Yorkville America Equities launched the Yorkville America Next Generation Memory Index ETF (NYSE:NRAM) last week, with Micron (NASDAQ:MU) as its largest holding.
CEO Steve Neamtz says the supply thesis behind the fund could take years to play out. New factories capable of materially easing the memory shortage may not reach full production until 2028 or 2029.
Neamtz, citing SemiAnalysis research, said memory can account for more than half the cost of an AI accelerator. He also said major suppliers have already committed their 2026 high-bandwidth memory, or HBM, output.
“The AI platforms get the attention, but the infrastructure underneath them is where the actual bottleneck sits,” Neamtz told Benzinga.
When Does the Memory Crunch End?
Producing HBM consumes roughly three times the manufacturing capacity per gigabyte of conventional memory, Neamtz said.
“There’s a real gap between when the shortage needs to ease and when the supply to fix it is available,” he said.
Micron CEO Sanjay Mehrotra made a similar point in June, saying supply should improve gradually in 2028 but the company had no "line of sight" on when it would catch demand.
The shortage has already started influencing AI hardware. TrendForce said Nvidia Corp. (NASDAQ:NVDA) cut planned memory capacity in its Vera Rubin systems after preliminary supplier allocations covered only around 60% of its estimated 2027 needs.
Has the Memory Trade Already Peaked?
A mainstream 32GB DDR5 kit climbed from roughly $110-$140 to around $400 in less than a year, according to research Neamtz cited. Memory stocks have surged alongside prices.
He argues the shortage is unlikely to disappear soon. Oversupply could emerge around 2028 or 2029, he said, but only if new capacity arrives just as AI demand slows.
The Bet Goes Beyond Micron
Micron currently leads NRAM’s 20-stock underlying index at 16.5%, followed by Sandisk (NASDAQ:SNDK) at 14.4% and SK Hynix at 14.1%.
“The current memory shortage isn’t confined to a single component or a small handful of producers,” Neamtz said.
That helps explain Sandisk’s large weight: Neamtz argues AI is straining NAND storage as well as the DRAM and HBM feeding data to processors.
NRAM also holds companies outside the major DRAM and HBM producers, including storage and controller names.
When Does the AI Trade End?
The bigger risk is that AI demand fades before new memory capacity arrives. So far, the numbers point the other way.
Anthropic recently told investors it expects positive adjusted operating income for a second straight quarter.
Benzinga reported that revenue across the top 10 AI apps jumped 68% from January to July, even as downloads fell.
TrendForce expects the world’s nine largest cloud providers to spend more than $886 billion this year, while DRAM and NAND could absorb 68% of major cloud providers’ capex in 2027.
Prediction traders appear bullish on the broader AI trade. Polymarket gives Nvidia a 71% chance of ending 2026 as the world’s largest company, on about $7.2 million in trading.
For Yorkville, the bet is that AI demand keeps growing faster than memory makers can add supply.
Image: Shutterstock
Login to comment