Marvell Technology (MRVL): AI Server Rack Exposure

Where it fits in an AI server rack

Develops custom silicon for AI and data-center customers; PAM and coherent DSPs, laser drivers, TIAs, silicon photonics and co-packaged optics; AEC DSPs used in active electrical cables for scale-up and scale-out AI; PCIe retimers; Teralynx switches up to 51.2 Tbps; and SSD controllers (FY2026 10-K). Data center was 79% of Q2 FY2027 revenue ($2.17B of $2.74B). Closed Celestial AI ($3.5B, Feb 2, 2026) and XConn ($469M, Feb 10, 2026), the latter adding PCIe and CXL switching.

What the evidence establishes

AI-rack evidence

Disclosed: Marvell’s FY2026 10-K describes its custom AI silicon, electro-optics DSPs, AEC DSPs for AI scale-up and scale-out, PCIe retimers and Teralynx switches.

An explicit AI-server or AI-rack product, design-in, supply role, program, reference design or partnership, such as a part named for rack-scale AI systems or a product sold for AI data centers. For a rack builder, it means the company’s own AI servers or racks. This does not establish revenue, market share, volumes or future sourcing.

Component role	Custom AI silicon, optical DSPs and AEC DSPs
Evidence	Disclosed: Marvell’s FY2026 10-K describes its custom AI silicon, electro-optics DSPs, AEC DSPs for AI scale-up and scale-out, PCIe retimers and Teralynx switches.
How we know	Disclosed The company’s own filing, release, annual report, product page or investor record describes the AI-server or AI-rack role. Where customers are unnamed, the listing says so.
Primary source	Marvell Technology — supporting disclosure Company or official disclosure; vendor claims are attributed to their source.
Checked	2026-09-24. This is the research review date, not the source publication or quote time.

Keep in perspective

Marvell does not name its custom-silicon customers (one customer was 14% of FY2026 revenue), and custom revenue is not reported separately. Management expects the custom business to accelerate from H2 FY2027.

Specialist business exposure. This describes the breadth or role of the business; it is not an investment rating. An AI-server product, design-in or partnership does not by itself establish material AI-rack revenue, market share, or future returns.

Connections to the assembly

These links explain the technology relationship. The model is illustrative and does not represent an actual manufacturer’s bill of materials.

GPU compute dies

Two reticle-sized logic dies joined into one GPU, built on a leading-edge foundry process. They run the matrix math of training and inference. Custom accelerators designed by or for cloud companies are the main alternative.

Scope of the company links: The compute dies. NVIDIA designs Blackwell and Blackwell Ultra (208 billion transistors on two reticle-limited dies, custom TSMC 4NP) and AMD designs the Instinct MI350 and MI455X; NVIDIA discloses that TSMC fabricates Blackwell. Custom accelerators are the alternative: Broadcom co-develops OpenAI-designed accelerators and racks (its other XPU customers are unnamed), Marvell develops custom AI silicon for unnamed customers, and the design-service houses Alchip (an N3 accelerator for a North American cloud provider, from May 2026) and GUC (AI accelerator design support) disclose programs without naming customers. Intel’s Gaudi 3 is a merchant AI accelerator with no documented rack-scale deployment. Hyperscaler chip designers (Google TPU, AWS Trainium) are not mapped because they do not sell the chips as components. Samsung is named as an NVIDIA wafer foundry but not for Blackwell GPU dies, so it is not mapped here.

E1.S NVMe SSDs

Flash drives in the slim EDSFF E1.S form factor: NAND flash, a controller and DRAM that hold the operating system, checkpoints and cached data. Reference trays carry four; some vendors fit up to eight in internal bays.

Scope of the company links: E1.S NVMe drives sit in each compute tray (NVIDIA lists four per GB200 tray). The rack-specific qualifications found are Sandisk’s DC SN861 E1.S (certified for GB200 NVL72 as Western Digital, Oct 14, 2024) and Micron’s 9550 and 7450 (on NVIDIA’s GB200 NVL72 recommended vendor list, per Micron). Kioxia (NX1 E1.S for GPU servers) and Samsung (PM1763 for AI infrastructure) position drives for AI servers, SK hynix reports expanding eSSD sales for AI servers, and its subsidiary Solidigm, which is not separately listed, shows a cold-plate-cooled SSD. Silicon Motion, Phison and Marvell sell SSD controllers with no named rack program. A vendor list shows eligibility, not volumes.

Switch ASICs

Scale-up switch chips that connect every GPU in the rack to every other GPU at full bandwidth. Other designs use Ethernet switch chips for the same job.

