Open Innovation in AI: Can Policy Promises Match Market Realities?

When the White House unveiled its AI Action Plan this summer, industry leaders applauded its call to prioritize open innovation. IBM CEO Arvind Krishna highlighted how openness and supportive regulation could cement U.S. technological leadership. On paper, the message is clear: openness is the path to progress.

But a recent NBER working paper by Pierre Azoulay, Joshua Krieger, and Abhishek Nagaraj offers a sobering counterpoint. Their analysis of generative AI competition warns that openness alone rarely guarantees competitive markets. Instead, control over complementary assets—from GPUs and training data to safety systems and benchmarks—will likely determine who profits from innovation.

This creates a paradox. Even as policymakers call for democratization, industry dynamics may funnel power into the hands of a few well-capitalized incumbents. The authors note that only the emergence of a “rogue” giant—such as Meta’s open-source LLaMA experiment—has the potential to reset the balance, much like Android disrupted the smartphone duopoly.

What leaders should take away:

• Policy is signaling intent, not enforcing outcomes. Don’t assume regulation will level the playing field in AI.

• Moats are shifting. Traditional IP is less important; instead, data pipelines, compute scale, and safety governance define defensibility.

• Openness can be a weapon. Strategically choosing openness—through partnerships, shared infrastructure, or transparent models—can be as much a power move as closing off ecosystems.

• Watch the “application layer.” Even if the foundation model market concentrates, opportunities for startups and enterprises will flourish in applied AI solutions.

In short, today’s leaders must navigate a dual reality: governments preach openness, while markets indicate control. The firms that thrive will be those that learn to balance both—leveraging openness to attract ecosystems, while securing enough control to capture value.