Capital continues to pour into AI. But among LPs, GPs, and family offices with long-duration exposure, the terms of engagement are beginning to shift.
As AI systems move from experimentation into core economic and institutional functions, issues once treated as governance or compliance concerns are re-emerging as binding commercial constraints: procurement risk, loss of operational control, regulatory exposure, and fragile dependencies embedded deep in the AI stack.
What we learned from discussions with over 60 institutional LPs in Europe, North America, and Southeast Asia is that what is changing is not investor appetite for AI, but how alpha, risk, and durability are being understood and evaluated in relation to short- and long-term incentives. Early system-design choices around data control, model dependency, interoperability, and accountability are increasingly shaping downstream outcomes that affect not just individual companies, but entire portfolios.
In an environment marked by geopolitical uncertainty and tightening public-sector requirements, governments and enterprises are signaling that control, resilience, and agency are no longer preferences. They are conditions for deployment.
The result is a quiet but material shift. Architecture is becoming an investable variable. Control is becoming a source of competitive advantage. And AI systems designed without regard for governance, agency, and long-term operability are beginning to face constraints that capital alone cannot overcome.
The debate is no longer only about transparency, alignment, or safety. It is increasingly about strategic necessity. As AI systems move from the periphery into the core of operations, questions of dependency, control, and human agency are becoming central to value creation itself. Demand is rising for AI systems that support interoperability, pluralism, and human oversight. Not as an ethical preference, but as a condition for institutional trust, market access, and economic resilience.
For investors working at the intersection of technology and long-term capital, this moment feels familiar. A decade ago, clean energy shifted from a values-driven niche to an industrial imperative. What mattered was not intent, but whether new systems could compete on cost, reliability, and resilience at scale. AI now appears to be entering a similar phase: moving from a debate about norms to a question of system design, competitive advantage, and durability. Together, these dynamics are beginning to define a new value stack for the AI economy, one that investors are starting to take seriously.
This makes it timely to revisit a basic question: what does it mean to invest in âgoodâ AI?
As I have argued previously, this is not about investing with AIâusing AI tools to optimize outcomes in sectors such as climate, health, or education. It is about investing in the AI stack itself, treating AI as digital infrastructure whose design shapes value creation, information flows, and decision-making, with direct implications for markets, security, data agency, and the long-term depth and resilience of investable economies.
A better AI stack
Impact investment in AI is best understood as a systemic approach to catalyzing a dynamic pro-human AI stack across three interlinked layers: infrastructure, applications, and assurance.
At the infrastructure layer, impact investment focuses on the shared technical building blocks that shape how AI is built, deployed, and governed: data architectures, models, compute governance, and protocols. This includes investments in privacy-preserving machine learning that reduce exposure of sensitive data, open protocols and interoperability standards that prevent lock-in, and foundational mechanisms that allow people and organizations to define how their data, identity, and intent are represented and respected across systems.
Rather than seeking to replicate hyperscale dominance with smaller players, these approaches aim to enable pluralistic and resilient AI ecosystems that remain economically viable while supporting digital self-determination. In practice, this responds to growing demand from governments, enterprises, and end users for infrastructure that is secure, interoperable, and auditable by design, strengthening data agency while reducing systemic concentration risk.
At the application layer, the emphasis shifts to how AI systems are integrated into everyday services used by individuals, enterprises, and governments. These applications increasingly operate in high-stakes environments, processing sensitive personal, financial, or strategic data and shaping access to information, opportunity, and essential services. Impact-oriented approaches prioritize business models that align commercial incentives with accountability, user trust, and human judgment, particularly as AI systems begin to act with greater agentic autonomy on behalf of users.
In this context, applications that operate within clearly defined boundaries of intent, consent, and responsibility are better positioned to earn trust, navigate regulation, and scale sustainably. The question is no longer whether AI can scale, but whether it can do so without creating new forms of dependency, opacity, or systemic fragility.
At the assurance layer, impact investment targets the rapidly expanding ecosystem of AI governance, safety, and accountability technologies. As AI systems become more autonomous and embedded across critical functions, human oversight no longer scales without dedicated tools for monitoring, auditing, explainability, risk assessment, and compliance. These capabilities are essential not only to limiting misuse, manipulation, and unintended harm, but also to verifying that AI systems continue to operate within the expectations, constraints, and permissions under which they were deployed.
As regulatory frameworks mature globally, assurance is emerging as a market in its own right, supported by recurring enterprise budgets, procurement mandates, and growing links to insurance, liability, and risk management. What was once treated primarily as a cost is increasingly recognized as a precondition for deployment at scale and a source of long-term legitimacy.
Capital is catching up
Across regions, there is a growing and increasingly consistent demand for AI systems that are more open, resilient, and aligned with human agency by design. This demand is emerging from people, enterprises, and governments simultaneously, and it is beginning to shape markets in tangible ways.
Users are becoming more attentive to questions of control, opacity, and dependency as AI systems increasingly act on their behalf in consequential settings. Enterprises are incorporating AI-related security, governance, and trust considerations directly into procurement and vendor selection.
Governments, as major procurers of digital infrastructure, are reassessing strategic dependencies and sharpening requirements around control, interoperability, and resilience. This is particularly evident among middle-power economies whose competitiveness depends on open markets, and who are increasingly considering alternative AI systems that preserve interoperability, agency, choice, and, ultimately, digital self-determination.
Together, these shifts are moving AI governance from abstract principle to concrete market signal.
