I’m thrilled to sit down with Vladislav Zaimov, a seasoned telecommunications specialist whose deep expertise in enterprise telecommunications and risk management of vulnerable networks offers invaluable insights. With years of experience navigating the complexities of radio access network (RAN) technology, Vladislav is uniquely positioned to shed light on the evolving landscape of RAN silicon, the shift to general-purpose processors, and the emerging opportunities for new players in this critical sector. Today, we’ll explore the decline of custom silicon, the challenges faced by major vendors, and the potential for innovation as the industry adapts to changing market dynamics.
Can you walk us through the current trends with custom RAN silicon and why it appears to be losing its foothold in the market?
Certainly, Andrew. Custom RAN silicon, which has long been the backbone of 5G baseband processing, is facing a slow decline due to a combination of economic and technological factors. The RAN market itself has shrunk significantly, dropping from $45 billion in revenue in 2022 to about $35 billion in 2024. This downturn makes it harder to justify the high costs of developing specialized chips. On top of that, the industry is seeing a growing preference for general-purpose processors—think of the kind used in PCs and data centers—because they offer more flexibility and lower costs over time. It’s a pragmatic shift; why invest heavily in niche hardware when broader platforms can do the job almost as well, especially as performance gaps narrow?
What’s driving the move toward general-purpose processors, and how do you see this reshaping RAN technology in the coming years?
The push toward general-purpose processors is largely about scalability and cost efficiency. These chips, unlike custom ASICs, aren’t tied to specific hardware-software pairings, which means vendors can deploy RAN software across multiple platforms with less hassle. Right now, virtual RAN—built on these processors—accounts for just 10% of the baseband subsector, but projections suggest it could exceed 20% by 2028. This growth is fueled by advancements like Intel’s Granite Rapids family, which has closed the performance gap with custom silicon, and by the industry’s broader desire for flexibility. In the long run, this shift could redefine how networks are built, making them more adaptable but also raising questions about performance in demanding areas like Layer 1 processing.
How are major RAN vendors like Ericsson, Nokia, and Samsung grappling with their reliance on specific silicon providers?
These vendors are in a tricky spot due to their dependencies. Ericsson and Samsung lean heavily on Intel for their RAN silicon, which poses risks if Intel faces supply issues or lags in innovation—something we’ve seen hints of with their recent financial struggles. Nokia, on the other hand, is tied to Marvell Technology, which limits their options compared to competitors who might pivot more easily to other providers. This lack of diversity in silicon supply chains can stifle competition and innovation, and it’s a concern for telcos who rely on these vendors for critical network components. Open RAN was supposed to bring more choice, but so far, it hasn’t shaken up the silicon market as hoped.
Speaking of alternatives, how do you see AMD positioning itself as a contender in the RAN space, particularly for vendors like Ericsson and Samsung?
AMD is emerging as a very compelling alternative, especially for Ericsson and Samsung, because it uses the same x86 architecture as Intel. This compatibility means software written for Intel chips can be adapted to AMD with minimal rework, unlike shifting to something like Arm-based processors. Financially, AMD is also in a strong position, with a 36% sales growth to $9.2 billion in their latest quarter, while Intel’s growth has been sluggish. However, there are hurdles—Ericsson, for instance, has concerns about AMD’s hardware accelerators for forward error correction, which aren’t integrated like Intel’s but come on a separate card. That’s seen as an added cost and complexity, though it’s not an insurmountable issue.
Samsung seems to be ahead of the curve in moving away from custom silicon. Can you elaborate on their strategy and what it might mean for their future offerings?
Samsung has indeed taken a bold leap toward virtual RAN, outpacing others like Ericsson. They’ve built entirely virtual RAN systems for operators like Verizon in the US and are expanding in Europe with Vodafone. Their speed comes from a clear focus on cost and performance—virtual RAN, in their experience, often outperforms traditional setups while being cheaper. Interestingly, Samsung is also experimenting with AMD processors in a way that doesn’t rely on hardware accelerators for forward error correction, banking on the high core count of AMD chips to handle it via software. If successful, this could lead to a commercial offering soon, potentially setting a new standard for flexibility in their product lineup.
Looking ahead, what’s your forecast for the future of RAN technology as custom silicon continues to wane and new players like AMD step in?
I think we’re on the cusp of a transformative period for RAN technology. As custom silicon fades, general-purpose processors will dominate, driven by their adaptability and the industry’s need to cut costs in a shrinking market. AMD has a real shot at becoming a major player, especially if they can address concerns around hardware accelerators and continue leveraging their financial momentum. However, the transition won’t be seamless—vendors will need to balance performance demands with the benefits of virtualization, and telcos will push for more diversity in silicon options to avoid over-reliance on a few providers. By 2028, I expect virtual RAN to be a significant chunk of the market, possibly even exceeding current projections if innovations keep pace. It’s an exciting, if challenging, road ahead.
