What happens when the power to shape AI’s future lies not in the smartest model—but in who controls the compute?
This week, I spoke with Lennart Heim, researcher at RAND and advisor to Epoch AI, to unpack the geopolitics of compute: why sheer model capability is no longer the only game in town, and how control over compute may define who leads in AI.
We discussed:
* Why China matching US models doesn’t mean it’s winning the AI race.
* The staggering scale of infrastructure: 100,000+ GPUs, billions in hardware, and spiralling test-time compute demands.
* Why algorithmic efficiency is slashing costs and redrawing the frontier faster than expected.
* The economic and national security tension between AI deployment and control.
* The risks of over-concentration and the rise of sovereign compute strategies from Europe to the UAE.
* What open-source models on abundant compute might mean for future power dynamics.
From token floods to trillion-dollar CapEx, we explored the delicate architecture of power in the age of AI. Enjoy!
📅 Catch me live every Friday at 9am PT | 12pm ET | 5pm UK.
