The hidden cost of AI: What’s the price of technological progress?

As artificial intelligence becomes increasingly embedded in our daily lives, concerns about its hidden costs are rising. While AI has the potential to drive economic growth and innovation, it also raises questions about fairness and governance. At the 4YFN event during MWC Barcelona, Irene Unceta, Academic Director of the Bachelor in Artificial Intelligence for Business at Esade, moderated a discussion on the social and economic implications of AI with leading experts in the field. 

The expert panel featured Marta Poblet, Senior Research Lead at The Data Tank and Adjunct Professor at RMIT University; Anna Colom, Senior Policy Lead at The Data Tank; and Julia Wallner, Chief of Staff at appliedAI Initiative GmbH. Together, they explored the consequences of AI-driven decision-making for inequality, labor and democracy, and the ethical responsibilities of companies and governments in ensuring fair technological development and implementation.

Is AI expensive? The energy and social cost that doesn't show up on any balance sheet

Before talking about social inequality, it's worth putting numbers on the cost of AI that is actually measured. According to Gartner data (2026), data centers consumed 565 TWh of electricity in 2026, 26% more than the previous year, and AI-optimized servers already account for 31% of that consumption.

The UN warns that by 2030 this demand could reach 945 TWh, comparable to Japan's current electricity consumption, with AI taking up as much as 40% of the total. The water cost is no small matter either: data centers used 4.5 trillion liters of water for cooling in 2025, a figure that could double by 2030 according to United Nations projections.

Inequality and exclusion in AI development

As noted by Unceta, AI is often framed as a force for progress—but is it equally positive for everyone? Who truly benefits, and who bears the cost? So far, AI development is concentrated in the hands of a few powerful companies and countries. And as AI systems mirror and amplify existing societal inequalities, those who are already marginalized risk being left even further behind. 

“It’s not that emerging technologies inherently come at the cost of progress,” Anna Colom explained. “But if they are not correctly developed, implemented, and governed, they do.” She emphasized that issues become particularly critical when AI is applied to public services such as healthcare, financial access, and social welfare. When algorithms determine who qualifies for a loan, insurance, or government support, biases in these systems can result in unjust outcomes. 

These are not just technical issues; they have real-life consequences for millions of people

Colom referenced a real-world case where the Dutch tax authority used an algorithm to spot suspected welfare fraud. Parents with foreign backgrounds were wrongly targeted by the algorithm and accused of making fraudulent childcare benefit claims. In some cases, the demanded repayments sum up to tens of thousands of dollars. Thousands of families were pushed into poverty, hundreds of children were taken into foster care, and some victims even committed suicide. 

It is a stark example of how an AI-based welfare system can cause serious harm when its implementation is flawed. The human cost of a badly designed algorithm can far exceed any efficiency it promised.

AI governance: who is heard?

The panel also addressed global governance of AI. The race for AI dominance is often framed as a struggle between major powers, but what happens to those left out of the conversation?

“The voices of the Global South have been largely overlooked,” Marta Poblet noted. “While we often talk about global convergence on AI principles, the reality is that fairness, transparency, and accountability are often defined by Western institutions. Yet, there are valuable initiatives emerging from the Global South that we rarely hear about.”  

If you lack technical expertise, you’re excluded from the conversation. That shouldn’t be the case

Concepts like Kaitiakitanga—a Māori term for stewardship that embodies guardianship, responsibility, and interconnectedness—or Ubuntu, a Nguni Bantu term from Southern Africa often translated as “humanity to others” or “I am what I am because of who we all are,” could be of the utmost value in developing AI governance that accounts for the deep and diverse ramifications of such technology. 

Poblet highlighted the need for inclusive policymaking, where affected communities have a say in the regulatory process. “If you lack technical expertise, you’re excluded from the conversation. That shouldn’t be the case.” Poblet advocated for integrating collective intelligence into AI governance, ensuring that public input extends beyond legislation into implementation and monitoring. 

The corporate responsibility in AI ethics

What role do companies play in addressing these gaps? “Companies need to define what fairness means in their AI ambition,” Unceta said. Understanding how AI works is crucial, and businesses should prioritize designing equitable systems from the start. 

According to Julia Wallner, fairness in AI can't be evaluated only at deployment; it must be built in from the start of development. "It's not just about regulatory compliance, but about transparency and accountability" she said. Getting this right early also keeps AI development cost under control, since fixing a biased system later is far more expensive.

Beyond compliance, Wallner emphasized the need for companies to proactively address labor displacement. “You are transforming not only your company but also people’s lives,” she said. It’s not just about staying competitive but also about investing in upskilling and reskilling their workforce. The World Economic Forum estimates that while AI will generate 11 million new jobs, it will replace 9 million. “But these aren’t one-to-one skill transitions—there’s a significant gap to bridge,” she noted. 

Colom agreed, adding that a major challenge is not just about upskilling but also about critically assessing which skills are prioritized and which are lost in the process. She warned that over-reliance on AI-driven decision-making in labor markets, such as algorithmic management in the gig economy, can exacerbate worker vulnerabilities. 

How can you reduce the operating cost of AI without sacrificing responsibility?

The question many executives are asking is not just whether AI is expensive, but how to manage that investment responsibly. Implementing AI without a clear ethical and governance framework not only creates reputational risk, it also drives up long-term costs. The Dutch algorithm scandal mentioned by Colom cost the government billions of euros in compensation and years of litigation. These are costs that no balance sheet captures at the moment the implementation decision is made.

