Broadcom Beat. The Stock Broke. The Layer Held.
Broadcom beat on revenue and earnings and grew its AI business triple digits — then fell 14.6% Thursday, because the forward AI guide missed an elevated bar and management did not raise the full-year number. But the read-through never came: Marvell fell 3.4%, Astera 2.8%, Credo 0.8%, Nvidia a tenth of a percent. The market repriced the most-crowded name, not the trade — and the mean analyst target never moved off $508.75.
Broadcom's one-day reaction to its own earnings beat
−14.6%
Last night's preview asked one question: would Broadcom sell the beat? It did — and harder than the model's soft-guide case. AVGO printed a clean beat (revenue +48%, AI revenue +143% year over year) and still dropped 14.6% to $409.15, because the Street was positioned for a beat-AND-raise and got a beat with an in-line-to-soft AI guide. That part graded the preview right: the cushion was too thin, the insiders had sold, and “watchful, not bullish” was the correct read. What the preview got WRONG is the more interesting half. The invalidation case said a soft guide would drag the custom-silicon layer down 10–20% on read-through. It didn't happen. Marvell fell 3.4%, Astera 2.8%, Credo 0.8%, Nvidia and Cisco a tenth of a percent each; only Arista (−7.4%) partly followed. The selloff was concentrated on the one name everyone owned, not the trade everyone is in. The Street confirms it: not a single analyst cut Broadcom, the mean target held at $508.75, and after the drop the stock sits 24% BELOW that target — the fall created room, it didn't break the thesis. So this was an expectations reset, not a demand break: Broadcom walked in overbought at RSI 73 and merely met a bar it had already been paid for, while it still models +61% revenue growth this year. The deeper cut is that “Broadcom vs Marvell” isn't one trade at all — AVGO is a quality-and-value stock (quality 62, value 71, momentum 26), MRVL is a pure-momentum stock (momentum 81, trading 40% above its target). They share a label and nothing else. Broadcom still owns the AI profit pool. Thursday just proved that owning it isn't the same as being paid for it.
The contradiction
Broadcom beat and still fell 14.6% — alone. Its peers barely moved and the Street cut nothing. The premium reset, not the demand.
What the headline says
Broadcom flopped — the AI chip trade is cracking
AVGO −14.6% on earnings; semis broadly lower with Micron and ARM; "earnings flop" leads the tape
What the data says
The leader reset; the layer barely moved
MRVL −3.4%, ALAB −2.8%, CRDO −0.8%, NVDA −0.1%; mean AVGO target held $508.75 → +24.3% upside; zero downgrades
Chapter 01
Broadcom Fell 14.6%. Its Custom-Silicon Peers Barely Moved.
Line up Thursday’s one-day reaction across the AI-networking complex and the picture is lopsided: Broadcom −14.6%, then a long gap to Arista −7.4%, Marvell −3.4%, Astera −2.8%, Credo −0.8%, Nvidia and Cisco a tenth of a percent each. A weak Broadcom print was supposed to be a sector event. Instead it was a single-name event with one partial sympathy move.
The tape that mattered was not Broadcom’s number — it was everyone else’s reaction to it. Broadcom fell 14.6% to $409.15, its sharpest single-session drop of the AI rally, after a quarter that beat on revenue and earnings. If the AI-silicon demand thesis had genuinely cracked, the read-through names would have followed. They didn’t. Marvell fell just 3.4%, Astera 2.8%, Credo 0.8%, and Nvidia and Cisco were essentially flat at −0.1% each; only Arista, the Ethernet-switching name closest to Broadcom’s networking franchise, partly followed at −7.4%. The median peer fell about 1.8% — a ~13-point gap to Broadcom’s drop. That spread is the whole story: the market repriced the most-owned name in the group, not the group.
One-day reaction the day after the print
Source: MarketDecode scanner, 1-day total return, 2026-06-04
The Break
AVGO −14.6%$409.15; beat on revenue + EPS, fell on the guide
The Non-Reaction
NVDA / CSCO −0.1%GPU + legacy networking essentially flat
The Gap
~13 ptsAVGO vs ~−1.8% peer median
Chapter 02
Priced for Perfection: the Day Cooled Broadcom, Not the Layer
Broadcom walked into the print overbought at RSI 73 — the most-loved it had been all year. One session reset it to 43.7, dead-neutral. But the selloff cooled only Broadcom: Marvell is still at RSI 79, Cisco 77, Astera 70 — all still in overbought territory. The layer never corrected; the leader did.
A 14-handle drop sounds like capitulation, but the momentum picture says it was a reset of one stretched name. Broadcom entered Wednesday’s print at a 14-day RSI of 72.5 — firmly overbought, the richest reading in the group — and exited Thursday at 43.7, a 29-point collapse into neutral. That is exactly what an expectations reset looks like: the air came out of the most inflated name. What it did NOT do is cool the rest of the layer. Marvell still reads RSI 79.3, Cisco 77.3, Astera 70.5 — all still overbought, none of them corrected. Nvidia sits at a calm 51.4. So the one name that mean-reverted is the one that printed; the names that are arguably MORE stretched are still sitting at altitude, which is the risk to watch from here.
RSI-14 across the AI-networking layer
Source: MarketDecode scanner, 14-day RSI, 2026-06-04
Most Overbought
MRVL RSI 79Never corrected; +82% over 20 days
The One That Reset
AVGO RSI 44Only name back to neutral
The Move
−29 RSI ptsAVGO 72.5 (Wed) → 43.7 (Thu)
Chapter 03
The Demand Didn't Break — the Guide Did
The thing that fell 14.6% was the expectations, not the business. Broadcom still models +61% revenue growth this year — second only to Nvidia’s +81% in the group, ahead of Marvell’s +40%, Arista’s +29% and Cisco’s +11%. The AI revenue line that "disappointed" still grew triple digits year over year. What missed was a single forward guide against a bar that had been raised again and again.
