nature-deal underwriting infrastructure designed for the deal, the geospatial truth, the entity record, the supply-chain provenance, the audit evidence, survives every regulatory cycle for the next decade.

The decisions an asset manager, DFI, insurer, or corporate carbon buyer makes in the next twelve to eighteen months about how to structure their underwriting workflow will outlive almost every regulation that pushed them to make those decisions.

That is the ten-year decision window.

Two traps

There are two predictable failure modes when institutions enter this window without a deliberate architecture.

The first is to wait until the deal arrives and then assemble the diligence chain from scratch. A live deal triggers a thirty-day rush to brief ERM or AECOM on land verification, a Brazilian or Indonesian law firm on cadastral and embargo screening, a sustainability advisory on supply-chain and counterparty diligence, a carbon ratings provider on methodology compliance, a physical-risk data provider on climate trajectories. Each engagement is bespoke. Each costs between $60,000 and $200,000. Each takes between four and twelve weeks. The aggregate is the $400,000–$700,000 number that defines most institutional nature-deal underwriting today. The memo gets to investment committee five months later. The investment thesis has already shifted by then.

The second trap is to buy a static rating product, a Sylvera AAA-class rating on a carbon credit, a BeZero AA rating, a Forest IQ portfolio screen, an MSCI GeoSpatial Asset Intelligence subscription, and treat the rating as the diligence. Static ratings are second-line opinions sold as data. They do not underwrite a specific deal. They do not produce an investment memo. They do not survive a credit committee asking "what is your confidence on the indirect-supplier supply-chain verification, and what evidence supports it?" Rating subscriptions cost between $30,000 and $300,000 annually depending on coverage. They are useful but not sufficient. The underwriter still has to assemble the workflow.

Both traps share the same root error: treating nature-deal underwriting as a series of one-off projects rather than as a recurring engineering problem with a designed architecture.

Design for the deal, not the framework

The right design driver is not the framework. It is the deal underneath.

A nature deal has five recurring components. A geospatial subject, the parcel, the polygon, the basin, the supply-chain catchment. An entity subject, the sponsor, the borrower, the counterparty, the beneficial owners. A financial subject, the structure, the term, the covenants, the offtake. A methodology subject, the standard the project applies under (VM0047, VM0048, VM0042, Gold Standard, IAPB biodiversity principles). A regulatory subject, the disclosure regimes that touch the deal (TNFD LEAP-FI, NGFS, NBRA, IFC PS6, CSRD ESRS E4, IFRS S2, SFDR, UK SDR, EUDR DDS).

Each of these five subjects has a durable existence independent of any single regulation. Each can be verified independently against multiple sources. Each carries a confidence level that can be expressed numerically and routed to human review when the level falls below a threshold. Each can be re-verified at any future date with the same architecture, regardless of which specific regulation is then in force.

This is the underwriting cascade we use at Kyroq. Five links. None of them optional. None of them defined by the regulation.

Deal candidate. Polygon (GeoJSON RFC 7946 or equivalent), entity (incorporation jurisdiction, beneficial-ownership chain), financial structure, claimed methodology, anchor LP.

Substrate verification. Multi-source cross-reference. For the polygon: AlphaEarth Foundations embeddings 2017–2026, Prithvi-EO-2.0 land-cover classification, MapBiomas Brasil v9.0 cross-validation, ESA Biomass mission AGB v6 (April 2025) for above-ground biomass. For the entity: CAR (Cadastro Ambiental Rural), SICAR, INCRA federal cadastre, IBAMA federal embargo list, state-level SEMA registers, federal Ministério Público prosecutorial records, Brazilian Junta Comercial corporate filings, sanctions lists (OFAC, EU, UK), NGO investigative archives (Global Witness, Mighty Earth, Greenpeace). For the supply chain: GTA animal-movement permits via INDEA-MT, Visipec, Cargill Maps disclosed sourcing zones, customs flows via SISCOMEX and TARIC, port AIS reconciliation. For the methodology: published Verra / Gold Standard / ART methodology specifications, ICVCM Core Carbon Principles category assessments, Cambridge over-crediting analysis cross-reference.

Confidence-tiered verdict. Per section. Per claim. Per evidence pin. Data quality tier A means high-quality source, multiple independent feeds, recent timestamp. Tier B means medium-quality source, primary plus inferences. Tier C means low-quality source, single feed or expert judgment under uncertainty. Aggregate section confidence is a weighted average. Any verdict below 60% routes automatically to senior underwriter review. Any verdict below 40% triggers external specialist engagement before the memo proceeds.

