What's really holding back AI's next breakthrough? Two things keep getting overlooked: our over-reliance on massive data centers and the narrow focus on language-only models.

Data centers are becoming a bottleneck—not just for computing power, but for sustainability and accessibility. We're pouring resources into centralized infrastructure when the real innovation might demand distributed solutions.

Then there's the elephant in the room: betting everything on language models. What about multimodal systems? What about models designed for different tasks, different domains? When everyone chases the same approach, you get diminishing returns.

These invisible constraints could define the entire decade ahead. Until we rethink them, we're just optimizing within a box instead of breaking out of it.