MRVL: Growth Lags NVDA; Rich Valuation Facing Multiple Compression?

Marvell (MRVL.O) released FY27 Q2 results (quarter ended Jul 2026) after hours on Aug 28 Beijing time.

1. Full-year outlook: FY27 revenue raised to $12.0bn (vs. prior guide $11.5bn), +45% YoY. FY28 revenue raised to $18.0bn (vs. prior $16.5bn), +50% YoY, with Data Center up 60%+ (vs. NVIDIA's 70%+ next-year outlook) and custom ASIC revenue still expected to more than double.

Beyond near-term prints, investors care most about the growth trajectory. Management provided top-line outlooks for the next two fiscal years during the post-print call.

Ahead of the print, the sell side broadly expected $Marvell Tech(MRVL.US) FY27/FY28 revenue of ~$11.8bn and $17.0–17.5bn. The new guides are only slightly above those estimates. Versus yesterday's major beat-and-raise from NVIDIA, Marvell's outlook looks notably underwhelming.

The recent Google agreement stoked hopes for custom ASIC wins (incl. TPU share gains). However, management did not lift custom ASIC guidance this time (no update for this fiscal year, FY28 still 'more than double'), which may read as a 'pay-to-play' framework rather than revenue-backed commitments.

2. Data Center: revenue was $2.17bn, +19% QoQ, driven mainly by interconnect products, with Data Center at 79% mix.

Within Data Center, interconnect remains the core driver, which is consensus. Relative to interconnect, the Street is more focused on custom ASIC performance, especially post the Google deal.

Today, Marvell's custom ASIC business is concentrated on Amazon's Trainium family. That business has underperformed given competition from Alchip, lagging Amazon's capex growth trajectory.

Shares spiked 10%+ when the Google deal was announced, largely on expected TPU-related upside (vs. Broadcom). But management did not raise ASIC guidance later; it did not reiterate this year's 20%+ target from last quarter, and kept next year's 'doubling' call. That leaves a question mark on whether Google can deliver incremental ASIC/TPU orders.

3. Operating metrics: revenue was $2.74bn, +13% QoQ, in line with consensus (~$2.75bn). The $0.3bn sequential uplift came entirely from Data Center.

GPM was 53.1%, up 100bps QoQ. GAAP gross margin is affected by acquisition-related amortization and does not fully reflect underlying operations. On an adjusted basis, we estimate GPM at 58.3%, flat QoQ. If lower-margin custom ASIC scales faster, GPM could face pressure.

4. Next-quarter guide: revenue of $3.15bn, a touch above Street (~$3.10bn). Implying Q4 revenue around $3.7bn, both quarters growing ~15–20% QoQ, led by interconnect demand. GAAP GPM guided at 52.9%–53.9%. With multi-year guides now provided, the quarterly guide matters less.

For the full call memo, see 'Marvell (Trans): No ASIC guide hike; Data Center growth 60%+ next year'.

Dolphin Research view: 'Pay-to-play' tie-up, outlook underwhelms

The print broadly met expectations, with growth almost entirely from Data Center interconnect. Adjusted GPM was 58.3%, flat QoQ, after stripping amortization and related impacts.

From this fiscal year, Marvell condensed disclosure from five segments to two (Data Center; Carrier & Other). Data Center was the main growth engine this quarter, +19% QoQ, driven by interconnect demand.

With full-year guidance in hand, the next-quarter guide is de-emphasized. We infer the next two quarters at roughly $3.15bn and $3.7bn, both up ~15–20% QoQ, again led by interconnect.

More importantly, management provided two-year revenue targets. FY27/FY28 are guided to $12.0bn/$18.0bn (vs. prior $11.5bn/$16.5bn), while the Street was already at ~$11.8bn/$17.0–17.5bn pre-print. Against NVIDIA's 'blowout' guide yesterday, Marvell's looks soft.

Dolphin Research sees two main reasons for the after-hours selloff: (i) despite the Google tie-up, Marvell did not raise custom ASIC guidance, making the agreement appear more 'pay-to-play'; (ii) Data Center growth for FY28 is 60%+, while NVIDIA, at a much larger base, guided 70%+ for next year.

Key focus areas beyond the headline numbers:

1) ASIC

Marvell recently announced its Google collaboration: it plans to issue warrants for 58.97mn shares to Google, granted on Aug 18, 2026, with a strike of $206.58.

Two parts: (i) an immediate grant of 1.36mn shares, vesting 0.34mn at 3/6/9/12 months; (ii) 57.61mn performance-based shares (240 tranches), starting Aug 1, 2026, with one tranche (~0.24mn) vesting for each $500mn of eligible revenue.

'Eligible revenue' covers Google's custom ASIC chips, including AI inference accelerators, storage controllers, NICs, memory interface controllers, and near-memory compute.

