The AI investment cycle is shifting from GPUs toward CPUs, with AMD and Intel (INTC) becoming essential for "Agentic AI" orchestration; Intel specifically offers a long-term floor near $20.47 due to heavy US government backing. High-conviction opportunities exist in the memory sector, where Micron (MU) and Western Digital (WDC/SanDisk) are seeing unprecedented demand for high-bandwidth memory and NAND flash that is already sold out through 2028. For diversified exposure to the memory bottleneck, the Roundhill Memory ETF (MEMY) provides a targeted basket of these high-growth hardware producers. Investors should look beyond chips toward power and infrastructure plays like GE Vernova (GEV), Constellation Energy (CEG), and Bloom Energy (BE) to solve the massive electrical shortages currently idling data centers. While NVIDIA remains the industry leader, the highest potential returns are moving "out the risk curve" to these secondary layers of the AI stack funded by the $1.1 trillion in projected capital spending from Google, Microsoft, and Meta.

Hyperscalers & Model Labs (The "Waterfall" Source)

• This layer represents the origin of AI investment capital. Major companies are seeing massive inflows from retail and enterprise users (via products like ChatGPT and Claude), which they are then reinvesting into infrastructure. • Key Players: Google (GOOGL), Microsoft (MSFT), Amazon (AMZN), and Meta (META). • Private Players: OpenAI and Anthropic are identified as the primary funnels for retail AI interest. • Capital Expenditure (CapEx): These companies originally committed $630 billion for 2026 but have revised that upward to $800 billion this year, with projections hitting $1.1 trillion next year.

Takeaways

• Sustainable Flywheel: Unlike the dot-com bubble, these companies are spending from cash flow and earnings, not debt. As long as AI remains useful, this "waterfall" of cash will continue to flow to the lower layers of the stack. • Platform Dominance: Google recently reclaimed the crown as the world's most valuable company, signaling that the market is rewarding the "distributors" of AI, not just the chip makers.

Central Processing Units - CPUs (AMD & Intel)

• A major "narrative violation" is occurring: while GPUs were the focus of the last year, CPUs are becoming equally critical due to "Agentic AI" (autonomous AI agents). • The Ratio Shift: Previously, the ratio of CPUs to GPUs in AI models was nearly zero. Today, it is roughly 1:1, and it is projected that the number of CPUs will soon outweigh GPUs to handle "agent orchestration." • Intel (INTC): Mentioned as a strategic play for the US government to reduce reliance on TSMC in Taiwan. The US government reportedly took a 10% stake in Intel at roughly $20.47 per share. • AMD (AMD): Reported to be up 320% over the last year, driven by the need for "brains" to coordinate AI agents.

Takeaways

• The "Agentic" Trade: Investors should look at AMD and Intel as essential infrastructure for the next phase of AI where agents must "think" and use tools autonomously, a task handled by the CPU. • Sovereign Investment: The US government’s involvement in Intel suggests a long-term floor for the stock based on national security and domestic manufacturing needs.

AI Memory (MU, SanDisk, & SK Hynix)

• Memory is described as the "secret monopoly" and the current primary bottleneck for AI hardware. It accounts for roughly 50% of the bill of materials for a high-end GPU. • High Bandwidth Memory (HBM): Dominated by Micron (MU), SK Hynix, and Samsung. This is the premium memory used for training frontier models. • NAND Flash: Specialized in by SanDisk (Western Digital). This is used for "persistent" memory and expanding context windows (the AI's ability to remember earlier parts of a conversation). • Supply Constraints: These providers are reportedly sold out of capacity until the end of 2028.

Takeaways

• SanDisk (Western Digital): Highlighted as a massive gainer (up 40x according to the transcript) due to the explosion in demand for NAND storage required by smarter AI agents. • Micron (MU): Up 7.25x over the last year; remains a core play in the HBM sector. • ETF Opportunity: The Roundhill Memory ETF (MEMY) was mentioned as a way to gain exposure to a basket of these memory stocks; it rose 72% in its first month. • Jevons Paradox: Even as AI models (like DeepSeek) become more efficient with memory, the total demand for memory increases because the lower cost allows for more agents to be deployed.

Power & Physical Infrastructure

• As thousands of GPUs are installed, the bottleneck shifts to power generation, cooling, and electrical regulation. • Key Players Mentioned: • Bloom Energy (BE): Mentioned as a significant gainer in the power space. • Corning (GLW): Recently partnered with NVIDIA to focus on optical glass/fiber for data centers. • GE Vernova (GEV), Constellation Energy (CEG): Essential for supplying and regulating the massive power loads required by data centers.

Takeaways

• The "Dark Warehouse" Risk: Companies like XAI are reportedly only utilizing 11% of their GPU power because they lack the electrical infrastructure to turn them all on. • Next Wave: The "puck is flowing" toward power and cooling. Investors should look beyond the chips to the companies that enable the chips to actually run.

Investment Themes & Risk Factors

• The "Bubble" Debate: The speakers argue this is not a traditional bubble because it is not "levered" (borrowed money). It is funded by the massive profits of the world's largest companies. • The Risk Curve: Money is moving "out the risk curve"—away from NVIDIA (NVDA) (where the growth is already "baked into the valuation") and toward secondary and tertiary layers like memory, CPUs, and power. • Raw Materials: Lithium Carbonate was mentioned as a foundational "Layer 6" material, with prices seeing extreme volatility and growth (from $75k to $187k in a short period).

Takeaways

• Cycle Awareness: Memory is historically a "boom and bust" industry. While executives claim "this time is different," investors should be wary of future supply gluts once current contracts are fulfilled. • Actionable Strategy: Monitor quarterly earnings of the "Hyperscalers" (Google, Microsoft, etc.). If their AI-driven revenue continues to grow, the "waterfall" to the semiconductor and power sectors remains intact.