Marvell shrinking share of the RAN custom silicon market & acquisition of XConn Technologies for AI data center connectivity

Samsung and Nokia currently use Marvell’s OCTEON Fusion baseband processors and OCTEON Data Processing Units (DPUs) in their 5G Radio Access Network (RAN) equipment.

1. Samsung has a long-standing relationship with Marvell, primarily using the latter’s silicon for purpose-built (traditional) 5G RAN products. 

  • OCTEON Fusion Processors: Samsung uses these baseband processors in its 5G base stations, particularly for massive MIMO (Multiple-Input Multiple-Output) deployments that require significant compute power for complex beamforming algorithms.
  • OCTEON and OCTEON Fusion Families: Samsung has leveraged multiple generations of these processors for baseband and transport processing solutions.

2. Nokia has been a significant customer and relies heavily on Marvell’s silicon for its RAN portfolio, incorporating them into its custom ReefShark chipsets. 

  • Customized OCTEON Silicon: Nokia uses customized Marvell OCTEON silicon across key applications, including multi-RAT (Radio Access Technology) RAN and transport.
  • OCTEON Fusion Processors: These are used for baseband processing in Nokia’s 5G products.
  • OCTEON TX2 and OCTEON 10 Families: These infrastructure processors are used for demanding tasks like packet processing, security, and edge inferencing within Nokia’s 5G infrastructure.
  • OCTEON 10 Fusion: Nokia is working with the latest generation of this 5nm baseband platform, which supports use cases from radio units (RU) to distributed units (DU) for both traditional and Open RAN architectures. 

Nokia has been dependent on Marvell for much of its RAN silicon, moving away from previous reliance on Intel. However, the company is also exploring the use of Nvidia GPUs for future AI-RAN and 6G network equipment as part of a new collaboration. 

………………………………………………………………………………………………………………………………………………………………………………….

Marvell’s sales of RAN silicon are on the decline.  Let’s examine why that’s happening:

Light Reading reports that Marvell’s custom silicon for Samsung’s future 5G and 6G RAN equipment (5G base stations and small cells)  has started to look economically unviable. “Samsung and Marvell continue to collaborate on chips,” said the company in a statement emailed to Light Reading. Yet it also acknowledges efforts to find alternatives, including work on its own silicon.

“Samsung has been designing and manufacturing its own in-house modem chips for over 30 years, differentiating it from other RAN vendors.  Samsung has also been collaborating with key partners other than Marvell on modem SoC [system-on-chip] chipsets designed to be integrated in hardware basebands.”  For example, Samsung makes Virtualized RAN (vRAN) products for Verizon and other carriers using general-purpose processors from Intel.

Nokia, Marvell’s other big RAN silicon customer, is shifting to general purpose processors for its RAN equipment rather than Marvell’s custom RAN chips.  Soon after Nvidia’s $1B investment in Nokia, the latter disclosed plans to design 5G AI RAN, 5G Advanced and 6G network equipment that will run on the Nvidia’s GPUs.

“We can shift investment into software and ultimately deliver differentiation and value where it matters – and this is the shift from proprietary to general-purpose hardware,” said Nokia CEO Justin Hotard at Nokia’s capital markets da. Nokia is trying to reuse as much of the software developed for Marvell’s chips as it can but maintaining separate Marvell and Nvidia development tracks for its 5G Advanced and 6G products would drive up costs.

……………………………………………………………………………………………………………………………………………………………………………….

Meanwhile, the total global RAN market has been declining for years as network operators slash investment in network equipment and cut jobs.  According to Omdia (owned by Informa):

  • Global RAN equipment sales fell from $45 billion in 2022 to $40 billion in 2023 and just $35 billion in 2024.  Nokia’s mobile networks business group suffered an operating loss of €64 million (US$75 million) on sales of €5.3 billion ($6.2 billion) for the first nine months of 2025.
  • For its 2023 fiscal year (ending in January 2023), Marvell’s carrier division made almost $1.1 billion in revenues, more than 18% of total company sales. Two years later, annual revenues had slumped to just $338.2 million, less than 6% of turnover.
  • Marvell’s carrier sales have also recently improved in fiscal 2026, rising 88% year-over-year for the first nine months, to $436.3 billion. However, that’s still half as much as Marvell made during the first nine months of fiscal 2024, and interest in the RAN has seemingly evaporated.
  • Samsung’s  share of the shrinking RAN market has declined. Amid contraction of the entire addressable market, revenues generated by Samsung Networks fell from 5.39 trillion South Korean won ($3.74 billion) in 2022 to just KRW2.82 trillion ($1.95 billion) in 2024. For the first nine months of 2025, Samsung reported network sales of KRW2.1 trillion ($1.46 billion). But it has also lost market share, which dipped from 6.1% in 2023 to 4.8% in 2024, according to Omdia.
  • Ericsson has two development tracks – one for purpose-built RAN products based partly on its own custom RAN silicon and the other for an Intel-based virtual RAN. In contrast to Samsung, the purpose-built RAN silicon portfolio today accounts for nearly all of the company’s sales.
  • Ericsson’s senior managers increasingly talk about virtualization as a means of developing one set of software for multiple hardware platforms. The hope is that software originally designed for use with Intel’s processors  could be redeployed on CPUs from AMD or  licensees from ARM Ltd. with minimal coding changes. Such optionality combined with the narrowing of the performance gap between CPUs and purpose built RAN silicon would make it hard for Ericsson to justify investment in its own custom silicon.

