An $18 billion contract sounds like a compelling investment proposition. Until someone asks whether the underlying infrastructure can actually deliver.
Consider Talen Energy’s agreement to supply electricity to Amazon Web Services from its nuclear facility in Pennsylvania. The arrangement has been described as representing approximately $18 billion in contracted revenue. Yet public documentation available in September 2026 suggests that the utility’s planning covers only around three-quarters of the contracted volume, while compensation for electricity not drawn could be considerably lower than the headline value. The discrepancy is revealing. A contract can be legally enforceable without translating into the cash flows investors anticipate. And the evidence needed to identify that distinction may sit in an entirely different set of documents.
This is a familiar problem in frontier technology and infrastructure investing. Valuations frequently depend on assumptions about regulatory approvals, industrial capacity, commercial commitments or financing milestones that receive surprisingly little independent scrutiny.
AI is beginning to change that. Not by eliminating investment uncertainty, but by making it considerably cheaper to investigate.
The economics of verification are changing
Historically, sophisticated due diligence has been expensive. Large infrastructure funds can commission extensive technical investigations, while family offices and smaller investment firms often lack the resources to examine every material assumption behind a transaction. The resulting asymmetry is significant. A €5 million investment in an industrial technology company may depend on the same grid connections, regulatory approvals or manufacturing constraints as a much larger transaction, without benefiting from comparable scrutiny.
AI is narrowing this gap.
Regulatory filings, permit registers, subsidy decisions and technical documentation can now be examined and cross-referenced in minutes rather than days. Information that was theoretically public but practically inaccessible is becoming available to a much wider range of investors. Yet faster document processing does not automatically produce better investment decisions. A subsidy award is not necessarily a cash disbursement. A projected commissioning date may depend on several counterparties whose own commitments remain conditional. The challenge is to establish what the available evidence actually supports, rather than what a financial model assumes it means.
And this requires looking beyond individual documents.
The Risks Between the Documents
Consider an industrial deeptech company preparing its first commercial manufacturing facility. The tech has been validated, customer demand appears credible, public funding has been awarded and an industrial partner has been selected. Each element looks reassuring when examined separately.
But the financing may depend on a subsidy disbursement, which requires eligible expenditures, which depend on equipment procurement, which itself depends on the availability of capital. Meanwhile, customer deliveries assume that the facility will be commissioned on schedule.
A delay at one point can undermine the economics of the entire project. This is particularly relevant to frontier technologies, where commercial success depends on coordinating scientific development, industrial capabilities, regulatory processes and capital across multiple organizations, sometimes in different jurisdictions. Traditional diligence tends to examine these components through separate legal, financial and technical workstreams. What is less systematically investigated is whether their underlying assumptions are mutually consistent.
The distinction matters for valuation.
A projected commercialization date is not simply a milestone on a timeline. It determines financing requirements, expected revenues and potentially the dilution required to reach profitability. A six-month delay can therefore have consequences far beyond the operational inconvenience it initially suggests. The objective should be to identify the few dependencies that genuinely determine the investment outcome, establish who controls them and assess the consequences if they do not materialize as expected.
Three questions to ask before signing:
• Contracted is not drawn. A signed volume is only revenue when someone uses it. Who is committed to take it, where, and from when?
• Whose date is it? A date the company sets is a plan. A date a grid operator, regulator or supplier has committed to is a fact.
• Approved is not paid. Grants, subsidies and permits arrive in stages, with conditions. What has been paid, and what is still conditional?
That is a different exercise from producing an exhaustive inventory of risks. It is also one that AI can make considerably more accessible.
Towards a different model for due diligence
My experience as a deeptech investor, European Innovation Council jury member and entrepreneur coach at ESADE has repeatedly exposed the same blind spot: investors scrutinize the technology, market and management team, while treating critical execution assumptions as relatively settled. The problem is rarely a complete absence of information. More often, nobody has been specifically tasked with connecting the evidence to the economics of the transaction.
This is what led me to develop DeepRadar, an independent practice that tests the assumptions an energy, infrastructure or deeptech valuation depends on. Part of it is free: anyone can see what the public documents on a company or project actually commit to. Anyone can examine what public documents establish about a company or project, without commissioning a lengthy research exercise. The reasoning is straightforward: automated document analysis is becoming a commodity. Its long-term value is unlikely to reside in summarizing information that everyone can increasingly access.
The real opportunity is to make independent verification available earlier in the investment process, particularly for investors who cannot justify traditional consulting engagements. Such investigations need not be exhaustive. A focused review of a handful of material assumptions can inform valuation, financing conditions or transaction structure before commitments become irreversible.
Importantly, the exercise still requires human judgment. AI can identify inconsistencies and surface evidence, but interpreting contractual dependencies, assessing execution credibility and translating uncertainty into investment terms remain matters of expertise. Independence also matters. Founders, advisers and transaction participants have legitimate but sometimes conflicting incentives. An external assessment can challenge assumptions that might otherwise survive simply because everyone involved has an interest in moving forward.
Over time, I expect this approach to become a more routine part of investment decisions, particularly in transactions where technical complexity has historically made independent scrutiny prohibitively expensive. The implications extend beyond efficiency. As the cost of verification declines, investors will have fewer reasons to accept material assumptions without testing them, and companies will face greater expectations to substantiate the milestones supporting their valuations.
AI is not making due diligence obsolete. It is making superficial due diligence increasingly difficult to justify.
—
About the contributor
Eden Djanashvili is the founder of DeepRadar. She has served on the European Innovation Council jury, invested in deeptech companies and teaches and coaches at ESADE.
One interesting consequence is that the cost of being wrong may remain unchanged, while the cost of discovering why you're wrong is falling dramatically. That should fundamentally change how smaller funds approach diligence. The question is whether investment practices will evolve as quickly as the tools.
Perhaps the biggest opportunity is to change the questions asked at investment committees. Rather than presenting 40 pages of risks, identify the three assumptions whose failure would materially change valuation or financing needs. That's a much more useful basis for an investment decision.