Framework-aligned output. The same underlying verdicts mapped to TNFD LEAP-FI Locate–Evaluate–Assess–Prepare, NGFS three-phase logic, UNEP FI NBRA workflow stages, NCFF credit-pack structure, IFC PS1–PS8 (especially PS6 critical habitat). If the regulation changes — Annex III scope expands, NBRA workflow evolves, TNFD LEAP-FI v2 publishes — only this output layer needs re-cutting. The verdicts and the evidence underneath do not change.

IC decision and monitoring covenants. Human underwriter signs the memo. Investment committee defends the decision. Covenant pack with cryptographic anchoring on every claim (SHA-256 + RFC 3161 timestamps) survives into the legal documentation. Monitoring runs continuously — typically weekly EO change-detection at 10m resolution, quarterly biomass re-baseline, annual full memo refresh. Covenant breach triggers escalation.

This cascade is the durable architecture. It is independent of which specific regulation is in force when the deal closes, when the annual review runs, or when the IC chairs the post-mortem six years from now.

Five deliverables that operationalise the architecture

If the cascade is the principle, here are the five artefacts that turn it into something a Chief Investment Officer, a Head of Sustainability, a Chief Risk Officer, and an internal Model Risk Management team can all defend.

Deal-type applicability map. A two-axis matrix. Rows are the regulations and frameworks the institution operates under (TNFD, NGFS, NBRA, IFC PS, CSRD ESRS E4, EUDR DDS, SBTi FLAG, ICVCM CCP, IAPB principles, SFDR Article 9, UK SDR labels, IFRS S2). Columns are the institution's deal types (Brazil cattle and soya, LATAM timberland, West African cocoa, Indonesian palm, Cerrado ARR, regenerative ag, REDD+, biodiversity credits, blue carbon, infrastructure-with-nature exposure). Each cell is a traffic light: relevant in the next 12 months, mid-term, watch-list, or not applicable. The map is boring. It does the most political work. It surfaces, on a single page, where actual exposure sits. It stops the all-too-common pattern of treating every framework as equally urgent for every deal type.

Deal-to-substrate mapping. For each deal type, walk the full cascade end-to-end. For a Brazilian Cerrado ARR deal: requirement (VM0047 + CSRD ESRS E4 + EUDR if cattle co-located) → KPI definition (annual VCUs, hectares restored, biomass trajectory, leakage discount, permanence buffer percentage) → data objects (parcel polygon, sponsor entity, baseline imagery archive, methodology document, supplier list, GTA records) → source systems (Kyroq substrate, customer-side ERP or fund management platform, external verification body) → owners (sustainability lead, deal-team analyst, IT or platform team). Do one deal type completely before generalising. A single fully walked cascade beats five abstract frameworks.

Substrate architecture. This is where most institutions over-spend. The instinct is to build a single sustainability data warehouse holding everything. That is the wrong architecture. The right architecture mirrors what works in financial-services data infrastructure: stable substrate stays in stable infrastructure; volatile reasoning gets its own layer. For nature underwriting that means commodity Earth-observation foundation models (AlphaEarth, Prithvi-EO-2.0, TerraMind, granite-geospatial-biomass) consumed via fine-tuned inference; primary-source administrative data (CAR, SICAR, INCRA, IBAMA, GTA, customs, AIS, court records) integrated via API where available and via structured ingestion where not; chain-of-custody graph primitives reconciling supply-chain transformation events; and a workflow layer above this substrate that produces framework-aligned outputs with cryptographic evidence pinning. The substrate is shared infrastructure that should be built once. The reasoning logic and the framework mapping are volatile and should be updated continuously.

Output layer and framework mapping matrix. This is the document a customer-side compliance team, MRM team, or external auditor reads first. Every section of every memo Kyroq produces maps explicitly onto TNFD LEAP-FI step, NGFS phase, UNEP FI NBRA stage, NCFF credit-pack slot, and IFC Performance Standard. The mapping is a deliverable artefact. When NBRA publishes a methodology update, the mapping changes — not the underlying verdicts. When TNFD LEAP-FI publishes v2, the mapping changes. The output layer is designed for change. Everything underneath it is designed for permanence.

Governance and MRM pattern. The single most under-engineered piece of every framework. For each domain (land verdict, counterparty, supply chain, carbon integrity, regulatory, water and biodiversity, physical climate, financial), three named roles. The Business Owner decides what the data has to do. Head of Sustainability, Head of NbS, Head of Climate Risk depending on the institution. The Data Steward operates the data day-to-day. Senior analyst, deal-team member, or external partner. The Technology Owner provides infrastructure. Customer-side IT, Aladdin Sustainability product team, or platform integration team. One source of truth per data object, mapped to one accountable role. That is what survives a regulatory shift, an MRM audit, and a regulator's question three years later about how a disclosure was calculated.

