In short: The better of the two AI-spend beneficiaries in the mailbag, on the gross-margin test: "Marvell is in a different bucket where they've got obviously very high gross margins. They're helping some of the best companies in the world to turn out ASICs like Google." "I think Marvell's much more interesting than a Celestica." Comparative preference rather than a sized call.
Marvell helps big tech companies like Google design custom AI chips (ASICs — chips built for one specific job). Niles uses gross margin — the share of each sales dollar left after the direct cost of making the product — as a quick test of how much real value a company adds. Marvell's is very high, meaning customers pay up for its engineering, so he finds it "much more interesting" than Celestica. The flip side, which he notes in the Nvidia discussion, is that Marvell's custom-chip work is exactly what creates tension between Nvidia and its biggest customers.
41:16Celestica, you can see what the gross margins are, get an idea what the value add is. Marvell is in a different bucket where they've got obviously very high gross margins. They're helping some of the best companies in the world to turn out ASICs like Google, right? There was an announcement recently of them working with Google and how much Google could end up with, that they spent a certain amount of money.
In short: Subscriber view, passing: named with Broadcom as the custom-silicon partners hyperscalers use to reduce their NVIDIA dependence — "but this takes time."
13:00Hence, the strategic logic of acquiring Hugging Face, which Nvidia acquired, I believe, last week. Hyperscalers hate this dynamic. They are responding by developing their own silicon, working with Broadcom, Marvell, and others, but this takes time. His point here is that the hyperscalers know that they are overly dependent on Nvidia, and they hate it, and they are looking for alternatives. I continue.
In short: Named once, as the second half of the failed-catalyst pair: high-beta momentum could not get "that shot of adrenaline off of Nvidia's earnings, the derivative trade, and off of Marvell's earnings" (Terranova). Cited as an input to the sentiment reading rather than as a call — a softening from the outright negative framing of 2026-AUG-31, since Terranova's conclusion here is that the group may get "a little bit of a recovery rebound."
In short: The clean illustration of the momentum failure. Wapner, citing Wolfe Research: "Wolfe points out on Marvell, the rally's come to a screeching halt. If that trade has come to a halt, is it a problem or not?" Talkington adds it to the tape reading on Dell — "with NVIDIA trading down, Marvell trading down, who knows what the stock will do." Nobody defended it.
Marvell designs networking and custom AI chips, and it is the show's clearest example of the momentum trade breaking: Wolfe Research's phrase, quoted by Wapner, is that "the rally's come to a screeching halt."
Nobody on the desk defended it, and Talkington cites it — alongside Nvidia — as the reason a good Dell report may not be rewarded. When the fastest-moving names in a theme stop responding to good news, that is the signal the money that was chasing them has already left.
In short: A genuinely large guidance raise that the tape had already spent. "Marvell upgraded its long-term outlook, projecting FY28 revenue to reach 18 billion, up 50% year-over-year, but the stock still didn't rally given this was priced in." The underlying quarter was fine (revenue $2.7B, +37%; data centre $2.2B, +46%; Q3 guided $3.15B vs $3.03B) and management now sees FY27 revenue up ~45% to ~$12B with custom chip revenue more than doubling and AI bookings "exceptionally robust." The lesson he is drawing is about the semiconductor tape rather than Marvell: in a group where SOX forward P/E has gone from 21-22× to ~15×, upward revisions no longer move price.
Marvell designs custom chips for specific customers — the alternative to buying Nvidia's off-the-shelf product. It raised its long-term outlook substantially, now expecting revenue of about $18 billion in the 2028 financial year, up 50% year on year, with custom chip revenue more than doubling.
The stock did not move. "The stock still didn't rally given this was priced in."
That non-reaction is the useful signal, and it belongs with the semiconductor de-rating above. In a sector where the multiple has fallen from 21-22 times earnings to about 15, a company can raise its forecasts materially and get nothing for it, because the market has already assumed the good news and is now discounting the risk that it does not persist. When upward revisions stop moving prices, the market has changed what it is worried about.
Full passage: premium transcript (PDF).
