Korean startup DeepX wins green tech label for low power AI chips

Most of the public debate this week is about slowing models, but DeepX’s certification is a reminder that the hardware race is still about scale and efficiency. If you can cut inferencing power by a quarter while holding performance steady, you make it cheaper to deploy smarter models into edge devices, factories and robots. That matters even more as regulators and communities balk at the energy footprint of hyperscale data centers.

For AGI, breakthroughs in low power inference do not directly change the core science of intelligence, but they do change the economics of saturating the world with capable agents. A future where every robot arm, car and camera can host reasonably strong on device models is one where emergent system behavior and coordination problems become much more complex. Those dynamics will arrive faster if chips like DeepX’s can be manufactured and integrated at scale.

In the Korean context, green tech certification also opens doors to public procurement, subsidies and preferential financing. That could give domestic AI hardware startups a wedge into global supply chains that have been dominated by US, Taiwanese and European players, diversifying the ecosystem that underpins frontier model deployment.