The Verdict — Marvell's Data Center Hits 76% of Revenue, Dell Blows Out 33%, and the Smart Money Quietly Moved Into the Buildings

The Lead

Issue 08 set the test for the picks-and-shovels layer: “If MRVL’s Q1 FY27 shows ASIC revenue accelerating, it validates that the picks-and-shovels layer is capturing value even as NVDA dominates the headline GPU number.” The preview that ran on 27 May raised a specific bar — data centre revenue growth above 35% year-on-year.

Marvell reported Tuesday night. Net revenue was $2,417.8 million, up 27.6% year-on-year and ahead of the company’s own $2.40 billion guide. Data center revenue was $1,832.7 million — 76% of the entire company, up from a 76% share a year ago on a base that was a third smaller. The market’s verdict was emphatic: the stock extended its 2026 doubling, analysts at RBC and Barclays raised price targets, and at least one outlet declared Marvell “Broadcom’s most credible rival.”

And yet our own bar missed. Data center grew 27.2% year-on-year — solid, but a full eight points below the 35% line we drew. This is the discipline the Digest is built on: the directional call (Marvell is executing on the connectivity-and-custom-silicon thesis) was right; the specific threshold was too aggressive. Forecast [2026-05-27-001] is graded busted. More on why the 27% number is more bullish than it reads below.

Two days later, Dell delivered the other half of the thesis. The stock surged roughly 33% on an AI-server demand signal so strong it dragged Super Micro up about 16% — indicted board member and all — and pulled HPE and Arista along behind it. The S&P 500 and Nasdaq 100 closed the week at record highs.

The thing to hold in your head as you read on: a meaningful slice of Friday’s record-high tape was a US–Iran ceasefire rumour, not an earnings number. That distinction is the subject of the fourth thread.

Threads

Marvell — data center is 76% of the company now, but our >35% bar missed

Marvell Technology Q1 FY27 (quarter ended 2 May 2026), per the 8-K earnings release and the 10-Q filed the following day:

Metric	Q1 FY27	Q1 FY26	Y/Y
Net revenue	$2,417.8M	$1,895.3M	+27.6%
Data center revenue	$1,832.7M	$1,440.6M	+27.2%
Data center % of total	76%	76%	flat
Gross profit (GAAP)	$1,260.8M	$952.4M	+32.4%
GAAP gross margin	52.1%	50.2%	+1.9pts
Operating income (GAAP)	$339.4M	$270.6M	+25.4%
Net income (GAAP)	~$34.5M	~$177.9M	—

Two numbers demand explanation.

The first is the 27% data center growth. On the face of it, that busts our 35% bar and looks like a deceleration story. It isn’t, for a structural reason: data center is now 76% of Marvell. When a segment is three-quarters of the company, its growth rate is the company’s growth rate — there’s no longer a small, fast-growing sliver flattering the average. Marvell has finished the transition from a diversified analog-and-storage business into an AI-data-centre business with a communications tail. The 10-Q’s end-market table makes it explicit: data center $1,832.7M, communications and other the entire remainder. The custom-silicon ASIC programs for hyperscalers — the part Baker’s thesis actually rides on — sit inside that data center line and were not separately broken out this quarter. So the verifiable metric (total data center, +27%) missed our bar; the sub-line we actually cared about wasn’t disclosed. We grade against what we can verify, and we missed.

The second is the GAAP net income collapse to ~$34.5 million. Operating income was healthy at $339.4M. Below the line, a $203.3 million “other expense, net” — versus $6.0M a year ago — gutted the GAAP bottom line. That charge is acquisition-related: Marvell closed its purchase of Celestial AI during the quarter, and the 10-Q also references an unrealised mark on a forward stock purchase contract. Celestial is a photonic-fabric, scale-up-interconnect company; the filing states the deal is meant to “accelerate the Company’s connectivity strategy for next-generation AI and cloud data centers.” Strip the one-off and the operating business is compounding revenue at 28% with margins expanding nearly two points. The non-GAAP EPS the street trades on — consensus was around $0.79 — was unaffected by the GAAP mark.

