Clean energy finance is often presented as a problem of capital supply. Institutional investors, infrastructure funds, banks, private-equity firms, corporations, and public agencies recognize the scale of investment required across renewable generation, storage, grid infrastructure, carbon management, distributed energy, and the increasingly power-intensive economy developing around artificial intelligence. Yet the presence of capital does not mean that capital can move efficiently into projects. Between an attractive concept and a financeable asset lies a demanding process through which technology, revenue, contracts, permits, engineering, operations, counterparties, and risk must be brought into a structure that investors and lenders can understand and trust.

As part of New York Energy Innovation 2026, Richard Deming, Founder and CEO of CEART Corp, joins Henning Stein, Partner at 1BusinessWorld, to examine why that process remains slow and how artificial intelligence can improve it. Deming’s perspective is shaped by approximately 25 years in renewable-energy development, preceded by a career in construction and real estate development. He approaches the problem not as a software entrepreneur entering an unfamiliar market, but as a project developer who has spent decades trying to move assets through permitting, contracting, diligence, financing, construction, and operation.

That experience leads to a central conclusion. The friction in clean energy finance does not arise because projects lack documents or because transactions lack professional advisers. It arises because hundreds of facts, agreements, responsibilities, conditions, and risks must be understood in relation to one another before anyone can determine whether a project is genuinely ready to receive capital. Lawyers, engineers, consultants, sponsors, lenders, investors, and counterparties may all review the same asset, but they often work through different systems, schedules, and definitions of readiness. The result is repetition, delayed discovery, inconsistent interpretation, and a transaction process in which clarity frequently arrives only after significant time and expense have already been committed.

Artificial intelligence can help reduce that friction, but Deming’s argument is more disciplined than the broad claim that a language model can read a data room and underwrite a project. General-purpose models can summarize documents, retrieve language, and produce fluent answers, but project finance depends on knowing which questions matter, what evidence should exist at each stage, how one contractual weakness affects another part of the transaction, and what conditions must be resolved before an asset can reach financial close. The decisive intelligence is therefore not the model alone. It is the combination of structured project context, sector-specific logic, professional diligence practice, specialized data, and the accumulated judgment of people who understand how energy projects succeed and fail.

Financeability Begins With Revenue Certainty

The development of a clean energy project may begin with technology, location, resource quality, or environmental potential, but financing begins with a more fundamental question. The capital provider needs to understand where the revenue will come from, how dependable it is, and whether it will remain available for long enough to support repayment of the investment.

Deming illustrates the point through his experience evaluating carbon-capture systems for biomass power plants. The technical opportunity receives serious attention. Engineers examine the technology, the development team spends money exploring the possibility, and the project moves well beyond a casual review. The barrier is not that carbon capture cannot work. The barrier is the absence of sufficiently dependable long-term revenue or offtake that would justify the expense and allow the investment to be underwritten.

When revenue depends on a market, that market must be sufficiently credible for investors to rely on it. When revenue is supported by an offtake agreement, the contract must be valid, executed, bankable, and long enough to support the capital structure. The counterparty must be capable of performing, and the conditions required for the agreement to become effective must be achievable within the development and construction plan.

Revenue certainty, however, is only the beginning of bankability. A strong offtake agreement has limited value if the project cannot secure its permits, retain site control, obtain interconnection, complete engineering, establish construction responsibilities, or put credible operations and maintenance arrangements in place. Each of these elements supports the project’s ability to reach operation and generate the cash flow described in its contracts and financial model.

This is why the diligence process cannot evaluate documents as independent files. A permit may appear valid until its timing is compared with a contractual milestone. An offtake agreement may appear attractive until its conditions are evaluated against the construction schedule. A site may appear suitable until interconnection, environmental, or access requirements are considered together. Project finance is a system of interdependent evidence, and the meaning of each document depends on the wider structure in which it sits.

The Cost Of Late Clarity

Deming’s experience shows how persistent this problem remains across different project sizes and stages of industry development. His early renewable projects encountered permitting, financing, and tax-credit complications even when the installations themselves were small. Years later, on larger and more sophisticated transactions, the same underlying pattern remained. A term sheet could establish broad commercial agreement, but detailed issues would emerge slowly as participants discovered clauses they interpreted differently, documents that were missing, or risks that had not been made visible at the beginning.