…………………………………………………………………………………………………………………………………………………………………………….

Today, Marvell announced it will acquire XConn Technologies for $540 million to boost AI/data center connectivity.  In late 2025, the company announced the acquisition of Celestial AI for up to $5.5 billion to expand its optical interconnects for next-gen data centers, solidifying its position in infrastructure semiconductors.

Adding XConn’s PCIe and CXL switching technology (see illustrations below), fills gaps in Marvell’s silicon portfolio and enables the company to expand into higher-speed interconnects (like PCIe Gen 6).

XConn Technologies XC 50256 chip: 256 lanes with total 2,048GB/s switching capacity

…………………………………………………………………………………………………………………………………….

XC50256 CXL 2.0 Switch Chip

……………………………………………………………………………………………………………………………………………………………

As AI workloads scale, data center system design is evolving from single-rack deployments to larger, multi-rack configurations. These next-generation platforms increasingly require a high-bandwidth, ultra-low latency scale-up fabric such as UALink to efficiently connect large numbers of XPUs and enable more flexible resource sharing across the system.

UALink is a new open industry standard purpose-built for scale-up connectivity, enabling efficient, high-speed communication so multiple accelerators can operate together as a single, larger system. UALink builds on decades of PCIe ecosystem innovation and incorporates proven high-speed I/O techniques to meet the bandwidth, latency, and reach requirements of next-generation accelerated infrastructure.

Together, Marvell and XConn will bring together a significantly larger, integrated team to fully address the rapidly emerging opportunity in UALink switching as well as comprehensively support the growing list of customers and partners who want to work with Marvell in evolving their next generation AI platforms.

About Marvell:

To deliver the data infrastructure technology that connects the world, we’re building solutions on the most powerful foundation: our partnerships with our customers. Trusted by the world’s leading technology companies for over 30 years, we move, store, process and secure the world’s data with semiconductor solutions designed for our customers’ current needs and future ambitions. Through a process of deep collaboration and transparency, we’re ultimately changing the way tomorrow’s enterprise, cloud and carrier architectures transform—for the better.

About XConn Technologies:

XConn is the innovation leader in next-generation interconnect technology for high-performance computing and AI applications. The company is the industry’s first to deliver a hybrid switch supporting both CXL and PCIe on a single chip. Privately funded, XConn is setting the benchmark for data center interconnect with scalability, flexibility, and performance. For more information visit: https://www.xconn-tech.com

……………………………………………………………………………………………………………………………………………………………

References:

https://www.lightreading.com/5g/fragile-samsung-deal-with-marvell-shows-challenge-for-ran-chipmakers

https://investor.marvell.com/news-events/press-releases/detail/1004/marvell-to-acquire-xconn-technologies-expanding-leadership-in-ai-data-center-connectivity

RAN silicon rethink – from purpose built products & ASICs to general purpose processors or GPUs for vRAN & AI RAN

Dell’Oro: Analysis of the Nokia-NVIDIA-partnership on AI RAN

Intel FlexRAN™ gets boost from AT&T; faces competition from Marvel, Qualcomm, and EdgeQ for Open RAN silicon

Analysis: Nokia and Marvell partnership to develop 5G RAN silicon technology + other Nokia moves

Samsung and Marvell develop SoC for Massive MIMO and Advanced Radios

China gaining on U.S. in AI technology arms race- silicon, models and research

Omdia on resurgence of Huawei: #1 RAN vendor in 3 out of 5 regions; RAN market has bottomed

Leave a Reply

Your email address will not be published.

You may use these HTML tags and attributes: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <s> <strike> <strong>

*