Energy Is the Real Bottleneck for AI
Energy is your foundational bottleneck. Information is an energetic artifact.
Asked to name the real geopolitical bottlenecks in the AI stack, whether chips, compute, or energy, Warwick Powell answers without hesitation: energy is the foundational bottleneck. His reasoning is memorable. AI, he says, has brought to the front a genuine understanding that information is an energetic artifact, measurable in energy consumed per tokens produced. That granular link between electricity and intelligence reframes the whole conversation.
From there he draws out the stakes. Solving energy at the national level is not enough if, in the process, a country creates an access problem for its own society. He warns of energy haves and energy have-nots, and points to the United States, where the expansion of hyperscale data centres is leaving a legacy of electricity disadvantage, in cost and sometimes in outright availability. It is a concrete, everyday consequence of the AI build-out that most coverage skips.
The clip works because it upends the common assumption that chips are the whole game. Powell does place chips second, acknowledging they must be manufactured through complex processes and secured. But by putting energy at the substrate, he gives viewers a clean framework that connects to the episode’s threads on sovereignty and lock-in, and to Armenia’s own plans to use nuclear power for data centres. For anyone following technology, the economy, and the U.S.-China contest, it is a compact insight with a warning attached: get energy right, or risk building a two-tier society around it.
Transcript
Warwick Powell: Well, I actually think energy is your foundational bottleneck. And because AI, interestingly, has brought to the front a genuine understanding that information is actually an energetic artifact. And in very measurable ways, measurable in terms of energy consumed for tokens produced. right so we now have an environment where we actually have a very granular understanding of of how energetic information actually is so my suggestion is is that our first order of business if you will particularly from the point of view of risk and lock-in and all of that is addressing this question of energy addressing this question of energy opens up a Pandora's box we touched on of energy sovereignty, but we also need to be thinking about questions of energy haves and energy have-nots. This is a critical issue for societies at a domestic level. So you might solve a national question around energy, but if you don't at the same time ensure that in solving your first problem you don't create an access problem for your society at large, I think you will end up creating more problems than you have solved. And we see that in the United States where the expansion of AI data centres,