Here is why this matters more than the busted bar. In Issue 08, Baker’s pivot looked like a stretch: he cut NVDA 77% and bought the connectivity layer (Astera Labs +109%, Credo NEW) on a thesis that AI scaling would become a bandwidth problem, not just a compute problem. Marvell just spent acquisition money to buy a photonic interconnect company and booked the charge — and in the same week, the press carried “Nvidia Bets Big on Photonics as AI Infrastructure Needs Grow.” When the GPU monopolist and the leading ASIC challenger both move on photonics in the same five days, the connectivity thesis stops being a contrarian’s hedge and becomes consensus infrastructure. Baker was early and wrong on NVDA timing. On the layer, he was right.

Forecast [2026-05-31-001]: AVGO Q2 FY26 (reports 3–4 June) will show AI/custom-silicon revenue growth exceeding 50% YoY — the fastest of the three ASIC reporters this fortnight. Confidence 0.60. Horizon: 5 days.

Dell’s 33% blowout confirms the supply chain end-to-end — SMCI rides along despite the indictment

If NVDA’s $81.6 billion quarter (Issue 08) was the chip, and Marvell’s $1.83 billion data center line is the custom silicon and interconnect, Dell is the box that ships to the customer. This week it confirmed the demand is flowing all the way down.

Dell surged roughly 33% in a single session on an AI-server outlook the market read as a step-change in hyperscaler spend. The read-through was immediate and indiscriminate: Super Micro added about 16%, Hewlett Packard Enterprise jumped on the sector sentiment, and Arista — the networking layer between the servers — caught a bid. The whole integrator-and-fabric tier re-rated on one company’s guidance.

The Super Micro move is the tell worth pausing on. The Issue 09 preview led with SMCI’s federal export-control indictment: the SDNY charged a sitting board member, and Jensen Huang used a Taiwan visit to publicly demand the company tighten compliance. That overhang has not resolved. And yet SMCI rallied 16% on Dell’s number — because when AI-server demand is the binding question, a governance cloud over one integrator is a second-order concern. The market is saying the pie is growing fast enough that even a compliance-impaired player gets fed. That is either confirmation of how strong the demand signal is, or a warning about how little the marginal buyer is discriminating. Probably both.

For the watchlist, the clean read is the one Issue 08 set up: NVDA’s print was not a single-company event, it was the top of a supply chain, and every layer below it — custom silicon (MRVL), system integration (DELL, SMCI, HPE), networking (ANET) — printed confirmation within nine days. The thesis arc from Issue 03 (hyperscaler capex running north of $400 billion annualised) now has corroboration at four distinct layers of the stack.

The convergence moved into the buildings — DLR, EQIX, CRWV

This is the thread that justifies tracking 13Fs instead of just reading earnings. The structured positioning data refreshed this week, and the story is no longer about who owns the chip.

The chip layer is saturated. NVDA shows the highest convergence in our universe — seven of eleven tracked funds: Atreides, Situational Awareness (Aschenbrenner), D1 Capital, Altimeter, Whale Rock, Dragoneer and Coatue. GOOG carries six (Atreides, D1, Lone Pine, Whale Rock, Coatue, Berkshire). META carries four. These are crowded, consensus longs — everyone who was going to own the hyperscalers and the GPU maker already does.

What’s new is one layer down, in the physical plant:

Ticker	What it is	Convergence	Funds
DLR	Digital Realty — data-centre REIT	score 23.5, bullish	Situational Awareness, Whale Rock
EQIX	Equinix — interconnection/colocation REIT	score 7.3, bullish	Coatue
CRWV	CoreWeave — GPU cloud	score 9.3, bullish (contested)	Atreides, Situational Awareness, Altimeter

These scores are small — this is early, thin positioning, not a crowded trade. But the direction is one-sided and the names are consistent. Aschenbrenner’s Situational Awareness is the purest expression: he holds NVDA (the chip), DLR (the building), and CRWV (the cloud that rents the chips inside the building). That is a vertically integrated bet on AI compute from silicon to slab. When the most thesis-driven allocator in the universe is positioned down the entire physical stack and the data-centre REITs are showing up in convergence for the first time, the smart money is telling you where the next constraint binds: not chips, not capital — power and physical capacity.

