Dan Ives: The AI Trade's Bottleneck Moved Twice, From Chips to Memory to Power
Memory will not clear until 2028 or 2029. Power needs three to four times what the grid has today. Chips, Dan Ives says, were never the hard part.
Ives, Wall Street’s loudest AI bull at Wedbush, now builds Yorkville Ives, a merchant bank. He told Anthony Pompliano the GPU bottleneck story is stale. The new constraints: memory and energy, neither with a clean fix.
Memory, not chips, is the binding constraint. No supply-demand equilibrium until 2028-2029, naming SK Hynix and Micron as beneficiaries Dan @ 17:49.
Energy is the deeper bottleneck. The US needs three to four times more power at just 20-25% AI adoption Dan @ 26:39.
Chinese open-source is a “mini DeepSeek moment,” not existential. Anthropic and OpenAI’s moat is enterprise data, not model quality Dan @ 1:04.
The real risk to the US lead is domestic. Ives calls New York’s proposed data center moratorium “dangerous” Dan @ 21:33.
This breakdown is for paid subscribers. Below: the full memory-bottleneck timeline behind Ives’s 2028-2029 call, the multiple dislocation across Micron, SK Hynix, and Intel three other analysts are fighting over this same week, and the exact case for why New York’s data center politics beats China as the real risk to the AI trade. Join to get full access.
