Here is a fact that should not coexist with a sell-off, but did.
Cisco grew fiscal Q4 revenue 18% to $17.3 billion, beat the $16.8 billion consensus, and then guided the current quarter to $18–18.2 billion against a $16.8 billion Street estimate. That is a beat stacked on top of a raise.
The stock fell 8.4% to close at $113.47.
Now widen the frame to a single 48-hour window. Cerebras raised its full-year core revenue outlook to $880–890 million, posted $210 million for the quarter, and had its CEO tell the tape AI demand was "through the roof."
Cerebras fell about 14%.
Intel put up 25% revenue growth — its fastest since 2011 — with data-center revenue up 59% to $6.3 billion. Intel sank toward its $95.68 support.
Three companies, three beats, two raised guides, one 59% data-center print. All three lower, hard, inside two days. The reflexive read is that AI demand cracked. That read is wrong.
Three companies, three beats, two raised guides, one 59% data-center print. All three lower, hard, inside two days. The reflexive read is that AI demand cracked. That read is wrong.
What cracked was the second derivative.
Start with the setup. Cisco walked into the print up more than 60% for the year, finally rerated as an AI-infrastructure winner after a decade of being priced like a bond proxy. When a stock has already run 60%, the beat isn't the news — the beat is the reason it ran 60%. The beat is priced in before the CFO opens the call.
So the only variable left to trade is the slope of the curve going forward. And that is exactly where management handed the sellers a weapon.
Cisco is guiding fiscal 2027 revenue growth of roughly 15%. Solid. But analysts already see fiscal 2027 growth dipping back into single digits — a deceleration from the 18% Q4 print that the market cannot un-see.
Piper Sandler said it plainly: the numbers were good, but the guidance "looks conservative given the current demand environment," and some investors "may start to nitpick that we're seeing peak growth." Piper rates it a hold.
"Peak growth" is the phrase that matters. Not peak demand — peak growth. The level is fine. The rate of change is what rolls over.
Chuck Robbins tried to reframe the caution as discipline. "We had a record year, we had a record quarter," he told Cramer, adding that when he issued a guide above consensus, people asked why he was being so conservative. His answer: "We're starting a new fiscal year… it's also a time that we're going to start the year being a little bit prudent."
Prudent is the right word for a management team. It is a terrible word for a stock priced for acceleration. Prudence, in a 60%-YTD name, reads as a ceiling.
The Cerebras case is the same trade wearing a startup's clothes. Feldman raised the full-year guide, said demand was through the roof, and pointed at $25.4 billion in remaining performance obligations as proof of "extraordinary future demand." He also told the market to expect revenue to triple next fiscal year.
The tape refused to buy the triple. Cerebras still carries a $450.5 million net loss for the quarter — most of it, $386.6 million, stock compensation — after a $309.5 million profit a year earlier. And when a company promises to triple, "through the roof" isn't a bonus; it's the baseline already embedded in the price. A raise from $855–865 million to $880–890 million is a nudge, not a step-change, when the stock is discounting a 3x.
That is the recurring pattern: each of these names had already sold its best-case future to shareholders. There was nothing left to surprise with, so any hint of a plateau became the trade.
Intel fits as the third data point, and only that. Its data center grew 59% and its stock still fell — but the specific mechanics there (the foundry losses, the capex burden) belong to a different conversation. What matters here is only the shape: strong number, forward caution, stock down. The Q3 guide midpoint that came in soft relative to expectations did the same work Cisco's 2027 single-digit outlook did. The market bought the deceleration, not the level.
Which brings us to the uncomfortable part for people who own CSCO.
The sell-the-news reflex isn't a one-off in this name. Two quarters ago, Cisco's Q2 2026 reaction was, per eToro's earnings data, -12.3% on the day. That wasn't a dip that got bought back — it was still down 10.2% five days later and down 8.4% at 30 days. The reaction held.
Go back further and the record is littered with it: -17.5% in Q3 2022, still -14.2% at 30 days. -9.6% in Q1 2024, still -6.2% a month later. Across 18 tracked quarters, the average earnings-day move is essentially flat, at -0.2%. This is not, arguably, a stock that reliably melts up on good prints. It reads as a stock that punishes crowded positioning.
So consider Cramer's call that Thursday's plunge is a buying opportunity. Strip away the framing and look at what the bet actually requires.
Buying here is a wager that embedded expectations reset lower faster than the business decelerates — that the price finds a level cheap enough to absorb a single-digit FY27 before the single-digit FY27 actually shows up in the numbers.
That can absolutely work. KeyBanc stays bullish, arguing Cisco takes share as hyperscalers and neoclouds ramp capex. The order book supports the story: $4 billion of hyperscaler infrastructure orders in the quarter, $9.3 billion for the year, with roughly $4 billion of hyperscaler revenue expected to nearly double to $7.5 billion in FY27. The demand is real and the pipeline is visible.
But none of that resolves the actual disagreement on the tape. The disagreement isn't about whether Cisco sells more gear next year. It's about whether the growth rate is peaking, and what multiple you pay for a business whose growth rate is peaking.
That is the whole story of these three days. The market stopped trading the level of AI demand and started trading its slope. When you own a name that ran 60% into a print, the beat was your entry ticket, not your catalyst — and the only thing left to move the price is the first credible sign that the curve is bending.
Cisco's own guidance, however prudently intended, was that sign. Cerebras's raise was too small to override the 3x already in the price. Intel's forward guide undercut its blowout quarter.
None of this says AI infrastructure spending is over. It says the easy repricing is over. From here, the trade is about the second derivative — and, if history holds, Cisco's sell-the-news reflex has enough teeth that patience, not conviction, is what gets paid.