🎙️ AI Infrastructure, minus the buzzwords.

In this episode, Ji Hoon Kwon (Managing Director, Samsung Ventures) explains where AI data centers really choke (power, thermal, and networking) why baseload energy matters, and how a CVC can move at VC speed while creating value beyond capital. He also pressure-tests “future” ideas like floating and space data centers with an operator’s lens.


Highlights

  • 🔌 Power is bottleneck #1. GPU draw jumped to ~1–1.2 kW per chip; racks push ~120 kW, and lost power turns into heat.

  • 🌡️ Thermal follows power. More watts mean more heat; rethink electrical and cooling together.

  • ⚙️ Next wave: power electronics. Efficiency in conversion and distribution becomes the lever.

  • 🔋 Baseload beats intermittency. 24/7 AI favors SMR nuclear and geothermal over variable-only supply.

  • 🧪 Diligence that saves time. Run technical and financial checks in parallel; decide fast or pass fast.

  • 🤝 Value after the check. Open doors across BUs—manufacturing, customers, and design-ins—not just capital.

  • 🛟 Future formats, reality check. Floating and space DCs have promise, but corrosion, maintenance, connectivity, and launch costs are tough.

  • 📈 What makes a “good” CVC deal. Strategic fit matters, but financial returns still keep the engine running.

Resources:

Rss Apple Podcaster
Podden och tillhörande omslagsbild på den här sidan tillhör Davide Ritorto. Innehållet i podden är skapat av Davide Ritorto och inte av, eller tillsammans med, Poddtoppen.