May 20, 2026
By Teylor Bremekamp , PMP, RCDD, Fiber Network Design Engineer, Sumitomo Electric Lightwave
In this post: why major transceiver manufacturers have aligned on Base-16, what the 800G-DR8 and 1.6T roadmaps mean for greenfield and legacy operators, how AI-era connectors are converging on 16-fiber formats, and what co-packaged optics means for Base-16 infrastructure.
Base-16 supports every major speed generation from 800G forward
The data center industry is in the middle of a generational infrastructure shift, with today’s architecture decisions carrying networks through multiple generations of speed upgrades. Based on the research I’ve done and my conversations with industry experts, the outlook is bright for Base-16.
The numbers are compelling, the transceiver roadmap is aligned, and the long-term economics are solid.
The Industry Is Aligned on Base-16
Major transceiver manufacturers are backing Base-16 for high-speed Ethernet applications, and the revenue data bears that out. 800G switch revenue surged 220% in the first half of 2025, according to IDC, and 800G is now the baseline for AI backends. The dominant 800G-DR8 standard, with 500m reach, uses the 16-fiber MPO as its interface: 16 fibers, eight fibers transmitting, eight receiving, all active – the current production standard for the world's most widely deployed high-speed transceivers. Across each of the 8 lanes, 100G is transmitted and received over each pair via 100G PAM4 modulation. Additionally, while 800G is also dominant on traditional front-end networks, legacy and existing data centers commonly use 2xDR4 optics for the front-end network. While still 16 fibers, or 8 lanes of 100G, traditionally this was deployed via two 12-fiber MPO connectors. Looking forward at new data center construction, the 16-fiber MPO connector remains dominant for both front-end and back-end DR8 deployments.
As network speeds scale to 1.6T, the same lane configuration applies at 200 Gb/s per lane, maintaining the Base-16 fiber count. Early 1.6T deployments are set to begin later this year, with LightCounting projecting 1.6T reaching 70-80% of new data center deployments by 2030. The chips driving this transition tell the same story: Broadcom's Tomahawk 6, NVIDIA's Spectrum-X, and Marvell and Cisco’s forthcoming 1.6T ASICs are all architected around the port configurations that Base-16 serves directly.
Greenfields and Base-16 Go Hand in Hand
Hyperscaler CapEx is expected to reach $600-750 billion in 2026, according to IEEE ComSoc, up 36-50% from 2025, with more than 75% directed at AI infrastructure. McKinsey projects $7 trillion in cumulative data center investment through 2030. Based on fiber's share of that buildout, the fiber optics TAM could reach $30 billion annually by 2030. AI clusters require 2-4x the fiber density of traditional data center architectures, and that number grows with each GPU generation. Multi-core fiber is on the cusp of adding options for density-constrained applications, but for the port-level architecture decision facing greenfield builders today, Base-16 fiber is the answer the roadmap supports.
Greenfield facilities designed end-to-end around Base-16 are next-gen ready from day one, optimized for density, footprint, and the migration path ahead – without recabling. Legacy Base-8 and Base-12 network operators have their own migration path that I may explore in a future blog.
AI-Era Fiber Connectors Are Base-16
Every major speed generation from 800G forward is likely to have a Base-16 connector format. Currently, the 16-fiber MPO connector (MPO-16) is the interface standard for 800G-DR8, recognized under IEC 61754-7-3. The density gains are concrete: MPO-16 delivers a 33% improvement in fiber count over MPO-12 in the same footprint. The 32-fiber MMC connector goes further, fitting three times the connections of a standard MPO in the same panel space. It positions data centers for the density demands of 3.2T down the road.
Multifiber very small form factor (VSFF) connectors – including the MMC and SN-MT – are increasingly specified for hyperscale deployments as per-rack fiber densities continue rising with each GPU generation.
The Roadmap Favors Base-16
Base-16 and VSFF connectivity are the infrastructure framework that will support 800G, 1.6T, and beyond without expanding the physical footprint. The industry has aligned on the upcoming 1.6T deployments via DR8 200G PAM4 per lane optics. A network operator who builds on Base-16 today for 800G deployments achieves 1.6T by swapping out transceivers on their own timeline – no recabling required.
A little further out, 3.2T networks will likely be built on a foundation of Base-16 infrastructure with VSFF connectors. For 3.2T networks, the industry remains split on the architecture solution. To utilize a Base-16 DR8-style optic for 3.2T, 400G per lane speeds must be achieved. Technical challenges remain today with the jump from 200G to 400G utilizing PAM4 modulation. Technical engineering solutions still need to be developed to solve for chromatic dispersion, high frequency noise, and other signal degradation issues.
Industry leaders in optics (such as Marvell, Broadcom, and Coherent) have successfully proven that 400G per lane speeds are viable up to 500m reach by using next-generation lasers (D-EMLs). However, this has largely been demonstrated in R&D-type settings. It’s likely that 400G DR8 lane speeds won’t be widely available on day 1 of 3.2T network rollouts. Thus, it’s fairly likely that initial deployments of 3.2T will utilize 16 lanes, or 32 fibers, of 200G PAM4 per lane. This will be achieved through 2xDR8 200G optics, or two 16-fiber connectors with OSFP-XD module footprints. In the event this is widely adopted, data center operations will still need to use a Base-16 fiber. However, this doubles the fiber count within the data center when migrating from 1.6T. If 400G lane speeds are not ready when 3.2T switches hit the market, data center operators will upgrade their fiber infrastructure in order to deploy the cutting-edge switches.
A similar story will likely unfold for the migration to 6.4T switches. The jump from 400G lane speeds to 800G lane speeds is an exponentially harder challenge. As 6.4T switches get rolled out, it’s increasingly likely that 400G lane speeds will be the highest available option for DR8 optics. Thus, for 6.4T, it is highly probable that these will be deployed via 2xDR8 400G optics, each with 16 fibers for a total of 32 fibers.
Beyond 500m, the 200G/400G PAM4 per lane modulation solution breaks down, and other solutions are emerging. Similar to FR8/LR8 used today, for reach beyond 500m cable lengths, FR8 solutions are being developed for campus-level connectivity utilizing Coherent-Lite modulation solutions such as DP-16QAM.
Impacts of Co-Packaged Optics and Near Packaged Optics
While adoptions of CPO/NPO will vary across data center operators over the next few years, for short-reach links like AI back-end networks, the Base-16 fiber deployments will remain largely unchanged. While pluggable transceiver form factors will remain Base-16 fiber, they will be available in MPO and VSFF-style connectors. For CPO/NPO, since the larger faceplate of pluggable optics is no longer part of the switch, these next-gen CPO/NPO switches will largely favor VSFF connectors due to their density advantages.
Legacy Base-8 and Base-12 fiber architectures deployed in existing data centers will require fiber upgrades to reach the 1.6T and 3.2T speeds and beyond. Operators will achieve this through connector changes, fiber count increases, and technicians equipped with 12-to-16 fiber ribbon fusion splicers.
Base-16 is the most streamlined path through a genuinely complex roadmap.
Contact us if you would like to talk through your network planning. For more Next Generation Thinking™, follow Sumitomo Electric Lightwave on LinkedIn and YouTube.
 |
About the author: Teylor Bremekamp, PMP, RCDD, is a fiber network design engineer with Sumitomo Electric Lightwave.
|