Where is AI Going Next?
Спикеры
Ejaaz Ahamadeen
— Соведущий, подкаст Limitless (Bankless)
Josh Kale
— Соведущий, подкаст Limitless (Bankless)
The hosts analyze the flow of AI capital expenditure from hyperscalers to downstream layers of the AI stack. They highlight memory and CPUs as current winners and identify infrastructure/power as the next frontier. The episode provides a framework for understanding the trickle-down economics of AI investment.
- Hyperscalers (Google, Amazon, Meta, Microsoft) are spending $800B+ in 2026, funded by cash flow.
- AI capital flows from retail (OpenAI/Anthropic) to hyperscalers, then to semiconductors, CPUs, memory, infrastructure, and raw materials.
- Memory (HBM and NAND) is a supply-constrained bottleneck with contracts through 2028.
- Agentic AI is driving a shift from GPUs to CPUs, benefiting Intel and AMD.
- AI infrastructure and power generation (e.g., Bloom Energy, Corning) is the next potential beneficiary.
- The current AI spending cycle differs from past tech bubbles because it is not leveraged.
- Jevons paradox is observed: efficiency gains in AI models increase total memory demand.
- NVIDIA is seen as past its peak relative value, with capital rotating to other layers.
Идеи
CPU demand driven by AI agents.
Agentic AI requires CPUs for orchestrating multiple AI agents, reversing the historical GPU-only focus. The CPU-to-GPU ratio is now moving from near zero to one-to-one and will soon flip, with CPUs outnumbering GPUs. Intel and AMD, dominant CPU manufacturers, are direct beneficiaries of this structural demand shift from AI inference and agent orchestration.
Memory supply constrained until 2028.
Memory is the critical choke point in the AI stack, making up 50% of a GPU's bill of materials. High-bandwidth memory (HBM) is supplied by only three companies (Micron, SK Hynix, Samsung) and NAND flash by SanDisk. Supply is sold out until 2028, and demand continues to rise due to AI agents and Jevons paradox (efficiency gains increase total memory consumption). These companies have pricing power and long-term contracts.
AI infrastructure power next layer.
The next wave of AI spending will shift from chips to power generation and infrastructure because data centers cannot utilize existing GPU clusters fully (e.g., xAI only 11% utilization). Companies providing power supply, regulation, cooling, and optics (Bloom Energy, GE, Vinova, Constellation Energy, CEG, and Corning) will see rising demand as hyperscalers and AI labs build out facilities.
Далее
This Bankless video, published May 07, 2026, features Ejaaz Ahamadeen discussing AMD, INTC, 000660.KS, MU, SNDK, 005930.KS, Vinova, GLW, BE, CEG, GE. 3 trade ideas extracted by AI with direction and confidence scoring.
Speakers: Ejaaz Ahamadeen · Tickers: AMD, INTC, 000660.KS, MU, SNDK, 005930.KS, Vinova, GLW, BE, CEG, GE