The AI chip squeeze, Dirty Soda mania, OCC approves Palmer Luckey’s Erebor | Diet TBPN
Speakers
John Coogan
— Co-Host, TBPN
Jordi Hays
— Co-Host, TBPN
Summary
The hosts focus on the AI supply chain, debating whether chips or energy are the bigger constraint on AI growth. They highlight TSMC and ASML as critical bottlenecks, while noting surging hyperscaler capex and Nvidia as a direct beneficiary. Other segments cover dirty soda's viral business model, Palmer Luckey's Erebor bank approval, AI.com, and debates about AI replacing software.
- AI chip bottleneck is currently more acute than energy.
- TSMC and ASML are key supply-chain constraints.
- Hyperscaler capex is rising sharply, including Amazon's $200B plan.
- Nvidia is expected to benefit from AI chip purchases.
- Energy may become a bigger bottleneck by 2027.
- Dirty soda trend raises questions about long-term moat.
- OCC approved Palmer Luckey's Erebor bank.
- Crypto.com founder bought AI.com for $70M.
Ideas
Energy is a future AI bottleneck.
Energy will be a bottleneck for AI and broader human flourishing, likely becoming a more acute problem by 2027. While chips are the current binding constraint, the speaker still sees energy as an important area to monitor because power generation capacity is hard to scale quickly.
AI chip bottleneck favors semiconductors.
AI compute demand is creating a semiconductor bottleneck. Chips are currently the more important constraint than energy because leading-edge fab capacity takes years and tens of billions to build while hyperscaler capex is surging. The bottleneck runs through advanced nodes and specialized equipment, supporting the semiconductor supply chain.
ASML EUV supply is constrained.
ASML is a bottleneck within the chip bottleneck. It ships only about 50-60 EUV machines per year at $350 million each, leading-edge fabs need dozens, and ASML's own supply chain is not very diversified. This makes ASML a critical, capacity-constrained supplier for advanced semiconductor expansion.
TSMC controls advanced node bottleneck.
TSMC controls about 90% of the advanced node market and has become the real bottleneck in AI chips. Its Arizona fab took five years and still is not at volume, showing how hard it is to replicate its process know-how and capacity. As hyperscalers push capex into hundreds of billions, TSMC's scarcity gives it pricing power and strategic leverage.
Amazon AI capex pressures profits.
Amazon's plan to spend $200 billion on AI buildout is worrying investors because it may pinch profits while waiting for returns. The stock has underperformed other hyperscalers, and the speaker characterizes Amazon's capex as playing defense rather than offense.
AI accelerates Meta's ads business.
Meta is a clear beneficiary of AI advancements, which are already showing up in an accelerating ads market. The speaker frames AI as directly strengthening Meta's core advertising business.
Microsoft's OpenAI stake remains solid.
Microsoft's massive position in OpenAI is in a solid place, and it holds the IP and a real hold on the AI question. Even though the stock has been beat up a little, the speaker views its AI positioning as strong.
Hyperscaler capex flows to Nvidia.
Hyperscaler AI capex is surging, including Amazon's plan to spend $200 billion this year. Much of that spending will go to training and inference chips, and the speaker expects a large share to flow to Nvidia, making it a direct beneficiary of the AI data center buildout.