Separate the two things that happen on an earnings night: what the business did, and what the stock was priced for. Broadcom’s business is fine. Consensus still models its revenue growing about 61% this fiscal year — the second-fastest in the AI-networking group behind Nvidia’s 81%, and well ahead of Marvell’s 40%, Arista’s 29% and Cisco’s 11%. The reported quarter beat, and its AI revenue grew triple digits from a year ago. What actually moved the stock was forward and narrow: the next-quarter AI chip guide came in roughly in line to soft against a Street bar that had been ratcheted up for months, and management didn’t raise the full-year AI figure. That is a guide miss against expectations, not a demand miss — a distinction the 14.6% drop blurs and the peer non-reaction confirms.
Consensus revenue growth, current fiscal year
Source: MarketDecode forward-estimate feed, current-year revenue growth, 2026-06-04
Fastest Grower
NVDA 81%The reference point — still the demand leader
Broadcom
61%2nd-fastest in the group, after the drop
The Laggard
CSCO 11%Legacy networking; slowest in the layer
Chapter 04
Two Businesses Wearing One Trade
Broadcom and Marvell get traded as one "AI networking" idea, but their factor fingerprints are mirror images. Broadcom scores high on value (71) and quality (62) and low on momentum (26) — a scaled, profitable, cash-rich name. Marvell scores high on momentum (81) and middling on value (49) and quality (48) — a pure price-trend name. The market just stopped paying for the difference.
The cleanest way to see why "who owns the AI profit pool" is the wrong question is to put Broadcom’s and Marvell’s factor profiles on the same axes. Broadcom’s shape leans hard toward value (71) and quality (62) with low momentum (26) — the signature of a large, profitable, cash-generative business with a second software engine bolted on, whose stock had simply run ahead of itself. Marvell’s shape is the inverse: momentum 81, value 49, quality 48 — a smaller, faster, story-driven name riding a price trend. These are not two versions of the same trade; they are opposite stocks that happen to share a customer base. That is why the same headline hit them so differently: a momentum name shrugs off someone else’s guide miss, while the quality name that was priced for perfection takes the full reset. Owning the profit pool (scale, margins, the dual chip-plus-software engine) is Broadcom’s; owning the momentum is Marvell’s.
Factor profile: Broadcom vs Marvell
Source: MarketDecode factor model (0–100 scale), 2026-06-04
Marvell’s Spike
Momentum 81Group-high; AVGO sits at 26
Broadcom’s Base
Value 71 / Quality 62Scaled, profitable, dual-engine
The Inversion
55-pt momentum gapOpposite profiles, one trade label
Chapter 06
Verdict: the Street Didn't Cut — Broadcom Fell Into 24% of Upside
Plot Thursday’s reaction against each name’s room to its analyst target and Broadcom lands in the telling quadrant: it fell 14.6% straight INTO +24.3% of upside, because the mean target held at $508.75 and not one analyst cut it. Marvell and Astera, by contrast, are still trading above their targets even after the dip. The drop removed Broadcom’s premium without touching the thesis.
End where the evidence points. The vertical axis is Thursday’s reaction; the horizontal axis is upside to the mean analyst target. Broadcom sits in the bottom-right — it fell hard, but the fall pushed it to +24.3% below an unchanged $508.75 target, with zero downgrades and a still-neutral options tape (put/call 0.43). That is an expectations reset, not a thesis break: the Street did not revise the business down, it let the price come back to the model. Marvell and Astera sit on the left — still 40% and 16% ABOVE their targets respectively, even after small dips, because momentum carried them past where analysts will follow. Nvidia is the calm reference: flat on the day, 35% of upside still on the table. So the profit-pool answer is unromantic. Broadcom still owns the pool — the scale, the margins, the dual engine, the analyst conviction (43 of 47 still at buy). Thursday simply proved that owning the pool is not the same as being paid a premium for it. The grade: the preview’s "watchful, not bullish" was right; the next test is whether AVGO reclaims its 20-day line into next week while the stretched peers hold.
One-day reaction vs upside to analyst target
Source: MarketDecode scanner + analyst targets, 2026-06-04
Fell Into Upside
AVGO 24.3%Below an unchanged $508.75 target
Above Target
MRVL −40.3%Momentum ran past where analysts follow
The Tell
0 downgradesMean PT held; options tape neutral (P/C 0.43)
Resolution window — 1 week
What would confirm or invalidate this read
Confirmation
Over the five sessions after Thursday June 4, the read-through stays muted: MRVL, ALAB, CRDO and NVDA do NOT fall an additional 8–15% on delayed Broadcom sympathy, and AVGO stabilizes near or reclaims its 20-day line (~$430) against an unchanged $508.75 mean target with no analyst downgrades. That confirms the thesis — Thursday was a single-name expectations reset, not a demand break, and the custom-silicon layer is intact.
Invalidation
In the same window the layer rolls over after all — MRVL/ALAB/CSCO mean-revert hard (their RSI 70s+ unwind) and/or the read-through names give back 10–20% on a broader “AI capex is slowing” narrative, with sell-side targets across the basket getting cut. That would mean Thursday was the first crack of a sector de-rating that simply started with the most-crowded name, and the peer non-reaction was a one-day lag, not a divergence.