Everything About AI
NextFin News - Citi's Stuart Kaiser is not ready to buy the semiconductor selloff. The bank's head of U.S. equity trading strategy said Friday that even as the S&P 500 clings to record territory, the rally underneath has turned painful, and with earnings expectations for 2027 and 2028 largely priced in, valuations are demanding enough that the risks of stepping into chip stocks now outweigh the potential upside.
The remarks land in the middle of a rout that has wiped more than $1.6 trillion from U.S. semiconductor market value over six sessions, on pace for a $2 trillion wipeout since last week. On Friday, the VanEck Semiconductor ETF (SMH) fell 0.8% to 554.19 and the iShares Semiconductor ETF (SOXX) dropped 1.1% to 509.01, while the SPDR S&P 500 ETF (SPY) was essentially flat at 772.09 — roughly 1.2% below the index's August intraday record of 7,816.70. Advanced Micro Devices slid 2.2%, Micron Technology fell 2.8%, and Broadcom gave up 1.5%, even as Nvidia eked out a 0.5% gain.
Kaiser's message is a timing call, not an obituary for the AI trade. His hesitation captures the central tension of this market: the long-term story for chips has never been stronger, but the price investors are being asked to pay for it has rarely been more unforgiving.
What Kaiser Is Actually Saying
The setup Kaiser described is a two-speed market. The cap-weighted S&P 500 sits near an all-time high, yet only about half of its 500 members trade above their 200-day moving averages, down from roughly 70% in midsummer. Breadth — the share of stocks participating in the advance — has deteriorated even as the headline index grinds upward. That divergence is what he meant by a painful week beneath the surface: the index is being carried by a narrow group of mega-cap leaders while the average stock weakens.
For Kaiser, the problem is not the earnings narrative. It is the discounting. When clients already view two years of forward earnings — 2027 and 2028 — as embedded in current prices, the bar for new capital shifts. The question stops being "is the AI buildout real?" and becomes "what surprise to the upside is left, and is it worth the risk?" In his own words, valuation-only arguments are a tough sell:
If all you're coming to me with is valuation, that's really a tough sell to me.
Kaiser has made the point repeatedly in recent television appearances: the AI trade still has upside, but the tactical entry point matters. With forward earnings discounted and market breadth thinning, the risk/reward for adding chip exposure at these levels tilts against the buyer.
The Mechanism: Why Priced-In Earnings Change the Math
To understand why Kaiser is standing aside, it helps to separate the direction of the trade from the price of the trade. The direction — AI-driven demand for compute, memory, and networking silicon — remains intact. Hyperscalers are still spending at a record pace. Alphabet has guided 2026 capital expenditure to $195 billion to $205 billion, more than double the prior year, and Amazon has lifted its capex outlook to $220 billion. Citi's own equity research team recently raised its growth forecast for cloud-provider capital spending among the "Big Five" to 90% year over year in 2026 and 46% in 2027, and the firm's semiconductor analysts have called the backdrop supportive for compute chips.
That is the structural leg of the argument, and it is not in dispute. The cyclical leg is the equity valuation attached to it. Semiconductor equities ran far ahead of even those elevated expectations: SMH has climbed roughly 76% from its 52-week low of 315.05, and Nvidia has gained about 33% from its year-low of 164.27. When prices incorporate multiple years of earnings growth, any stumble in the delivery — a guidance miss, a supply hiccup, a financing scare — produces an outsized drawdown because there is no valuation cushion left to absorb it.
This is the transmission channel Kaiser is flagging. In a normally valued market, a good long-term story plus a temporary setback equals a buying opportunity. In a fully discounted market, the same setback triggers a de-rating because investors have already been paid for perfection. The October pullback is not a repudiation of AI demand; it is a repricing of the certainty with which that demand was underwritten.
The valuation dispersion inside the sector shows why a blanket "buy the dip" is the wrong reflex. Memory is trading at a forward price-to-earnings ratio of about 6 times, and Nvidia at roughly 20 times — both below the S&P 500's own forward multiple of about 20.3 times. But AMD and Broadcom trade at trailing multiples above 45 times. The chip sector is not uniformly expensive; it is unevenly expensive, and Kaiser's caution is aimed at the parts of the complex where the narrative has already been fully monetized.
The Second-Order Risk: How the Buildout Gets Financed
There is a deeper concern beneath the valuation call, and it is one Kaiser has raised in recent appearances: the financing of the AI boom. When equity valuations are elevated and interest rates remain restrictive, companies have a strong incentive to fund growth by issuing stock rather than borrowing. Equity becomes the cheaper currency. That dynamic has already begun to show up in the supply of shares hitting the market — through IPOs, follow-on offerings, and secondary sales — and it creates a mechanical headwind for existing holders.
The chain runs like this: high valuations invite equity issuance; new share supply absorbs the very liquidity that was supporting the rally; and the marginal buyer, already cautious about breadth and valuation, demands an even larger discount to absorb the paper. Kaiser has described the semiconductor selloff as a function of exhaustion and rising equity supply, questioning whether today's prices make stock issuance more attractive to companies than debt. That is a second-order effect most investors are not watching — they are focused on GPU orders and data-center announcements — but it is the kind of flow dynamic that can keep a stock pinned even when the fundamental story is intact.
Higher-for-longer rates amplify the feedback loop. As long as borrowing costs stay elevated, the equity-issuance channel stays open, and the supply overhang persists. The moment valuations compress enough to make debt the cheaper option, that pressure eases. Until then, every rally into strength risks being met with new paper.
