Bullish and bearish opinions expressed in this episode, paired with supporting transcript quotes. The quote confirms what was said—not whether the opinion is correct.
Bullish
$AMD— AMD chips are undervalued due to perception issues rather than actual performance problems. The speaker sees alpha in buying AMD capacity that others are sleeping on, as the chips work well at the right price and major buyers like Meta and OpenAI are already purchasing significant quantities.
$INFRA— Distributed inference infrastructure represents a major opportunity as the market shifts from training-focused large data centers to inference-optimized smaller facilities. The speaker sees significant demand for 1-megawatt data centers that others are overlooking, with the ability to accept lower uptime requirements enabling access to cheaper power and facilities.
$OPENAI— OpenAI is positioned as a major buyer of alternative chip architectures, demonstrating willingness to diversify beyond NVIDIA and validating the market for non-NVIDIA compute solutions.
Bearish
$NVDA— NVIDIA's performance improvements from generation to generation (Hopper to Blackwell to Rubin) are not as dramatic as perceived, with performance per watt not improving significantly. The speaker also notes that TSMC's lead over Western fabs is smaller than commonly believed (only ~2x performance per watt), reducing the perceived moat.
$DATACENTER— Large centralized data centers (100+ megawatt facilities) face structural challenges as the market shifts from training to inference workloads. The speaker notes it's 'basically impossible' to build gigawatt data centers in the US anymore, and even 100 megawatt facilities are increasingly difficult, creating a supply constraint for traditional large-scale approaches.
$AILABS— The premium that frontier AI labs charge for being 3-6 months ahead may not be sustainable, as enterprise customers don't adopt new models at that pace and are often still using models from 6+ months ago. The speaker questions whether the premium for being ahead will last long-term.