This rhymes with Issue 08’s power thread (the NextEra–Dominion $67 billion deal, driven explicitly by Virginia data-centre demand). The buildings and the grid that feed them are becoming the scarce asset. CRWV’s contested score (a 48% contradiction rate — some evidence points bearish) is the honest caveat: CoreWeave is the riskiest expression, levered and customer-concentrated, and not every fund agrees. But the REITs are clean bullish.

Forecast [2026-05-31-002]: At least one tracked fund will initiate or add materially to a data-centre-REIT position (DLR or EQIX) in the Q2 2026 13F filings (mid-August). Confidence 0.55. Horizon: 80 days.

Record highs on a ceasefire rumour — the overcrowding tell

Now the steel-manned bear case, because the tape this week was not purely a fundamentals story.

The S&P 500 and Nasdaq 100 closed the week at record highs. Part of that was Dell and the AI-server complex. But a meaningful part was a US–Iran ceasefire rumour — “Stock Market Hits Highs On Iran Deal Hopes,” “S&P Futures Gain on Hopes for U.S.-Iran Deal” ran across the wires the same day. When an index makes an all-time high partly on a geopolitical headline that can reverse on the next news cycle, the move is borrowing from sentiment, not earnings.

And the same press feed that carried the Dell euphoria carried the counter-signal: “The AI Trade Is Getting Overcrowded. Here’s Why You Should Opt for the iShares Semiconductor ETF” — the classic late-cycle tell, where the advice shifts from picking winners to diversifying because everyone already owns the winners. “Super Micro Stock Rally Misses the Point From Dell Earnings” was running alongside the +16% move.

The positioning data makes the overcrowding concrete rather than vibes. When seven of eleven elite funds own NVDA and six own GOOG, the question isn’t whether the thesis is right — it plainly is, the earnings keep confirming it — but who is the marginal buyer. A maximum-conviction, maximum-weight consensus long has no one left to convert. That is precisely the configuration where good news produces a muted reaction (NVDA’s 1.5% drop on a $30 billion beat, Issue 08) and any disappointment produces an outsized one. The single genuinely contested name in our convergence set is SQ (Block) — bearish, 80% contradiction, one fund — the lone short-ish signal in a sea of crowded longs. That asymmetry is the risk: the book is one-directional.

None of this says the AI-infrastructure thesis is wrong. Every layer printed confirmation this fortnight. It says the trade is crowded, the marginal new dollar is buying an ETF rather than a single name, and the most recent leg of the index move leaned on a ceasefire headline. When the fundamentals are this good and the stock reactions are this muted, you are being paid less and less to hold the consensus.

Since the Preview (Issue 09, 27 May)

The 27 May preview flagged two earnings-week stories before the results were in. Resolutions:

Story	Preview read	Now
MRVL Q1	”Reports tonight — Baker’s thesis on trial”	Beat on revenue (+28%), data center 76% of company, Celestial closed. >35% growth bar busted at +27%; directional call right.
ZS FCF guide-down	-19% on capex confession; “FCF is the new growth filter”	Stands. The market’s profitability filter is the bear case running underneath the whole AI-software complex; watch PANW (2 June) for the peer read.
SMCI indictment	Board member charged; “bearish-watch”	Overhang unresolved, but the stock rode Dell +16%. Demand is overwhelming governance — for now.
DELL	”the NVDA flow-through proxy”	Confirmed emphatically: +33%, dragged the whole integrator layer.

