VanEck product manager Nick Frasse stated that physical construction delays for chip suppliers like ASML Holding NV (NASDAQ:ASML) and Taiwan Semiconductor Manufacturing Co. Ltd. (NYSE:TSM) control the expansion pace of data centers, while investor focus shifts toward memory producers Micron Technology Inc. (NASDAQ:MU) and SK Hynix Inc. (NASDAQ:SKHY).

Data Centers Function as Token Factories

Frasse, who is part of the management team of VanEck’s $67.79 billion VanEck Semiconductor ETF (NASDAQ:SMH), stated in a conversation with Phil Rosen that artificial intelligence (AI) deployment remains in its early infancy, remarking, “we’re probably still in the batting cages for it.”

He explained that data centers operate essentially as “token factories” that burn tokens at accelerating rates as agentic AI executes broader prompts. Frasse noted overall semiconductor demand has exceedingly outpaced supply across all technology sectors.

Integrated Semiconductor Supply Stack

Frasse outlined how semiconductor companies operate through collaborative supply chain layers rather than direct head-to-head competition.

Fabless design firms outsource chip manufacturing to foundries like TSMC, which produces hardware for Nvidia Corp. (NASDAQ:NVDA), Apple Inc. (NASDAQ:AAPL), and Broadcom Inc. (NASDAQ:AVGO).

TSMC relies on ASML, with Frasse emphasizing “there’s only one ASML in the entire world” manufacturing EUV lithography machinery. Consequently, demand increases at foundries automatically drive revenue for equipment suppliers.

Rotation Into Memory Bottlenecks

Following early packaging constraints on GPUs, Frasse identified memory as a current major bottleneck in the AI buildout.

Producing high-bandwidth memory requires building new fabrication plants, extending supply timelines. Frasse noted multi-year commitments from corporate balance sheets reduce historical industry cyclicality.

Frasse said that there are cycles even in bull markets and thinks there could be reversions to the mean. He added that there are investors who “don’t want to pay up for Nvidia right now” but may put capital into memory suppliers like Micron and SK Hynix.

Physical Constraints Act as AI Governor

Addressing investor concerns regarding massive capital expenditure and potential market bubbles, Frasse argued that physical infrastructure timelines naturally govern industry expansion.

Constructing a semiconductor fabrication plant or standing up a data center requires two to three years. Frasse explained that these physical time limitations create an “inherent governor” that will “help a giant bubble from being created,” as compute constraints prevent overbuilding.

How Have These Stocks Performed?

Stocks1-Month6-MonthsYTD1-Year5-Years
ASML-13.46%17.31%52.00%86.83%89.53%
TSM-0.16%24.36%43.62%63.73%265.40%
MU-3.39%111.72%242.49%510.98%1215.61%
SKHY6.77%7.64% (Since Listing)

Disclaimer: This content was partially produced with the help of AI tools and was reviewed and published by Benzinga editors.


Image via Shutterstock