Star Analyst Recommends Trimming Cloud Giants, Pivoting to Semiconductors, Warns AI Rally Will Hit "Speed Bumps"
N.R. Finch
Dot-com-era semiconductor analyst Dan Niles warns the AI trade is hitting a 'speed bump' and urges a shift from cloud hyperscalers to semis — as enterprises pivot from 'token maximization' to 'token minimization,' September-quarter guidance could disappoint.
What is the shift from "token maximization" to "token minimization"?
Months ago companies encouraged employees to use AI as much as possible — generate tokens freely. (Tokens are the smallest billing unit for AI usage; more tokens = higher cloud bills.)
That has reversed. Enterprises realized they could burn through their entire annual AI budget in four months at that pace.
This means → companies are actively cutting AI consumption, and the "revenue acceleration" story for cloud providers may stall in the second half of the year.
Why shift from cloud giants to semiconductors?
Dan Niles's logic: regardless of how enterprises cut token usage, the money to build data centers and buy chips has already been spent — semiconductors are the most direct beneficiary of AI capex.
The market is already pricing this in: year-to-date, Amazon, Google, Microsoft, Oracle, and Meta are down ~6% on average, while semiconductor ETFs (SMH, SOXX) have risen.
In plain terms = cloud companies sell a "pay-per-use service" — less usage, less revenue. Chip companies sell "one-time hardware" — the purchase is already booked.
Strong in June, pressure in September — what's the timeline?
June-quarter earnings are expected to remain strong because revenue from the earlier "token maximization" wave is still flowing through.
But enterprises are routing more requests to open-source models — some costing just one-eighth of premium models.
This means → September-quarter guidance is the key checkpoint. A mass shift to cheaper models could visibly slow cloud revenue growth.
How is Dan Niles positioning his own portfolio?
He is gradually trimming hyperscaler exposure while also cutting some semiconductor holdings.
His reasoning is blunt: "The semiconductor index has already doubled" — valuation headroom is narrowing.
In plain terms = he is not saying semis will keep rallying. He is saying that as the AI thesis evolves, investors need to be far more selective rather than holding passively.
Who disagrees? What does Columbia Threadneedle say?
Tiffany Wade, senior portfolio manager at Columbia Threadneedle Investments, is more bullish: AI infrastructure companies remain in a strong growth cycle, with "revenue and earnings estimates still being revised upward."
She cited two recent data points: Micron's AI infrastructure deal with Anthropic and continued expansion of large data-center projects in Texas.
In her view, the recent tech pullback is short-term noise — it does not change the long-term trajectory of expanding AI investment.
AI trade: speed bump or just a pothole?
BULL
Compute demand stays strong
Micron + Anthropic deal, Texas data-center expansion — spending is still ramping.
Pullback is short-term noise
Revenue and earnings estimates still being revised up; long-term trend intact.
BEAR
Token minimization eats revenue
Enterprises are rationing AI usage — cloud's pay-per-use model takes a direct hit.
Open-source models divert spend
Some models cost one-eighth of premium; September guidance at risk.
Valuations are no longer cheap
Semi index has doubled — even if the thesis is right, upside is compressing.
In plain terms = the two sides don't disagree on whether AI matters. They disagree on whether enterprises will cut spending faster than infrastructure can expand — September-quarter earnings are the deciding test.
What should ordinary investors watch?
There is one verification point: September-quarter earnings — can hyperscalers' massive capex translate into visible revenue growth?
If September guidance is strong, AI demand is absorbing the "token minimization" headwind — the bulls win.
If September guidance is weak, Dan Niles's "speed bump" call gets priced in, and the divergence between cloud giants and semis could widen further.
Content is for reference only, not financial advice.