The Counter-Case: Why Standing Aside Could Cost You
The strongest argument against Kaiser's caution comes from the other side of the same bank. Citi's equity research analysts have argued that U.S. semiconductor stocks are due for a rebound and have positioned the firm as a buyer on the pullback, preferring semiconductor-equipment makers over chip designers on the back of accelerating estimate revisions. The logic is straightforward: capital spending is still accelerating, equipment vendors are the first to get paid in every new fab cycle, and the earnings revisions for 2026 and 2027 remain positive for the group.
More broadly, the counter-thesis holds that waiting for a better entry point in a secular uptrend is a losing strategy. The semiconductor cycle has surprised skeptics repeatedly over the past three years. Every time the market feared the AI capex cycle was peaking, hyperscaler spending guides moved higher. Broadcom's chief executive reiterated that AI semiconductor revenue would exceed $100 billion in fiscal 2027 even after the company's June guidance miss. If that pattern holds again, the pullback is a cyclical de-rating inside a structural bull market — and the cost of waiting is missing the next leg up.
Kaiser's answer to this is implicit in his framing: he is not calling for the trade to end. He is calling the current risk/reward unattractive. The distinction matters. A tactical pause is not a structural exit, and an investor with a multi-year horizon faces a different calculus than a trading desk managing quarterly risk. But for the marginal dollar — the money deciding whether to deploy today — Kaiser's view is that the asymmetry currently favors patience.
Cyclical or Structural: The Call That Decides the Trade
Here is the judgment that separates this pullback from a top: the AI infrastructure buildout is structural, but the equity multiple attached to it is cyclical and mean-reverting. The structural evidence is the capital commitment itself — Alphabet's $195 billion to $205 billion guide, Amazon's $220 billion, the multi-year power and data-center construction pipeline. These are not one-quarter decisions; they are multi-year programs that cannot be unwound quickly. Demand for advanced logic, high-bandwidth memory, and networking silicon is locked in for years.
The cyclical evidence is the price action. SMH is down about 18% from its June peak of 671.83, and Broadcom is down roughly 32% from its 52-week high of 495.00. That is a classic valuation de-rating, not a demand collapse. The earnings are still coming in — TSMC reported record third-quarter net income of NT$763.06 billion on Thursday, up 68.7% year over year, with fourth-quarter revenue guided to $52.3 billion to $54.3 billion — but the multiple investors will pay for those earnings has compressed.
So the base case is a cyclical pullback within a structural uptrend: painful in the near term, resolved higher once valuations digest and the next wave of earnings confirms the demand trajectory. The risk case is that the financing channel Kaiser flagged becomes self-reinforcing — that equity supply keeps arriving faster than demand can absorb it, turning a de-rating into a longer consolidation.
What Would Prove Kaiser Wrong
This view is falsifiable, and the signals are concrete. Kaiser's "wait" call breaks if two conditions print together: first, the Philadelphia-style semiconductor indexes reclaim and hold their 50-day moving averages — roughly 579.66 for SMH and 538.11 for SOXX — on sustained volume; second, market breadth re-expands, with the percentage of S&P 500 stocks above their 200-day moving averages climbing back above 60% from the current roughly 50%. If both happen, the exhaustion diagnosis was premature and the tape is telling you the buyers have returned.
On the fundamental side, the disconfirming signal is an upward revision in hyperscaler capital spending that is met with a rally rather than a sell-the-news reaction. If Alphabet, Amazon, or Microsoft raise guidance again and chip stocks respond positively instead of fading, the market is signaling that earnings growth can still outpace the discount rate — and the "priced in" argument collapses.
Outlook: Three Horizons, Three Trades
Short term (weeks): volatility stays elevated. The combination of hotter-than-expected inflation — the annual consumer-price rate accelerated to 3.7% in September, above the 3.6% consensus — narrow breadth, and equity supply means rallies are likely to be sold until the tape proves otherwise. The exposed are the high-multiple names where the narrative was most fully monetized: AMD, Broadcom, and the equipment complex that ran hardest. The relatively insulated are the low-multiple memory names, where forward earnings are already discounted at single-digit multiples.
Medium term (quarters): the direction depends on earnings delivery. If the next round of data-center revenue guides confirms the capex trajectory, the de-rating reverses and the sector grinds back toward its June highs. If guidance merely holds steady, expect a range-bound market as valuation and earnings grow into each other.
Long term (years): the structural thesis survives. The power, cooling, and silicon requirements of the AI buildout do not disappear because equity multiples compress. Investors who can tolerate a 12-to-18-month consolidation window are more likely to be rewarded than punished — but only if they respect the valuation discipline that Kaiser is enforcing today.
The scenarios, then, are clear. The base case is a 10% to 20% range-bound consolidation as earnings catch up to prices. The upside case — a reacceleration in hyperscaler spending met with improving breadth — would reopen the rally toward new highs. The downside case — a financing scare that forces equity issuance at distressed prices — would extend the de-rating into a deeper, longer correction.
Kaiser's restraint is a reminder that in a market where two years of earnings are already discounted, the most disciplined position is sometimes the one you do not take. The AI trade is not broken. But the price of admission has finally caught up with the story — and until it works its way back down, the best trade in chips may be the one you wait for.
Explore more exclusive insights at nextfin.ai.