What survives a regulatory shift

If an institution builds the architecture this way, here is what is still standing when the regulation moves.

The deal does not change. Whether TNFD asks for one biodiversity metric or three, whether NGFS adds a fourth phase or refines the third, whether NBRA workflow evolves, the underlying parcel, sponsor, supply-chain, and methodology are the same.

The substrate does not change. AlphaEarth keeps publishing annual embeddings. Prithvi-EO-2.0 keeps running in orbit. CAR keeps registering rural environmental boundaries. INCRA keeps maintaining the federal cadastre. IBAMA keeps publishing the embargo list. The substrate compounds in value as more deals run through it.

The ownership model does not change. The Head of Sustainable Investing still owns nature exposure whether the report goes to a regulator, a customer-side ESG team, or an internal investment committee.

What changes is the output layer. The specific KPI definition. The disclosure format. The framework-version label. Those are the variables that should change easily. Re-mapping the output layer is the work of weeks. Re-architecting the substrate is the work of years.

This is also why a nature-underwriting architecture is, in the end, not really about underwriting in the narrow sense. It is about whether an institution treats nature-deal evaluation as a recurring engineering problem (good) or as a series of one-off project diligences (very expensive).

The 12–18 month sub-window inside the 10-year window

The ten-year decision window is the strategic frame. There is also a tighter sub-window inside it that matters specifically for institutions building underwriting infrastructure in 2026–2027.

On 17 April 2026 the US interagency replaced the SR 11-7 framework with SR 26-2, OCC Bulletin 2026-13, and the FDIC FIL. The new guidance contains a critical scope limitation: "Generative AI and agentic AI models are novel and rapidly evolving. As such, they are not within the scope of this guidance." The agencies have committed to publishing specific AI guidance through an RFI cycle that will not close until 2027 or later. Practically, this creates a twelve-to-eighteen-month window in which supervised institutions classify agentic AI as outside formal MRM scope by analogy to SR 26-2, and during which vendors who establish credibility through SOC 2 Type II, ISO 27001, ISO 42001, model card libraries, validation benchmarks, scientific advisory boards, and indemnification posture set the de-facto template for what an MRM-credible nature-underwriting agent looks like.

The EU AI Act Article 6 / Annex III high-risk classification covers creditworthiness of natural persons. It does not cover the kind of corporate counterparty, parcel-level, or supply-chain underwriting that nature-deal evaluation runs on. The Digital Omnibus proposal may push the August 2026 binding date to December 2027 for certain provisions.

For an institution thinking about the substrate-and-workflow architecture in 2026, this matters. The vendor selection, the architecture decisions, the MRM evidence package — all are easier to negotiate now than they will be in 2028 after specific AI guidance lands. The window is open. It will not stay open indefinitely.

Where to start this quarter

A pragmatic sequence we have seen work, both in our own deployment patterns and in institutional partners we have worked with on scoping conversations.

Two weeks: stakeholder alignment. Head of Sustainability, Head of Risk, deal-team lead, IT or platform owner, external counsel, MRM team if applicable. Get the sponsor named. Decide whether the nature-underwriting architecture is being designed inside an existing fund-management or Aladdin-class platform initiative, or alongside it as a discrete project.

Four weeks: Deal-Type Applicability Map plus one fully walked deal cascade. Pick the deal type that hurts the most across the most frameworks. For most LATAM-exposed allocators this is Brazil cattle plus VM0047 ARR; for West Africa allocators it is cocoa supply-chain plus CSRD ESRS E4; for SE Asia allocators it is palm-oil supply-chain plus IFC PS6.

Four to six weeks: MVP substrate-and-workflow architecture plus governance pattern, plus a clear interface specification with whatever ERP, PIM, fund-management, or Aladdin platform programme is already in flight.

Two to three months total. The output is a framework an institutional CIO can defend to the board, integrate into existing risk infrastructure, and re-use for every regulatory cycle over the next decade.

One final thought

The funds, DFIs, sovereigns, and corporates that handle the next regulatory cycle well will not be the ones with the biggest sustainability teams or the most expensive software. They will be the ones whose underwriting architecture was designed to absorb regulatory change rather than be re-cut by it.

The window to make that decision well is open right now — alongside Aladdin Sustainability product updates, fund-management platform decisions, and the MRM Trust Package and certification cycles that are already on the boardroom agenda. Once those decisions close, the architecture is locked. For most institutions, that is a decade.

Use the window, or get locked into the next ten years of one-off diligence projects.

Kyroq builds AI-native underwriting infrastructure for nature capital. The architecture described in this article runs in our platform. We work with farmland funds, DFIs, carbon principal buyers, and insurance NbS desks on deal-level diligence at institutional grade. Contact: contact@kyroq.com · www.kyroq.com