Licensing remains the biggest hurdle for enterprise AI adoption, says Freepik CEO

Freepik CEO Joaquín Cuenca Abela believes the generative AI boom is sustainable despite growing concerns about business models and market saturation. In an exclusive interview, Abela compared potential AI market corrections to the dot-com bubble of 2000—temporary setbacks in an otherwise transformative technology that’s already generating billions in revenue from real users at an unprecedented pace.

What you should know: Freepik has positioned itself as a bridge between AI innovation and enterprise compliance, addressing one of the biggest barriers to AI adoption.

• The company offers end-to-end legal protection and indemnity through its Enterprise plan, which has been “warmly received by Fortune-500 design teams.”
• “Correct licensing is still the single biggest hurdle for enterprises rolling out AI,” Abela explained, emphasizing that clients want certainty their work complies with IP laws.
• Energy efficiency in AI generation has improved dramatically, with per-image energy consumption dropping by 3x in just three months.

The licensing landscape: Three models dominate AI training data approaches, each with distinct trade-offs that could reshape the industry.

• Opt-in model: Practically impossible due to the need for gigantic datasets like “the whole internet” where finding individual authors is unfeasible.
• Opt-out model: Adopted by Europe but faces resistance because it requires companies to disclose their datasets—currently “a major part of the secret sauce.”
• All-in model: Used by competitive models in the US and China, with recent California rulings suggesting training on legitimately purchased content may qualify as “fair use.”

Energy consumption reality: Current AI generation is far more efficient than critics suggest, with costs continuing to plummet.

• A single high-quality image consumes energy equivalent to “an old 60W lightbulb running for 16 seconds.”
• Video generation requires about 100x more energy than images, equivalent to “a lightbulb on for ~30 minutes.”
• Most common video use cases involve product shoots and marketing campaigns, potentially replacing traditional crews and exotic location shoots.

Regulatory predictions: Abela expects significant market fragmentation as different regions take divergent approaches to AI regulation.

• “Many major companies will rather shut down their AI in Europe than disclose their datasets and fully comply with the AI Act.”
• The US needs consolidation of multiple federal laws and higher court ratification of recent rulings.
• Asian markets, particularly China, appear to be training models “with a wide variety of datasets, just like all the competitive models in the US.”

What excites him most: Coding agents represent the technology’s most transformative potential because they enable “recursive self-improvement.”

• “If we truly reach a point where AI can improve itself without any of the current bottlenecks, we will get unprecedented levels of intelligence to solve our problems.”
• Future applications could include “personalized education, medicines developed for your particular body, and many of the wonders that we can’t just yet think of.”
• Despite being “equally thrilling and terrifying,” Abela predicts it will become “the new normal” within a few years.

Looking ahead: The next generation of AI tools will likely prioritize simplicity and multimodal capabilities.

• ChatGPT’s main weakness is complexity—users struggle to understand “which model they should use for which questions.”
• Future versions will likely automate model selection and expand beyond current text, files, audio, and images to include video streams.
• AGI applications will be ubiquitous: “The difficult thing about AGI is finding something where it cannot be applied.”