How is AI expansion driving demand for networking and servers? Three growth signals from Goldman Sachs’ notes
650 Group is optimistic about connections within clusters, links between data centers and CPU demand from AI agents. But server price increases currently stem mainly from cost pass-through, while equipment vendors also face intensifying competition.
Original source：Goldman SachsFirst published on Unhidea：Original source date：
Key takeaway
Goldman Sachs’ latest expert discussion shows that 650 Group expects AI to expand demand for both network connectivity and traditional servers. Whether market growth translates into vendor gains depends on product differentiation, customer workloads and the competitive landscape.
What will sustain data center networking growth after 2030?
In an expert discussion hosted by Goldman Sachs, 650 Group forecast a 29% compound annual growth rate for the data center networking market from 2030 to 2035, with the largest contribution coming from scale-up, or tightly coupled connections within AI clusters. This scope includes scale-out, expansion across data centers, non-AI networking and third-party optical modules, but excludes co-packaged optics (CPO).
650 Group believes HPE and Celestica, as early entrants into Ethernet scale-up switches for AMD Helios racks, could benefit from their computing equipment experience. But as the technology matures and solutions become further disaggregated, more participants will enter and their first-mover positions will face competition.
Growth is not confined to scale-up: 650 Group expects demand for front-end networks and scale-out Ethernet switches to continue growing at a high double-digit percentage rate.
Why will connections between data centers become another growth driver?
Constraints on power, land and campus design mean AI factories need to be distributed across multiple buildings. When GPU clusters span physically separate data centers, connecting the clusters itself creates new equipment demand.
650 Group expects the AI data center interconnect market to grow from about 500 million dollars in 2025 to 37 billion dollars in 2030, a compound annual growth rate of 144%. Over the same period, the non-AI market is forecast to grow from 9 billion dollars to 21 billion dollars, at a compound annual growth rate of 19%. The two types of demand use broadly similar equipment and suppliers, giving existing vendors a point of entry to capture incremental AI demand.
Why was the traditional server forecast raised, and what roles do higher prices and incremental demand play?
650 Group raised its previous forecast for the traditional server market by an average of about 52% per year over 2026–2030, driven by both prices and new demand. Current price growth mainly reflects the pass-through of higher memory costs; server core counts have also risen from 20–30 per machine to more than 200, increasing configuration density.
Looking ahead, it believes enterprises will accept higher prices in exchange for greater density and performance; at the same time, AI agents need to interact with the application layer, creating additional demand for CPU servers. 650 Group expects server unit numbers to grow, with core counts growing faster.
Customer workloads determine the strength of demand: inference providers have more pronounced CPU needs, while newer cloud providers focused on training or GPU rental have lower demand. 650 Group expects the overall medium-term shares of branded and white-box servers in the latter market to remain stable, although customers with stronger technical capabilities may gradually shift toward white-box systems.
Will open networking standards erode the value of branded equipment vendors?
Despite open network infrastructure solutions such as Meta IcePack, 650 Group believes leading branded vendors can still create value for hyperscale and newer cloud customers: experience with complex deployments, hardware capabilities and differentiated chips form the competitive foundation of companies such as Cisco and HPE. The case for their benefiting rests on these specific capabilities, rather than network market expansion alone.
Why is it difficult to extrapolate the near-term tailwinds for enterprise campus networks?
Recent demand comes from additional budgets for AI security threats, Wi-Fi 7 upgrades, and replacement of equipment from before 2020 after support ends. Wi-Fi 7 requires multi-gigabit, higher-power switches; the Internet of Things and industrial equipment increase demand for ports, offsetting slowing growth in PC and phone connections.
For 2027 and beyond, 650 Group is more cautious: demographic trends depress demand from primary and secondary school customers; Huawei is expanding its European share with low-priced products, putting pressure on Cisco and HPE’s selling prices; and Ubiquiti is moving upmarket with an expanded product portfolio. Growth in replacement demand and pressure on vendor pricing may occur at the same time.
