As the competition for AI dominance heats up, companies are beginning to think of how they can leverage the proprietary data they sit on to get their own piece of the AI pie. While the AI boom has been driven by faster chips, bigger models and ever-expanding compute power, foundational models like GPT, Gemini and Claude are becoming more accessible and commoditized. And that’s resulted in AI companies — like OpenAI, xAI, DeepSeek, Anthropic, Google, Ali Baba and others — launching more and more AI models that do arguably the same things, with some models edging out others on certain benchmarks.

Experts are now saying that the real competitive advantage is shifting to proprietary data, with some even arguing that the companies that control exclusive, high quality datasets are the ones that will set the terms on how AI is developed and used across industries in the near future.

Boe Hartman, cofounder and CTO of Nomi Health, sees this shift playing out in various sectors. “Data providers have the edge. Model providers are in a race to the bottom — eventually, they’ll be commoditized. The real value is in the data that makes those models smarter and more useful, especially as agentic AI comes into play,” he said in an interview.

It’s a lofty claim with real potential to actually change the future of AI development. But this emerging trend also comes with new challenges: more legal battles over data ownership, stiffer privacy regulations and rising costs of data acquisition.

Publicly available and synthetic data were once seen as the foundation of AI development. But as Andy Thurai, principal analyst and VP at Constellation Research, explained, their usefulness is hitting a ceiling. “Most AI companies have exhausted the openly available internet data. The differences in data, compute, infrastructure and algorithms are all narrowing, which means models will struggle to be differentiated. Exclusive, high-quality datasets are now the real differentiator.”

Proprietary data allows companies to fine-tune AI models with domain-specific knowledge, creating applications that outperform generic models trained on public data. This is especially true in highly specialized industries. Healthcare providers, for instance, are leveraging private patient records (with the right privacy protections) to train AI models that can diagnose diseases faster and more accurately than general-purpose AI.

However, acquiring and maintaining exclusive datasets isn’t easy. As Tzvi Kopetz, VP of marketing at Lusha, told me, “The biggest problem is signal to noise. Acquiring valuable data, not just noisy data, requires processing enormous quantities as well as sifting and filtering. Moreover, it also requires proprietary AI to do that.”

Money is king in the business world, isn’t it? And the fast-paced world of AI isn’t precluded from that reality. In the spirit of mining and monetizing the proprietary data gold, some companies — like social media platforms, for example — are quietly using user-generated content to train AI models that are then licensed to third-party businesses. Others, such as AI-powered financial services, are using proprietary transaction data to build predictive models that help optimize investments.

“There’s so much happening in this space, it’s hard to tell which business models will stick,” Hartman said. “I saw a report that there are 28,000 different ‘AI’ companies in the world — that’s a lot of spaghetti thrown at the wall. Time will tell who wins, but the companies with proprietary datasets will be the winners.”

Thurai sees this shift as inevitable. “By adding proprietary or private exclusive data, models go up in value tremendously. Especially models trained with domain-specific data, like in healthcare or finance, add a ton of value.”

While proprietary data may be the new AI gold, there are thorns and thistles on the way there. For example, adhering to regulations — which are now poised to become even stricter — and meeting ethical AI requirements are no walk in the park. The sheer cost of acquiring and maintaining clean, structured datasets is another major barrier.

Inna Tokarev Sela, founder and CEO of Illumex, noted how regulation is shaping the playing field: “The real struggle will be over who actually owns the data—the company collecting it or the person creating it. And who decides what’s an appropriate use of that data when it’s being generated constantly all over the world?”

Privacy laws like GDPR, CCPA and HIPAA are already reshaping how data can be used, and even more regulations are on the way. At the same time, there’s a growing realization that data sharing — when done responsibly — can be mutually beneficial. In Europe, for example, industry-specific data sharing agreements, such as healthcare data exchanges, are creating new models for balancing privacy with innovation.

With the balance of power shifting to exclusive datasets and smaller, leaner and more industry-specific AI models, companies that own high-value data are in a position to dictate how it’s used and who gets access to it. This trend suggests that the future of AI may not be dominated by model developers, but rather by the companies that provide high quality data for those models.

“Model providers will continue to provide value to a large host of service providers,” Kopetz said, “but I see a development of tiers of model providers. Just as DeepSeek constructed a low-cost solution on top of other models, similarly, there will be tiers of model service providers.”

That’s a notion Sela agrees with, noting that “The power dynamic is evolving more toward data holders setting terms for data consumers rather than foundational model developers holding the power. If I own an exclusive dataset, and you want to use a new DeepSeek model with my proprietary data, it’s you as the end consumer who will pay for this data exclusivity, not the model provider.”

The world of AI is fast-paced and new innovations emerge almost daily. So, it’ll be interesting to see how companies leverage their exclusive datasets, how this might change the way AI is developed and at what cost.

But one thing is certain: As AI models increasingly become more accessible, it’s not the algorithms that will determine success, but the data that fuels them. And companies that can collect, manage and monetize proprietary datasets will define the next era of AI — while those that rely solely on public data may find themselves struggling to compete.