In one example, transactions that Deming believed should have progressed from agreement to close within a matter of months required roughly a year and a half. The delay was not attributed to one extraordinary failure. It grew from the accumulation of ordinary transaction friction, including unclear documentation, late questions, overlapping professional review, scheduling constraints, and the gradual discovery of issues that could have been identified much earlier.

The consequence is larger than additional legal or advisory expense. Late clarity affects which projects receive attention and which projects can support the transaction costs required to become financeable. Large assets may be able to absorb extensive diligence because the economics of the transaction are sufficient to carry the expense. Smaller or more distributed projects often face many of the same legal, technical, and commercial questions but have less margin available to pay for prolonged review.

Deming connects this structural problem to the development of data centers. A distributed model of smaller facilities could fit more naturally within communities, place less concentrated pressure on the grid, expand local tax bases, and spread employment and infrastructure investment across a wider geography. Yet smaller projects are difficult to develop and finance because the layers of diligence, negotiation, utility coordination, contracting, and professional cost can overwhelm the economics of a more modest transaction.

The finance process can therefore influence the physical structure of the market. Extremely large projects are not always favored solely because they are the most efficient or useful configuration. They may also be favored because only very large margins can absorb the time, uncertainty, and transaction expense required to bring a complex asset to close. Improving diligence efficiency could broaden the universe of projects capable of receiving serious capital consideration without weakening the standards applied to them.

AI Infrastructure Raises The Underwriting Stakes

The challenge becomes even more pronounced when infrastructure has a long development life but the technology it supports changes quickly. Data centers require major commitments to land, power, construction, equipment, and utility capacity, while the processors and computing configurations inside them may evolve much faster than the infrastructure itself.

Potential tenants need computing capacity but may resist long agreements because they cannot predict their requirements several years into the future. Capital providers, meanwhile, need enough contractual certainty to justify financing long-lived assets. The underwriting process must reconcile the duration of the infrastructure with the shorter life of the technology and cannot rely solely on the assumption that future demand will validate every project.

Deming sees continuing demand for artificial intelligence but remains measured about how the related infrastructure will develop. Smaller and more distributed facilities may ultimately provide a better fit for certain communities and electric systems, but developing them requires new forms of utility coordination, commercial commitment, and project structuring. The fact that the underlying demand appears real does not remove the need to establish how each asset will secure power, serve customers, earn revenue, and remain viable as technology changes.

This environment increases the need for disciplined underwriting rather than reducing it. Capital must understand the durability of demand, the quality of the counterparty, the availability of power, the development schedule, the infrastructure dependencies, and the exposure created by a technology cycle that may move faster than the asset itself.

AI Cannot Define The Underwriting Standard

The growth of large language models creates an understandable temptation to treat document ingestion as diligence. A data room can be uploaded, users can ask questions in natural language, and the model can produce summaries that appear comprehensive. For preliminary review, document navigation, or information retrieval, these capabilities can be useful. They do not, by themselves, constitute a reliable underwriting system.

Deming emphasizes that general-purpose models can produce confident answers even when the source material is incomplete or when the model has misunderstood the relationship among the documents. In legal work, inaccurate citations and fabricated authorities can be discovered relatively quickly because the underlying references can be checked. In financial diligence, the consequences of an incorrect interpretation may remain hidden until a project fails to perform years after the transaction has closed.

Confidentiality adds another layer of risk. Project data rooms contain proprietary contracts, financing terms, technical information, commercial assumptions, counterparty details, and other material that sponsors and capital providers cannot allow to circulate beyond the transaction. Deming describes CEART’s attention to security controls, SOC 2 requirements, and the need to ensure that client information is not used to train external frontier models.

The company can use smaller models in more controlled environments and selectively use frontier models where their capabilities produce a clear benefit, but the architecture is designed so that one provider does not become the foundation of the entire platform. This model neutrality reflects a broader point. AI capabilities will continue to change, and particular providers may become stronger or weaker at different tasks. A domain platform should be able to use the most appropriate model without allowing its core intellectual value to depend on the model’s general fluency.