The structure shows Marvell negotiating from a weaker position. The goal is clear: secure Google's custom ASIC orders, while offering warrants upfront as a gesture, broadly mirroring the prior arrangement with Amazon.

2) Networking and interconnect

Acquisitions of XConn (PCIe/CXL switching), Celestial AI (photonic interconnect), and Polariton (electro-optic modulators) give Marvell a complete AI networking stack across Scale-Out, Scale-Up, and Scale-Across, offering high-speed interconnect from in-rack to inter-DC.

In today's Scale-Out optical landscape, pluggable optics remain dominant. NVIDIA, Google and others mainly use pluggables, and Marvell's PAM4 DSP is relatively leading in this market. Management previously said its 1.6T solutions have begun volume production and will ramp quickly in FY27, a key interconnect growth driver.

3) CXL memory expansion/pooling (Structera): AI inference demands massive KV cache. HBM is expensive and DRAM capacity is limited, so NAND/SCM-based expansion is attractive.

Marvell's CXL lineup includes Structera A, Structera X, and Structera S, serving as near-storage accelerators, memory expansion controllers, and memory pooling/switching, respectively.

With a current market cap of ~$211.1bn, Marvell trades at roughly ~40x FY28 adj. net income, assuming revenue CAGR ~53%, adj. GPM 57.5%, and adj. tax rate 12.5%.

Some major brokers now anchor valuation on FY29, reflecting a focus on long-term growth. Given uncertainty further out, Dolphin Research references FY28 (roughly CY27).

Versus NVIDIA (14–15x PE) and Broadcom (19x PE), Marvell screens expensive, embedding expectations for AI share gains and strong interconnect growth. Yet its FY28 Data Center growth of 60%+ trails NVIDIA's 70%+ next-year guide, which is unlikely to satisfy the market.

Part of the recent share rally reflected the Google deal. Judging by both the Google and Amazon agreements, Marvell appears to be negotiating from a weaker hand. The unchanged ASIC guide makes the deal look more like a 'gift' than a revenue-backed contract.

Net-net, the quarter was fine, and the after-hours selloff stems from an underwhelming outlook. With investors focused on forward growth, recent developments such as open-source model momentum and Anthropic ARR slope moderation could amplify volatility.

The AI chip field has few players, and Marvell is a rare US-based ASIC partner to multiple hyperscalers. Broadcom, buoyed by Google's TPU, reached a ~$2tn market cap at peak, while Marvell sits near $200bn. If Marvell captures 10–20% of TPU share, it could add several billion dollars of annual revenue (vs. current custom ASIC at ~ $2bn), which underpins the bull case.

On the unchanged ASIC guide, management added it will elaborate with scenarios at the Oct 6 Analyst Day. Watch for specifics then.

The full-year outlook is not great, but Marvell has 'paid up' with both Amazon and Google, keeping optionality for future orders. Near term, the stock could correct given a rich ~40x PE, yet '50%+ growth over the next two years + visible growth drivers' should support a premium multiple (30x+). For those betting on potential share wins, pullbacks could offer better entry.

Dolphin Research on Marvell (MRVL.O) results and details:

Note: 'Adjusted' operating metrics include SBC, but mainly strip amortization and similar items to better reflect underlying performance.

<End>

Dolphin Research's prior pieces on Marvell (MRVL.O):

Hot topics:

Aug 20, 2026: Marvell: 'Pay-to-play' tie-up with Google — can it pry share from Broadcom?

Earnings:

May 28, 2026 call memo: Marvell (Trans): FY28 $16.5bn guide could still move higher

May 28, 2026 note: Marvell: In the AI inference era, is interconnect more valuable than compute?

Mar 6, 2026 call memo: Marvell (Trans): Raising revenue again; a second XPU to ramp in FY28

Mar 6, 2026 note: Marvell: ASIC card miss, winning back AI via interconnect

Dec 3, 2025 call memo: Marvell (Trans): AWS becomes Celestial AI's lead customer

Dec 3, 2025 note: Marvell: Filling product gaps via M&A — a faster path to 'NVIDIA-lite'?

Aug 29, 2025 call memo: Marvell (Trans): Still no direct response to Alchip's challenge

Aug 29, 2025 note: Marvell: Cloud capex was generous — why didn't ASIC benefit?

May 30, 2025 call memo: Marvell (Trans): AI to become half of total revenue

May 30, 2025 note: Marvell: AI sequential growth stalled — where is the next trump card?

Mar 6, 2025 call memo: Marvell (Trans): Data Center growth collapsed

Mar 6, 2025 note: Marvell: Pouring cold water on AI again — ASIC flashing warning signs

Deep dives:

Jan 14, 2025: ASIC battle royale — can Marvell beat Broadcom?

Jan 2, 2025: Marvell vs. the trillion-dollar Broadcom — can ASIC ignite a comeback?

Risk disclosure and disclaimer: Dolphin Research disclaimer and general disclosures