To reduce the operating cost of AI sustainably, the experts on the panel agreed on three principles:

• Responsible design from the outset: build in bias audits, social impact testing and fairness criteria before deployment, not after. An error fixed at the design stage costs between 10 and 100 times less than one detected in production.
• Transparency as a competitive advantage: companies that document how and why their systems make decisions not only comply with the EU AI Act (in force since 2024), but also build institutional trust that reduces regulatory risk.
• Proactive reskilling: investing in upskilling before job displacement happens is 3 to 5 times cheaper than managing turnover and external hiring after automation, according to WEF 2025 data.

AI and democracy: a growing risk?

Beyond inequality and labor concerns, AI also has profound implications for democracy. Colom warned that the concentration of AI power within a handful of tech giants risks shifting from technocapitalism to techno-authoritarianism. According to her, figures like Elon Musk shape AI narratives in ways that could be harmful to democracy.  

Other risks include atomization, increased individualism, and a loss of shared social understanding. “For democracy to work, it is important that we understand each other, recognize our interconnectedness, and build bridges”, she stated. AI itself is not the problem, but it needs to be built on democratic principles and consider human connections. 

If we regulate AI as one big concept, we risk creating legal monsters. It’s like regulating football by only focusing on the ball

Poblet also stressed that AI governance must take a holistic approach. “The first thing when talking about AI governance is that we need to understand the object of the governance,” she explained. “If we regulate AI as one big concept, we risk creating legal monsters. It’s like regulating football by only focusing on the ball. The ball is essential, but what about the rest of the elements?”  

She emphasized that AI is just one part of a larger digital ecosystem with deep implications in democracy, which also includes data governance and digital services. A sector-specific and layered approach is necessary to address the complexity of AI governance, and bringing the right stakeholders into the conversation is crucial. “It takes time, but I remain optimistic about the regulatory space,” she concluded. 

Should we pause AI development?

One of the most debated questions in AI ethics is whether some AI developments should be slowed down or even halted. Poblet argued that while an outright ban on AI is unrealistic, certain autonomous systems should be paused until proper regulatory frameworks are in place. 

“Some systems shouldn’t be fully banned but postponed for a while,” she said. That’s the case for fully autonomous systems since “we don't fully understand their implications and we need to establish a proper regulatory framework”. “Some things need to slow down while we reflect on the world we want to live in”, Wallner added. 

Colom agreed: “Temporary bans can be a legitimate legal tool when necessary. The food and aviation industries don’t allow untested products to enter the market—why should AI be any different?”.  

The value of AI: How to build technological progress without passing on the cost

Wallner concluded that AI regulation and corporate responsibility must go hand in hand. “Companies are increasingly aware of these issues, but in the end, ethical AI is about what you do when no one is watching. It shouldn’t just be a marketing strategy—it should be embedded in the company’s values.” 

As AI continues to evolve, the key challenge is ensuring that technological progress does not come at the cost of fairness, inclusion and democracy. With thoughtful regulation, ethical corporate practices, and an inclusive approach to governance, AI can truly serve the interests of all—not just a privileged few.

The real value of AI isn't measured only in productivity gains or lower operating costs; it's also measured in how many people are left out of its benefits and how many natural resources are consumed to generate them. That is the AI cost that still doesn't appear on any corporate balance sheet, and the one that regulation over the next decade will have to learn to calculate.

Frequently asked questions

Is AI expensive compared with human labor?

AI is expensive in upfront investment, but its cost per task falls as volume grows. In 2026, tech startups are spending 50% more on AI than on payroll, according to industry data. However, the real AI cost must include development, infrastructure, energy, AI maintenance cost, staff training and regulatory risk management, factors many companies underestimate when calculating ROI.

How can you reduce AI operating costs without losing effectiveness?

To reduce AI operating costs, experts recommend:

• Using small language models (SLMs) when the task doesn't require a full LLM, since they can be up to 10 times cheaper per query.
• Optimizing latency and batch processing to reduce token consumption.
• Negotiating enterprise contracts with API providers before reaching critical volumes, which can cut costs by 30% to 40%.
• Building bias audits into the design stage to avoid the legal and reputational costs of flawed systems.

What is the difference between AI development cost and AI maintenance cost?

AI development cost covers everything needed to get a system into production: data collection and labeling, model training or fine-tuning, engineering talent and testing. AI maintenance cost is recurring: inference and hosting, monitoring for model drift, retraining, security updates and compliance audits. Many companies budget carefully for the first and underestimate the second, even though it runs for the whole life of the system.

What is the real value of AI for organizations?

The value of AI for organizations isn't just lower operating costs. According to the Penn Wharton Budget Model (2025), AI will raise productivity and GDP by 1.5% by 2035. Companies that adopt AI-driven automation report faster processing times and lower compliance costs, but sustainable value is only achieved when implementation is ethical, transparent and people-centered.

What hidden AI costs do companies fail to account for?

The most common hidden costs are energy consumption (AI servers already account for 31% of data center electricity use, a share set to rise toward 40% by 2030), the water cost of cooling, regulatory risk under the EU AI Act, the cost of reskilling displaced employees, and the legal and reputational liabilities that stem from algorithmic bias.