Episode Details

AI’s real choke point may not be compute — it may be energy, memory bandwidth, and the politics of building at scale. In this condensed summary of Harry Stebbings’ 20VC conversation with Thomas Somas, co-founder and chairman of Positron AI, you’ll hear why inference is becoming the next battleground in generative AI, how the “memory wall” is reshaping semiconductor economics, and why cached tokens can be so profitable. The discussion also digs into data centers, power use, regulation, geopolitics, and the growing backlash against infrastructure. Somas makes the case that the U.S. should keep building, that energy is civilization’s true bottleneck, and that local models may end up serving as filters for cloud-scale intelligence. If you want the key ideas from the full-length episode in a much faster listen, this summary covers the biggest insights in minutes. Listen now to get the key ideas in minutes.