Watchlist Updates

Ticker	Direction	Δ vs prior	Driver
MRVL	bullish	↑ (from watch)	Q1 $2.42B (+28%), data center 76% of company, Celestial photonic deal closed. RBC + Barclays PT raises.
DELL	bullish	→	+33% on AI-server outlook. The integrator-layer confirmation of the NVDA demand signal.
SMCI	bearish-watch	→	Rode Dell +16% but the SDNY export-control indictment is unresolved. Demand masking governance.
AVGO	bullish	→	Reports 3–4 June. The cleanest custom-silicon read of the fortnight. Aschenbrenner NEW $1.0B.
NVDA	bullish	→	7-of-11 fund convergence — most crowded long in the universe. Issue 08 thesis intact; marginal-buyer risk rising.
DLR	bullish	NEW	Data-centre REIT enters convergence (Aschenbrenner, Whale Rock). The physical-layer bet.
EQIX	bullish	NEW	Interconnection REIT, Coatue. Thin but directional.
CRWV	bullish (contested)	NEW	GPU cloud, 3 funds, but 48% contradiction. Riskiest expression of the physical-layer thesis.
CRDO	bullish	→	Baker’s connectivity-IC position; Marvell’s Celestial deal validates the layer.
ZS	bearish	→	FCF guide-down stands as the template for the AI-software profitability filter.

The Week Ahead

Date	Event	What confirms / what breaks
Mon 2 Jun	PANW Q4 FY26, GTLB Q1 FY27	PANW is the ZS comp. Does the FCF-margin filter hit the cybersecurity leader too, or was ZS idiosyncratic?
Tue–Wed 3–4 Jun	AVGO Q2 FY26, CRWD, HPE, CIEN, MDB	The fortnight’s main event. AVGO custom-silicon/AI revenue is the cleanest share-shift read vs NVDA. Beat validates Aschenbrenner’s $1.0B AVGO and Baker’s connectivity thesis simultaneously. CIEN is the optical-transport read-through on the photonics theme.
Tue 9 – Wed 10 Jun	ORCL Q4 FY26, CRWD	Oracle’s RPO backlog (Issue references: $400B+ class) is the cloud-capex demand signal. The read-through to the DLR/EQIX physical-layer thesis.

Issue 10 ships with the AVGO verdict on Forecast [2026-05-31-001], the PANW FCF-filter read, and the first Q2-positioning hints if any early 13F/13G amendments drop.

Methodology and Disclosures

Filings Intel Digest is a bona fide financial publication. Nothing here is personalised investment advice. The editor may hold positions in companies discussed; current positions and policy are at /about. All claims are sourced to publicly filed documents or identified press sources. Backtested or historical figures are direction-adjusted and calibrated against the signal lab; calibration is partial — treat any forward statement as a hypothesis, not a forecast.

Frozen as of close-of-trading Friday 29 May 2026, covering the week ended Saturday 30 May. The MRVL financials are drawn directly from the company’s Q1 FY27 8-K earnings release (filed 27 May, accession 0001835632-26-000014) and 10-Q (filed 28 May) — read from the primary filings, not the automated extraction pipeline, which returned zero structured signals for the window. Dell, Super Micro, HPE and Arista figures are from press coverage (Yahoo Finance, Barron’s, Investor’s Business Daily) and are directional pending each company’s primary filing. Positioning data is from the 13F convergence layer (Q1 2026 filings, eleven-fund universe), which was rebuilt and verified this week, including a fix to dual-class share aggregation and institutional value normalisation.

Watchlist now tracking 11 funds (9 Tiger-Cub-style growth + Druckenmiller + Berkshire).

Pipeline note: the automated signal-extraction pipeline returned 0 signals for the 23–30 May window — one new filing was ingested (the MRVL 8-K) but not run through LLM extraction. This issue’s structured analysis is sourced from a direct read of the MRVL 10-Q and the verified convergence layer. The editor should confirm whether extraction should be run manually on the MRVL filings to backfill the signal record.