In short: Custom silicon breakout — and a beat-and-raise that still fell. Q2 FY27 (ending August 1) revenue +37% Y/Y to a record $2.74B ($30M beat) with non-GAAP EPS of $0.94 ($0.01 beat). Data Center revenue jumped 46% to $2.17B, 79% of total sales, "as demand for AI infrastructure remained exceptionally strong." The next leg is custom: management expects "a significant acceleration beginning in H2 FY27, followed by custom silicon revenue more than doubling in FY28 as hyperscalers increasingly design their own AI chips," and Marvell expanded its Google partnership across AI accelerators, storage, networking and near-memory compute, "backed by a six-year warrant agreement." Connectivity is the second engine — optical interconnect demand accelerating as AI clusters need more bandwidth, with the CXL memory-expansion business called a "home run" across multiple hyperscalers. Guidance moved up on both years: FY27 to ~$12B (from $11.5B) and FY28 to $18B (from $16.5B), with Q3 revenue of $3.15B well ahead of consensus and implying "another ~15% sequential jump." Bottom Line: "But expectations are already enormous: the stock trades over 50x forward earnings, and much of the Google opportunity through FY28 was already embedded in guidance. That helps explain why shares fell despite Marvell raising both FY27 and FY28 outlooks."
Marvell designs two things for AI data centres. First, the chips that move data around — optical interconnects that carry information between racks as light, and memory-expansion parts that let a server address more memory than it can physically hold. Second, and increasingly, "custom silicon": when Google or Amazon wants its own AI chip rather than buying NVIDIA's, it hires a company like Marvell to help design and produce it.
The quarter was excellent. Revenue grew 37% to a record, the data-centre business grew 46% and now makes up 79% of the company, and management raised its forecast for both this year and next — to $12 billion and $18 billion respectively. It expanded its partnership with Google across accelerators, storage and networking, backed by a six-year warrant agreement, which means Google took the right to buy Marvell stock. A customer accepting equity exposure to its supplier is a genuine commitment signal.
The stock fell anyway, and understanding why is the transferable lesson. Marvell trades at more than 50 times expected earnings, which means the price already assumes years of the growth just described. And the newsletter identifies the specific problem: much of the Google opportunity through 2028 was already in the guidance before this announcement. So the news added enthusiasm without adding numbers — investors learned nothing that changed the forecast they were already paying for.
That is why the view here is neutral rather than positive despite an unambiguously strong print. The business is doing well; the question is whether anything is left over after the expectations are subtracted. Analysis, not a recommendation.
In short: Cited at "200 billion" in Brown's list of AI-levered market caps (with Arista at $235B and Broadcom at $1.7T) that will "trade up or down based on what the AI Fed tells us" next Wednesday. Two sessions after the Alphabet custom-chip option that moved it 10% (2026-aug-19), it gets no fresh committee stance here.
In short: The day's big mover — up nearly 10% and off the best levels — after a custom-chip deal that gives Alphabet an option to buy a $12.2 billion stake. Terranova owns it and gets first crack: "this is consistent with the type of financial arrangements that we've heard with the hyperscalers. It's very important to have customized chips now existing with Alphabet — that is a relationship with Broadcom." He reads Broadcom's decline as a second-order worry rather than a share-loss verdict: "I don't think necessarily Broadcom is down on the concerns that Marvell is going to take market share on the custom chips for the TPUs. It's more about in the future, if Marvell builds the relationship with Alphabet, do they turn to Marvell for silicon purchases — that would be detrimental to Broadcom." On the stock: "Marvell needed this. The stock had a bit of a pullback recently towards the end of June. It is restarting the momentum. This is a good deal." The structure matters — of the ~59 million shares, 57 million are contingent on the relationship and on Alphabet actually delivering future purchases.
Marvell designs custom chips — semiconductors built to one customer's specification rather than sold off the shelf. Today Alphabet gave Marvell a deal that includes an option for Alphabet to buy a $12.2 billion stake in it, and the stock jumped about 10%.
Read the structure before the headline: of roughly 59 million shares in the arrangement, 57 million only vest if the relationship actually produces future chip purchases. So it is a very large option on execution, not cash in the door. What it does buy immediately is credibility — Joe Terranova calls it the same kind of financial arrangement the big cloud companies have been striking all year, and says the stock "needed this" after a pullback at the end of June: the momentum is restarting.
The subtlety worth carrying forward is what it means for Broadcom, which has been Alphabet's custom-chip partner. Terranova argues Broadcom's decline today is not about losing share on the current chips — it is about the possibility that, once Marvell is embedded, Alphabet buys more of its silicon from Marvell in future.
In short: Link owns it — up 14% on the week as semis (both the SOX and SMH) post their best week since mid-June and momentum has its best week since mid-June, +8% from the July 29 low. Part of the laggard-catch-up trade she is leaning into alongside Broadcom and NVIDIA.