The open tension is no longer whether demand for better-designed AI systems exists, but whether utilization and adoption can catch up with the pace of capital deployment. Much of todayâs investment remains concentrated at the infrastructure layer, while application-level revenues, productivity gains, and procurement budgets are still uneven. This gap is increasingly shaping underwriting decisions, particularly in sectors where deployment depends on reliability, accountability, and long-term operating viability.
For LPs, the entry point has so far been risk and governance. Questions of fiduciary duty, systemic portfolio exposure, and long-term market concentration loom large. LPs are increasingly concerned not only about individual portfolio companies, but about compounding effects across labor markets, economic agency, and the future size and diversity of investable markets themselves.
At last yearâs AI Action Summit in France, a survey conducted by ReframeVenture and the Project Liberty Institute found that three out of four institutional LPs expressed serious concern about AI-related risks. Many already discuss how to apply AI risk frameworks such as NIST, AI Act or OECD standards in side letters and ongoing GP engagement, but many understand that the technology is scaling faster than principled approaches can catch up. A smaller but growing group is beginning to explore, from a systemic perspective, the investable opportunity set across AI infrastructure, applications, and assuranceâthough this remains early.
A small but growing group of impact-oriented GPs is beginning to treat the AI stack as a commercial opportunity rather than a concessionary one. Regulation, procurement rules, and enterprise risk tolerance are now defining real markets for AI systems built with human oversight, accountability, and user agency at their core.
For venture firms, the calculus is shifting under pressure from both capital providers and customers. In a tighter funding environment, LPs are pressing GPs to explain how AI-related risks and regulatory exposure are managed at the portfolio level, while governments, large companies, and end users are signaling demand for a different class of AI productsâparticularly in high-stakes settings where safety, accountability, and agency are requirements, not add-ons.
This is why ReframeVenture, ImpactVC and Project Liberty Institute launched a first-of-its-kind AI Due Diligence Tool for VCs in December 2025. A first cohort of specialist AI-focused funds, many still under USD 100 million, has begun to test these markets across infrastructure and applications, helping to de-risk impact investment in a better AI stack, a category that once sat at the margins of venture capital.
The role of these funds today echoes the early phase of climate-tech investing, before the emergence of billion-dollar platforms like Breakthrough Energy Ventures, which helped turn climate from a niche into a core venture category. In response, some firms are backing teams that build governance and control into their systems from the start, unlocking access to customers and procurement channels that remain out of reach for less mature offerings. They turn what once looked like a compliance burden into a source of competitive advantage.
Family offices occupy a distinct position in the AI investment landscape. Many are understandably cautious about the speed and scale of capital flowing into the sector, and reluctant to compete in increasingly large and fast-moving funding rounds where price, rather than judgment, often sets outcomes. At the same time, there is growing interest among family offices in more systemic approaches to AI investing, particularly where long-term value creation intersects with governance, risk management, and public interest.
With the ability to deploy capital across philanthropic, catalytic, and commercial instruments, family offices are well positioned to shape early markets that are likely to define the durability of AI returns. In environments where technological change, liability, and societal impact are tightly intertwined, aligning profit with public interest is increasingly a source of strategic advantage and capital protectionânot a concession.
Long-term value
Impact investment in AI remains early, but the direction of travel is no longer ambiguous. The market signals are real and increasingly consistent, even if the field itself is still forming. Public debate continues to oscillate between hype and fear, while more durable approaches are only beginning to translate into strategies investors can underwrite with confidence. What is missing is not demand, nor the scale of the opportunity, but a clearer articulation of how value is created, protected, and captured across the AI stackâand how those dynamics translate into concrete investment decisions.
That gap is now becoming the next agenda for investors. For LPs and GPs, the question is shifting from whether AI system design and governance matter to how they are priced. This shift is already showing up in practice: in diligence processes that probe data rights, model dependency, and downstream liability; in term sheets that reallocate risk through warranties, exclusions, and indemnities; and in procurement, insurance, and compliance requirements that increasingly determine which AI systems can be deployed at scale. These mechanisms remain uneven and emergent, but together they are beginning to draw a clearer line between speculative exposure and durable returns.
Measurement remains a parallel challenge. The socio-economic impacts of AI are diffuse, second-order, and often indirect, making them harder to quantify than emissions or energy intensity. Yet investors are converging on pragmatic proxies that matter commercially: reduced concentration and lock-in risk, improved audit and compliance readiness, clearer allocation of accountability, lower incident severity, and faster regulatory clearance in high-stakes markets. Translating concepts such as agency, accountability, and resilience into investable theses is not a solved problemâbut it is increasingly tractable, and increasingly demanded, by both capital providers and customers.
The broader implication is now coming into focus. As AI shifts from a technological novelty to a foundational layer of economic activity, the distinction between âimpactâ and âmainstreamâ investment is likely to erode. The relevant question for investors is no longer whether AI will matter, but which architectures, business models, and governance choices will remain deployable as AI systems become embedded in markets that operate under regulatory scrutiny, procurement discipline, and real liability.
Capital that helps build an AI stack capable of scaling while internalizing these constraints is likely to be better positioned to capture long-term value. Not because it optimizes for values alone, but because it preserves market access, protects optionality, and reduces the risk of obsolescence.
The next phase of opportunity in AI will not be defined by who can move fastest, but by who can remain viable longest. That is where the market is now headed.
_______________________________________________________________________________________
Paul Fehlinger is Senior Director of Policy, Investment & Innovation at Project Liberty Institute, where he leads global strategic initiatives at the intersection of capital, entrepreneurship, and governance to shape the future of the AI and data economy.
