AI Demand Concerns: Nvidia, Hyperscalers, and Chip Supply Dynamics

The AI Chip Market: Supply vs. Demand

• 💡 The core tension in the AI chip market is between Nvidia's desire for frequent chip upgrades and hyperscalers' need for longer-lasting, cost-effective hardware.
• 🧠 Both Nvidia and hyperscalers have valid points: Nvidia emphasizes efficiency gains in new chips, while hyperscalers focus on capital expenditure and depreciation.
• ⚠️ Current AI workloads are still being served by older generation chips (like V and A series Nvidia GPUs), indicating that the newest variations may not be immediately essential for all users.

Equilibrium and Depreciation Risks

• 🎯 The AI chip market currently lacks equilibrium, with demand outstripping supply for every available chip.
• ⏳ If market equilibrium is reached towards the end of the decade, the depreciation issue could become significant for companies like CoreWeave, which have a six-year depreciation model.
• 📈 Hyperscalers like Google, Microsoft, and Amazon can absorb this risk due to their own workloads and existing data center infrastructure, but companies building on spec are more exposed.

Market Indicators and Valuations

• 📊 Despite concerns, supply chain checks do not currently indicate a slowdown; demand for chips, networking, and CPUs remains high, as evidenced by AMD's CEO Lisa Su's confidence in doubling their TAM.
• 📉 However, if demand begins to slow even slightly, it could impact investor views on valuations, especially for companies whose models are not based on a simple on/off switch for demand.
• ⚠️ Companies like CoreWeave face risks from their six-year depreciation model and collateralized GPUs, particularly as value-added services in bare metal offerings are limited.

Energy and Capacity Constraints

• ⚡ Energy is identified as one of the biggest constraints in the AI complex, alongside memory.
• ⚠️ Companies offering only more compute risk being overexposed if hyperscalers, who prefer not to hold excess capacity, no longer need immediate solutions.