(Photo by iStock/Gerville)
For the past few years, artificial intelligence has felt almost miraculously accessible. Nonprofits, schools, public agencies, and social enterprises have been able to use advanced AI tools at little or no cost. Grant proposals, impact evaluations, program curricula, community outreach campaigns, and policy briefs are now routinely “co-written” with AI. This accessibility has been widely described as the “democratization” of AI. But it rests on a fragile foundation.
The reality is the current era of “free” or heavily subsidized AI is a temporary phase, not a stable feature of the technology. As AI shifts from experimental tool to core infrastructure, its underlying economics such as energy, hardware, privacy, and market power are beginning to assert themselves. That will have serious consequences for equity, public interest work, and the organizations that serve communities most affected by social and economic inequality.
The question is no longer whether AI will become a paid, utility-like service. It is whether social sector institutions will help design that future or simply be forced to adapt to it on unfavorable terms.
From Free Tool to Metered Utility
We are unlikely to see a single moment when AI “stops being free.” Instead, access will tighten over several years, in a pattern that follows other digital platforms. A plausible trajectory looks like this:
- 2025–Early 2026: Free and low-cost AI tools remain widely available, but advanced features start to cluster in paid tiers.
- Late 2026–2027: Free tiers become slower and more limited, while high-context, privacy-preserving, and domain-specific models move behind paywalls.
- 2028–2029: Frontier models and enterprise-grade systems become primarily subscription or usage-based services, negotiated like cloud contracts.
- 2030 and beyond: AI functions as a metered utility, essential, priced, and embedded in critical systems, much like broadband or electricity.
Why Free AI Won’t Scale
This transition is not driven by hype cycles alone. It is anchored in material constraints.
1. Energy demand. Data centers already consume a significant share of global electricity, and AI is now a major driver of growth. The International Energy Agency projects that electricity demand from data centers will more than double by 2030, reaching around 945 TWh—roughly equivalent to the current annual electricity consumption of Japan, with AI-optimized servers accounting for a large share of this increase.
At that scale, unlimited free access to powerful models becomes economically and environmentally unsustainable. Someone must pay for the energy, the grid upgrades, and the associated climate and water impacts.
2. Hardware bottlenecks. High-performance AI runs on specialized chips that remain scarce, expensive, and geographically concentrated. Demand for these accelerators is growing faster than supply. Scarcity requires prioritization, and prioritization almost always leads to pricing that favors large, well-resourced customers.
3. The hidden costs of privacy. As AI moves into health care, education, social services, and public administration, organizations will increasingly need systems that can handle sensitive information responsibly. This requirement will fundamentally alter the economics of AI. The existing tools designed for open, general use will be insufficient when organizations need to protect personal data, reflect local context, and meet regulatory obligations.
Organizations will need AI systems that can work with internal knowledge such as case notes, policies, or curricula, without exposing that information to public training pipelines. Achieving this level of privacy requires isolating data and customizing systems, which increases computing demands and reduces the efficiencies that keep large-scale AI inexpensive.
Privacy also determines how information is stored and retrieved during everyday use. When AI systems rely on stored representations of documents or conversations that contain sensitive information, those representations must be encrypted and carefully managed, adding further infrastructure costs. In addition, many organizations require auditable records of AI activity to ensure accountability and regulatory compliance. Secure, trustworthy AI is not simply a software feature but an infrastructure investment that increases costs and which many smaller organizations will find difficult to afford.
4. Consolidation and platform power. The AI ecosystem is consolidating around a small number of firms that control foundation models, cloud infrastructure, and distribution channels. This mirrors patterns documented in the broader digital economy: once platforms achieve scale and user dependence, monetization accelerates and bargaining power shifts away from end users and public-interest actors.
A New Kind of Digital Inequality
The social sector is at risk of a new digital divide, not over devices or internet access, but over who can afford high-quality, private AI tools. Many nonprofits serving marginalized communities already rely on AI for essential tasks yet operate with limited budgets. As free AI access diminishes, they will face hard choices between losing capacity or diverting funds away from direct services. Without intervention, well-resourced institutions will advance with powerful AI systems while under-resourced organizations fall behind, even as demand for their support increases.
What Social Sector Leaders Can Do Now
This transition is not a reason for organizations or leaders to disengage from AI; it is a reason to engage differently, with a focus on infrastructure, governance, and equity. Foundations, nonprofits, public agencies, and universities all have a role to play in shaping systems that are sustainable and accessible. The first step is to treat AI as core infrastructure rather than a free add-on. Organizations should create explicit AI budget lines, model different pricing scenarios over several years, and assess which programs now rely on AI and how deeply. Thinking of AI like connectivity or cloud storage prepares institutions for the moment when pricing or access changes abruptly.
The second step is to build shared capacity instead of attempting to develop isolated AI capabilities within every organization. Regional coalitions, anchor institutions, and library systems can pool resources to create shared AI labs and compute hubs that negotiate better terms with vendors, host open-source models for common tasks, and offer training and technical support to smaller organizations. This “rails, not rockstars” model reflects emerging global thinking on public interest technology and ensures that infrastructure, rather than one-off pilots, becomes the main vehicle for innovation and access.
Third, organizations can adopt multi-model, resource-aware practices. Not every task requires the most advanced frontier model; small, efficient models can handle routine drafting and summarization, while more capable systems can be reserved for complex or high-stakes decisions. Choosing providers that are transparent about energy use and that offer usage controls helps reduce both costs and environmental impacts while maintaining access to quality tools.
US nonprofits don’t need to build AI alone. The most practical near-term path is to join a regional “AI utility” anchored by universities, library systems, and major nonprofits that can broker shared computing resources, shared vendor terms, and shared support. National models already show how this works, however they may not cater to the unique needs of nonprofits: the NSF-led National AI Research Resource pilot coordinates access to compute, models, data, training, and user support across partners, while the community-owned National Research Platform shares compute power, storage, and networking across 50+ institutions. Layer this with shared governance using the National Institute of Standards and Technology’s AI Consortium standards to keep systems ethical and accountable for mission delivery.
Social sector leaders must also invest in AI literacy that explicitly includes the concepts of power, cost, and infrastructure. Training should go beyond prompt writing and help staff, partners, and community members understand how models are trained and deployed, what kinds of energy and resource demands they carry, and how to decide whether AI is appropriate for a given context. This deeper literacy strengthens both procurement and advocacy.
Finally, it is essential for the social sector to engage directly in policy and governance debates. As AI becomes foundational to work, education, health, and public administration, decisions about pricing, transparency, access, and data governance will shape equity for decades to come. Social sector institutions can help advance models such as public or cooperative AI utilities, push for regulation that protects affordable access, and support incentives for open-source and public-interest models. Their participation is necessary to ensure that AI infrastructure evolves in ways that serve the public good rather than deepening existing inequities.
A Narrow Window to Shape the Future
The era of widely accessible, low-cost AI has allowed many social impact organizations to experiment and innovate. But it has also created a quiet dependence on business models that are now changing.
We still have a narrow window in which to design shared, equitable, and sustainable AI infrastructure before access hardens into a set of private, unaccountable utilities. If AI is going to be as fundamental as electricity or broadband, then civil society cannot remain a passive customer. It must become a co-designer: building shared infrastructure, insisting on accountability, and ensuring that the next phase of AI serves not only markets, but the public good.
Read more stories by Phillip Olla.