Deming identifies the real driver as the context graph, the domain knowledge, and the understanding of how the components of an energy project connect. The model may accelerate extraction, comparison, classification, and analysis, but the underwriting standard must come from the industry knowledge embedded around it.

This distinction separates a specialized system from a model wrapper. A wrapper places an interface around a general model and asks it to interpret whatever documents are supplied. A domain system begins with an explicit understanding of the transaction, the project stage, the evidence required, the questions professionals would ask, and the relationships that determine whether an answer is acceptable, incomplete, contradictory, or material to the investment decision.

From Data Room To Decision Infrastructure

The starting point for this process is often a project data room, but data rooms are rarely as orderly as the term suggests. Some contain only a few early-stage documents. Others contain hundreds or thousands of files, including multiple versions of contracts, materials from several projects, scanned PDFs, maps, KMZ files, photographs, emails, engineering information, and documentation that may be either essential or irrelevant to the immediate decision.

Before meaningful diligence can begin, the system must understand what the data room represents. It must identify whether the materials relate to one project or several, determine the stage of development, separate current documents from outdated versions, and establish what evidence should reasonably exist at that point in the project’s life. An operating asset, a preliminary development opportunity, and a project approaching notice to proceed should not be evaluated against the same expectations.

CEART addresses this problem by transforming the data room into a structured dataset and context graph. The objective is not merely to extract text, but to establish relationships among documents, parties, assets, obligations, locations, project components, and development stages. A permit is connected to the relevant site and schedule. An offtake agreement is connected to its counterparty, term, conditions, and revenue assumptions. Engineering and construction documents are evaluated within the project stage and the contractual structure intended to deliver the asset.

Deming says this initial structuring process can take approximately two hours, even when the source data room is large and disorganized. That time creates a coherent representation of the project on which the detailed diligence process can operate.

The structured project is then evaluated through approximately 900 questions derived from the work of lawyers, engineers, consultants, developers, lenders, and investors. These are not open-ended prompts asking the system to provide a general opinion. They represent the recurring questions professionals use to determine whether the project satisfies specific requirements of legal, technical, commercial, and financial readiness.

The system also contains an understanding of what an acceptable answer should look like and how the answer affects the next stage of review. When the documentation supports the expected conclusion, that component of the project can be assessed more favorably. When the evidence is incomplete, inconsistent, or different from what a viable project would require, the system can pursue additional questions and adjust the analysis.

This branching structure is important because project diligence is conditional. A particular response regarding site control may lead to questions about term, assignability, or access. An offtake issue may require further review of creditworthiness, pricing, conditions, or termination rights. A construction concern may redirect attention toward engineering responsibility, completion guarantees, or schedule exposure. Agentic AI accelerates these paths, but the paths themselves are defined by professional knowledge.

Risk Scoring As A Roadmap

CEART translates the analysis into an overall risk score from one to 100, supported by component-level assessments covering areas such as site, offtake, engineering, construction, and other parts of project readiness. The purpose is not to suggest that a complex asset can be understood through one number. The value lies in making the composition of the risk visible.

A project may have highly credible offtake but an incomplete construction structure. Another may have strong site control and engineering but insufficient revenue certainty. By separating the components, the system allows participants to understand quickly where the project is strong, where it is incomplete, and where further work is required.

The result becomes both an assessment and a development roadmap. Instead of allowing problems to emerge slowly through months of fragmented review, the project team can see the principal gaps earlier and direct legal, technical, commercial, and financial work toward resolving them. The score does not replace the underlying evidence. It provides a navigable structure through which participants can move from a high-level assessment into the documents and conditions supporting each conclusion.

This structure can also improve communication among sponsors, capital providers, lawyers, engineers, and consultants. Each participant can work from a more consistent representation of the project rather than maintaining separate and partially overlapping views of its readiness. The result is not the removal of professional judgment, but a clearer foundation on which professional judgment can operate.

