Last week, Marvell announced another giant deal with the acquisition of Inphi. We had an opportunity to talk with members of Marvell’s executive team to garner insights on the acquisition, which we share in this blog.
We view this acquisition as strategic and complementary, rather than opportunistic, given Inphi’s strong positioning in electro-optics interconnect both inside data centers as well as connecting data centers. Inphi also has a strong and growing position in 5G backhaul, mid-haul, and front-haul. Its technology and addressable market complement Marvell’s storage, networking, processor, and security portfolio and accelerate its growth in Cloud and 5G infrastructure. Note that more than 70% of Inphi’s fiscal 2020 year-to-date revenue is derived from Cloud and 5G. Marvell’s executive team expects this acquisition to help expand its footprint with large Cloud Service Provider (SP) customers. Four additional networking Cloud customers with greater than $100 M in annual revenue will be created by the combined company.
While the pandemic may have slowed investments in some areas of the network, we have highlighted in our reports that spending on Cloud data center infrastructure as well as 5G networks will remain robust and may even accelerate because of the pandemic.
Our Data Center Capex five-year forecast projects spending on Cloud data center infrastructure to grow at 11% CAGR, reaching $140 B by 2024. Additionally, Dell’Oro Group’s 5-Year RAN Forecast shows that 5G NR investments are anticipated to advance five to ten fold over the next five years.
Nevertheless, although growth in Cloud SPs as well as 5G networks presents attractive opportunities for many suppliers, it also carries margin challenges, technology challenges, and the rising need for custom solutions:
It is widely known that Cloud SPs have a relentless need to lower the cost of deploying and operating their data centers. Incremental savings on each piece of purchased equipment translates into significant improvement in their return on investment (RoI) as well as other profitability metrics. This is because of the scale at which they operate data centers with hundreds of thousands of server installed base. Our interviews with players throughout the supply chain revealed how vigorously suppliers are struggling to win Cloud SPs business (viewed as strategic) without taking a margin hit. This low-margin business is driving small companies to get acquired and large companies to try to grow even larger as scale is the only way to remain competitive. There are tremendous synergies when you can sell more products to a single customer. There are also synergies when you can partner with the same manufacturers. Additionally, the development costs of some advanced—and, in some cases, custom—products required by Cloud SPs is so high that they can be offset only through scale. Marvell’s executive team expects both Marvell and Inphi to benefit from the larger scale in R&D—in particular, in process technology. Note that Marvell recently announced its 5 nm platform and started its research in 3 nm. This will help Inphi drive higher performance, lower power, and potentially more custom products.
The transition to a digital world with artificial intelligence (AI), mobility, and distributed computing at the edge has accelerated the explosion of traffic and data movement inside Cloud SP networks, across data centers and at the edge. This means that networks must become faster to satisfy the insatiable demand for bandwidth. Our data center switch five-year forecast report shows that about 30% of data center switch ports shipped by 2024 will be at 400 Gbps speeds and higher. However, as network speeds continue to increase, they will create challenges. For example, pluggable optics (currently the form factor of choice for network connectivity inside the data center) will hit density and power issues. When this occurs, the industry will be forced to adopt alternative technologies, such as co-packaged optics (CPO), with optics co-packed and integrated on the switch chip. We expect this trend to favor vertically integrated companies, at least in the early stages of adoption, due to the lack of a standardization level that would allow for a more diverse ecosystem. I mentioned this trend at the beginning of the year in my blog (Ethernet data center switch trends in 2020 and beyond) and predicted that CPO will drive numerous acquisitions, consolidations, and partnerships among switch chip vendors, switch system vendors, and optical transceiver vendors. Note that Marvell has a switch chip business, which may be bolstered by Inphi’s strong optics technology.
Rising need for custom solutions:
The data and compute intensity of modern AI and machine learning workloads is putting tremendous pressure on the performance of Cloud data centers. While computing demand continues to surge, CPU performance improvements are slowing down, as Moore’s law is reaching its limits. We expect this to drive new ways to design modern data centers to become giant compute engines with hundreds of thousands of compute nodes. This, in turn, will drive innovations to interconnect the different nodes without compromising latency and performance. As Cloud SPs will try to differentiate themselves through these architectural innovations, their need for custom—and sometimes proprietary—solutions will increase, potentially forcing their suppliers to own different pieces of technologies that they can integrate. Additionally, this differentiation will require a large scale to drive synergies, while developing custom-made solutions.
Marvell’s acquisition of Inphi was the chip sector’s second acquisition last week, following Advanced Micro Devices’ acquisition of Xilinx. The deal followed on the heels of an already active year of giant tech acquisitions with Nvidia’s acquisition of ARM and Analog Devices’ acquisition of Maxim Integrated. We expect more acquisitions to follow as the only way to survive in this highly competitive market is to join forces and get “stronger and better together.”