Scope of the company links: Scale-up and fabric switch chips. NVIDIA’s NVLink Switch chips sit in the 1RU switch trays of GB200/GB300 NVL72 (NVIDIA OCP contribution, Oct 15, 2024), and its Spectrum-6 Ethernet ASIC runs 102.4 Tb/s. Broadcom sells Ethernet switching silicon to hyperscalers and to makers of AI servers, switches and racks (FY2025 10-K); Cisco’s Silicon One G300 (102.4 Tb/s) powers its own systems; Astera Labs’ Scorpio X-Series is a scale-up fabric switch for up to 80 accelerators; Marvell’s 10-K lists Teralynx Ethernet switches of up to 51.2 Tb/s. Celestica designs AMD Helios’s scale-up switch trays on other companies’ silicon and is listed under the top-of-rack switch, not here.

PCIe retimers & switches

Chips that recover and re-send high-speed PCIe or CXL signals over long board traces and cables, or fan one set of lanes out to many devices.

Scope of the company links: Astera Labs sells Aries retimers and Scorpio P-Series PCIe 6 switches, which an NVIDIA executive says integrate with Blackwell-based MGX (Mar 18, 2025). Marvell’s Alaska P PCIe 6 retimers were adopted by Supermicro, TE Connectivity and TeraHop (Dec 9, 2025). Credo’s Toucan Gen6 x16 retimer was PCI-SIG listed on Aug 31, 2026; Broadcom’s PCIe interconnect chips serve AI and non-AI uses (10-K); Microchip’s Switchtec Gen 6 switches are sampling; Montage calls itself one of the two major PCIe 5.0 retimer suppliers and says retimers are core components of AI servers (2025 results). No source says which retimer or switch a given NVL72 tray uses.

Active electrical cables

Copper cables with a retimer or DSP chip in each plug that cleans up the signal, extending copper’s reach to several metres at a fraction of the power of optics.

Scope of the company links: Copper cables with signal chips in the plugs, mostly for NIC-to-switch and switch-to-switch links. Credo sells complete ZeroFlap AECs to hyperscalers; Marvell’s Alaska A 1.6T AEC DSP names Amphenol, Molex and TE as cable partners (Jun 27, 2024); Astera Labs sells Aries and Taurus smart cable modules for AECs; BizLink and Luxshare sell retimed active copper cables (Luxshare showed a 1.6T OSFP AEC at OCP 2025); Microchip sells the META-DX2C 800G AEC retimer. Semtech’s CopperEdge redrivers power linear active copper cables (ACCs), such as Amphenol’s 1.6T OSFP ACC, a lower-power alternative to retimed AECs. No source ties a specific cable to NVL72. Montage launched PCIe 6.x and CXL 3.x AEC solutions in January 2026.

Lasers & photonics

Electro-absorption modulated or continuous-wave lasers, and silicon-photonics chips, that turn the signal into light inside the transceiver.

Scope of the company links: InP lasers (EML, CW, VCSEL) and silicon-photonics chips inside modules and co-packaged engines. Lumentum and Coherent make EMLs and CW/high-power lasers and hold NVIDIA purchase commitments for lasers (Mar 2, 2026); Mitsubishi Electric estimates it led data-center EML chips in FY25 and plans over 3x EML capacity by FY30; Sumitomo Electric makes EMLs and CW lasers for AI data centers; Broadcom runs in-house InP/GaAs laser fabs and sells a pluggable laser source for its CPO switches. TSMC provides the silicon photonics in NVIDIA’s CPO switches, Tower runs silicon-photonics foundry platforms for 400G–1.6T modules, TFC makes external-laser-source modules, Furukawa sells CW-DFB chips for silicon photonics, Marvell includes silicon photonics in its interconnect portfolio and MACOM sells lasers and photodetectors. AXT supplies InP substrates upstream. AOI makes laser chips only for its own modules and is mapped under transceivers. TrendForce (reported, context only) puts combined EML and CW-DFB capacity near 50.7 million units a month in 2026.

Optical DSPs & drivers

PAM4 digital signal processors, laser drivers and amplifiers that encode, equalize and retime the signal at each end of an optical link.

Scope of the company links: PAM4 DSPs retime and equalize each lane; drivers and TIAs amplify the electrical signal on each side of the laser and photodiode. Marvell (PAM DSPs, laser drivers, TIAs and LPO chipsets for AI data centers, FY2026 10-K), Credo (Bluebird, a 3 nm 200G-per-lane DSP for 1.6T modules, plus 100G-per-lane families, FY2026 10-K) and MACOM (TIAs and modulator drivers for 800G–3.2T modules, FY2025 10-K) are mapped, and Semtech’s TIA and modulator driver were demonstrated in NVIDIA’s 1.6T DR8 OSFP transceiver at OFC 2026. Broadcom’s optical DSP line could not be read from a primary source in our review, so Broadcom is mapped under lasers only. Linear-drive (LPO/LRO) modules drop some DSP functions, and NVIDIA says its co-packaged switches need no DSP retimers.

Related research

Companies with overlapping sector exposure. Their products and evidence may differ; this is not a list of equivalent investments.

Suggest a correction · Research and market-data disclosures

Marvell Technology