We have published three different numbers for the size of AI networking — nearly $20B for a single year, more than $300B a year before this decade closes, and above $1T annual in the early 2030s — and we stand behind all three. Headlines now attach a trillion-dollar figure to AI networking, and buyers who only see that number next to our 2025 figure can conclude the market is being double-counted. It is not. The numbers sit on one published curve, read at different years and different scopes, and every step between them is already in our record. We think the gap between those two figures is where most AI networking sizing mistakes happen, so it is worth walking the curve.
The single-year ruler comes first. In June 2024, we published that data center AI networking will surge to nearly $20B in 2025, led by Ethernet, InfiniBand, and 800G optical transceivers across the switch, NIC, and interconnect layers attributed to AI deployments. That number sits on a ramp we have tracked in public since the beginning: the AI networking market reached $2B in 2022, with InfiniBand carrying 75% of the revenue, grew to over $3B in 2023, and reached over $15B in 2024 on its way to over $25B by 2028. The composition shifted underneath the totals: Ethernet AI networking revenue surged over 150% Y/Y in 2024 with bandwidth up over 200%, and we forecast 2025 scale-out networking to exceed $8B with frontend networking above $5B.
The decade ruler is the same curve, read later. In early 2025 we sized AI networking infrastructure above $50B a year as the buildout broadened beyond hyperscalers, and in March 2026 our Widening Chasm analysis put AI networking approaching $200B+ annually by decade’s end. This month, after Marvell’s investor day laid out a $400B TAM by 2030 backed by nearly $3T of AI capex, we published the steepest step yet: the systems opportunity exceeds $300B by the end of the decade, and with the incorporation of CPO we see the market for AI networking exceeding $1T in the early part of the 2030s. Our manufacturing-focused analysis landed the same view: AI networking is on a path to exceed $300B this decade and approach $1T early next decade as scale-up, scale-out, and scale-across fabrics all expand. A $1T run-rate is arithmetic at the far end of this ramp, not a separate claim.
Two rulers apply, and buyers who mix them will size anything incorrectly. Total data center networking — AI and non-AI — stands at slightly over $50B in 2026, up from $40B a year earlier, and we forecast $200B by 2032 at a CAGR over 20%, with the AI-attributed slice the fastest-growing portion. Total-market figures, AI-attributed figures, and layer-specific figures are three different instruments, and we publish all of them, quarterly, by segment and vendor.
The segment build-up is what makes the far-end arithmetic credible, and every layer is already public. We forecast AI scale-out Ethernet revenue to surge to over $100B by 2030, scale-up switching to exceed $30B by 2030 with optics adding $10B+, alongside NVLink at $25B+, Ethernet scale-up at $8B+, and PCIe/UALink at $3B+ by 2030. After this year’s rack-scale systems reset the architecture, our scale-up view steepened to passing $100B by 2030 — a revision worth noting, because it shows the curve moving up in public, not in retrospect. Deeper in the stack, RDMA switching has grown from a $1B market to one that will exceed $18B in 2028, with RDMA networking overall exceeding $22B by 2028. Sum the layers, add the optics, and the totals we publish follow.
Here is the rule we would offer any buyer reading a trillion-dollar headline: ask which year the figure describes, which layers are in scope, and whether it is one year or several. Our published record answers all three for our own numbers — nearly $20B in 2025, more than $300B by decade’s end, above $1T annual in the early 2030s. We believe the firms that size the market with the right instrument for the right decision will allocate capital better than those that quote whichever number is larger. We track this market quarterly and publish both rulers, by segment and vendor, in our AI networking research.
650 Group researches several of the technologies we covered in this blog, including Data Center, Co-Location & Cloud and Enterprise, and more.