Industry Knowledge Becomes A Compounding Asset

The quality of a domain-specific system depends not only on its questions and logic, but also on the evidence against which project documents can be compared. General-purpose models are trained across an enormous range of public material, but much of the information required to understand market practice in clean energy finance is private.

Power purchase agreements, financing documents, development contracts, technical schedules, and project data rooms are rarely available at meaningful scale on the open web. A model may understand the general purpose of a power purchase agreement while lacking enough relevant examples to determine whether a particular provision is customary for the technology, jurisdiction, project stage, and transaction structure under review.

Deming says CEART began with a substantial research effort focused on publicly available agreements and project materials. The company then worked with early pilot partners across hundreds of data rooms. It does not retain identifiable client information, but it extracts anonymized data and benchmarks that strengthen the platform’s understanding of project structures and market practice.

At the time of the session, Deming describes an internal dataset of approximately 34,000 documents, with the number continuing to grow. The value of this dataset lies not simply in its volume. It lies in its relevance to the precise decisions the platform is designed to support.

CEART concentrates on the period from immediately before the term sheet through financial close. That focus narrows the problem to the stage at which sponsors and capital providers make consequential decisions about whether the project is sufficiently credible to justify continued investment and what must occur before funds can be committed.

The platform’s defensibility therefore comes from several compounding assets. It reflects decades of direct industry experience, a diligence framework derived from professional practice, a growing body of specialized documents and anonymized benchmarks, a structured understanding of project relationships, and a narrow focus on one of the most demanding stages of project finance.

This combination cannot be recreated simply by obtaining access to the same frontier model. Model access is widely available. The more difficult asset is the accumulated understanding of what to ask, what a viable answer looks like, how evidence should be connected, and which differences are material to project viability.

Faster Finance Through Better Clarity

The purpose of this operating model is not to eliminate professional judgment. Lawyers, engineers, consultants, lenders, investors, and developers remain essential because project finance involves negotiation, accountability, commercial judgment, technical interpretation, and risks that cannot always be reduced to standardized rules.

The opportunity is to change how those professionals use their time. When the basic structure of the project has already been organized, missing evidence has been identified, recurring questions have been applied consistently, and potential risks have been surfaced early, experts can focus on the issues that genuinely require their judgment. They spend less time locating information, coordinating repetitive review, or discovering late in the process that another participant has been working from a different understanding of the transaction.

Developers gain an earlier and more credible view of how capital providers are likely to assess the project. They can address weaknesses before entering an expensive transaction process, distinguish between documentation gaps and fundamental economic problems, and direct resources toward the work most likely to improve financial readiness.

Investors and lenders gain a more consistent basis for comparing opportunities. They can distinguish projects with strong fundamentals but incomplete preparation from assets whose revenue, contracts, or execution structure remain too weak to support financing. This does not remove risk, but it improves the visibility of risk and allows capital to be allocated with greater discipline.

Earlier clarity can also influence the composition of the market. When diligence becomes less expensive and less dependent on months of manual coordination, smaller and more distributed projects may become easier to evaluate. The benefit is not merely faster closing for large transactions. It is the possibility of expanding the range of projects capable of accessing sophisticated capital.

The leadership implication is that speed and rigor do not need to be opposites. Deals often move slowly not because the review is exceptionally thorough, but because information is fragmented, questions are duplicated, responsibilities overlap, and material issues are discovered too late. Better structure can make diligence faster precisely because it makes the process more coherent.

AI’s role is to provide scale and consistency to that structure. Deep industry knowledge determines the questions, relationships, evidence, and standards through which the technology becomes useful. The model accelerates the work, while the domain system makes the result relevant and trustworthy.

Clean energy finance will continue to require patient capital, strong counterparties, credible contracts, experienced professionals, and disciplined execution. What can change is the operating infrastructure through which those elements are evaluated and brought together. When a project can be understood earlier as a connected commercial and technical system, capital providers can make decisions sooner, developers can focus on the gaps that matter, and fewer resources are consumed by transactions that never had a credible path to close.

That is the deeper meaning of faster deal flow. It is not about moving capital before the risks are understood. It is about understanding the risks sooner, with greater consistency and enough context to determine what the project needs in order to become financeable.