In short: Baruch bought it — the committee move — after a 37% July drawdown: "absolutely, very high quality name." He waited for confirmation ("we didn't want to really buy the name just yet… let's see how South Korea goes overnight and see if there's follow-through"), and after it rose 14% on Tuesday he bought anyway, "half as much as we were hoping for, which is a 1% add to our portfolios." The thesis: "they're emerging as the connectivity backbone with AI custom silicon, optical networking and AI memory solutions. CXL — Compute Express Link — allows hyperscalers to expand and pool memory outside of processors and works as a complement with high-bandwidth memory. I think they're going to be a huge leader as this AI build-out continues."
Marvell makes the chips that move data between processors rather than the processors themselves — custom AI silicon, optical networking, and memory-expansion technology called CXL that lets data centers pool memory outside the chip itself, complementing the expensive high-bandwidth memory everyone is short of. Bill Baruch calls it "the connectivity backbone" of the AI build-out.
Just as instructive is how he bought it after a 37% July collapse: he deliberately waited overnight to see whether South Korean markets followed through rather than catching the falling knife, and when the stock had already jumped 14% he bought anyway — but only half the size he'd planned, a 1% portfolio addition. Confirmation first, then a smaller position to pay for the worse price.
In short: One of the four names Terranova says the market must not re-concentrate into if the broadening narrative is to hold. No individual call this episode.
In short: Link owns it and has been trimming into strength ("it's up a whole lot") but "still like it very much" — part of her barbell of owning both semis and software. Down almost 8% on the day as the hawkish-Waller hit landed hardest on the long-duration semis.
Marvell makes custom AI and networking chips. Stephanie Link owns it and has been trimming (selling a bit) into its big run to lock in gains, but "still like it very much" — it's part of her deliberate "barbell" of holding both chip stocks and software. It fell almost 8% on the day because the hawkish Fed news hit the fast-growing, high-valuation chip names (so-called "long-duration" assets) the hardest.
In short: Stephanie Link owns it — up ~100% and "that thing scares me because it goes up a lot and goes down a lot"; her slice of semiconductor momentum, held alongside the buy-low NVIDIA she prefers. (Also cited with Intel/AMD among the semis bouncing hard on the day.)
Marvell makes custom chips used in data centers and networking — a direct AI beneficiary. Stephanie Link owns it and is candid that it "scares me because it goes up a lot and goes down a lot"; she's up about 100% on it. It's her deliberate slice of high-volatility semiconductor momentum, which she balances against the calmer, buy-low approach she took with Nvidia.
In other words, she's willing to hold one wild momentum name for the upside, as long as the rest of her semi exposure is the cheaper, steadier kind.
In short: One of her two semis (with Broadcom). $10 of earnings power by 2027; Nvidia invested $2 billion; a $2B buyback; the optical business can grow 50% for a couple of years and custom-ASIC (80% share with Broadcom) is expected to grow 20–100%. At 27x not super cheap — she'd buy a bigger dip.
Marvell designs specialized chips for data centers — including the high-speed "optical" parts that move data between AI servers and the "custom ASICs" (chips built to order for one customer, like a hyperscaler's in-house AI processor). Link owns it as one of just two semiconductors she holds, and values it on what it can earn down the road: about $10 of earnings power by 2027. The supporting facts she cites — Nvidia putting $2 billion into the company, a $2 billion buyback (the company buying back its own shares), an optical business that can grow 50% for a couple of years, and a custom-chip business that could grow 20–100% (it shares ~80% of that market with Broadcom).
At 27 times earnings she admits it isn't cheap, so her plan is patient: "I'd buy a bigger dip" — wait for a deeper pullback rather than chase it here.
In short: Referenced as the destination for Adobe's departing CFO — Dan Durn leaves June 15 to become Marvell's CFO, the leadership-turnover detail behind Adobe's read.
In short: Exhibit A of the pump dynamic: Nvidia put $2B in, Jensen went on live TV calling it "a trillion-dollar company," and the stock jumped ~25% — days before another big AI selloff.
37:54and investors are worried about circular deals being set up with SPVS where Nvidia will invest in a supplier just the other day it was very funny that Marll you know where effectively Jensen invested two billion of from Nvidia's balance sheet he went on live TV and said Marll is going to be a trillion dollar company and the stock was up like 25%.
Nothing matches this filter.
Verbatim excerpts from the public transcripts (auto-pulled at each mention's timestamp, lightly cleaned). Timestamps link into the video; "source page" opens that commentator's full analysis at this row. Click a mention's header line to expand it (one open at a time). For personal study — not investment advice.