Semtech has joined the eXtra-dense Pluggable Optics Multi-Source Agreement (XPO MSA), adding its 200G-per-lane modulator drivers and transimpedance amplifiers to an industry effort focused on higher-density optical interconnects for AI data centers. The move comes as AI clusters push networking toward higher bandwidth, tighter power budgets and denser optical designs.
The next bottleneck in AI infrastructure may not be compute capacity but how efficiently data can move between increasingly dense clusters of accelerators. Semtech is addressing that challenge by joining the eXtra-dense Pluggable Optics Multi-Source Agreement (XPO MSA), an industry initiative developing a new optical transceiver form factor for high-bandwidth AI networks.
As an XPO technology partner, Semtech is contributing a portfolio of 200G-per-lane modulator driver and transimpedance amplifier (TIA) integrated circuits intended for XPO transceiver development and eventual volume deployment.
The timing reflects a rapid escalation in AI networking requirements. IDC says worldwide Ethernet switch revenue reached $18.9 billion in the second quarter of 2026, up 43.4% year over year. Data-center Ethernet revenue grew even faster, reaching $12.3 billion, a 64.5% increase driven by continued AI training and inference infrastructure investment. Meanwhile, 800GbE accounted for 41.2% of data-center Ethernet revenue.
Higher-speed switches create a corresponding requirement for faster optical links. As AI clusters expand, network designers need more bandwidth per rack and per switch while limiting power consumption and physical footprint. XPO approaches the problem by increasing the density of pluggable optical interfaces rather than simply adding more conventional modules.
Semtech’s contribution is focused on the semiconductor layer inside those optical systems. Its XPO portfolio includes quad and octal TIAs, along with quad and octal Mach-Zehnder modulator (MZM) drivers. The devices are available in different channel-pitch configurations, including 750-millimeter, 625-millimeter, 500-millimeter and 375-millimeter designs, according to the company’s announcement.
That range is intended to give optical-module and photonic-integrated-circuit designers flexibility when balancing density, signal integrity and thermal constraints.
Semtech’s GN1834L and GN1834DL quad TIAs, for example, target a 750-millimeter channel pitch, while the GN1838L and GN1838DL octal TIAs use a 500-millimeter pitch. The GN42T380 provides an octal TIA at 375 millimeters. On the driver side, the TN1877 and TN1887 support quad and octal configurations, while the GN42M380 provides an octal MZM driver at 375 millimeters.
The company’s portfolio also includes on-chip equalization and telemetry features. Those capabilities are intended to compensate for different electrical and optical channels, tune links and provide additional visibility during development and operation.
The larger trend is a move toward linear optical architectures and higher-speed electrical interfaces. Semtech previously introduced 224Gbps-per-lane TIA and MZM driver products designed for LPO, LRO, XPO, near-packaged optics and co-packaged optics at 800G, 1.6T and 3.2T.
XPO sits within that broader transition. Instead of treating the optical module as an isolated component, the ecosystem is increasingly optimizing the electrical interface, photonics and packaging together to reduce power and improve bandwidth density.
The commercial opportunity is expanding alongside AI infrastructure spending. IDC estimates global AI infrastructure spending will reach $487 billion in 2026, up approximately 53% from 2025, and projects the market to exceed $1 trillion by 2029. Cignal AI’s current optical-components research likewise tracks accelerating shipments of 800G and emerging 1.6T datacom optics, illustrating the speed at which the optical market is moving toward higher-capacity systems.
For companies building AI fabrics, bandwidth density is increasingly as important as raw port speed. NVIDIA and other AI infrastructure suppliers are deploying increasingly large accelerator clusters, while hyperscalers are evaluating optical architectures that can deliver higher throughput without proportionally increasing power and rack space.
That puts semiconductor suppliers such as Semtech at an important point in the stack. The company’s CDR, TIA and modulator-driver technologies do not define the AI workload itself, but they influence how efficiently the underlying network can move data between compute systems.
The XPO MSA could also help address interoperability. Multi-source agreements establish common specifications that allow multiple suppliers to build compatible components, reducing dependence on a single proprietary optical architecture. For operators deploying AI infrastructure at scale, that can create more flexibility in sourcing modules and components.
Semtech’s membership therefore signals a broader industry shift: AI networking is moving beyond the race toward faster switches and into a parallel race for denser, lower-power optical connectivity. As 200G-per-lane interfaces become part of that roadmap, XPO is positioned as one potential form factor for connecting the next generation of AI compute fabrics.
Market Landscape
AI infrastructure is driving an unusually rapid networking upgrade cycle. IDC reports that data-center Ethernet revenue grew 64.5% year over year in Q2 2026, while 800GbE already represented 41.2% of data-center Ethernet revenue.
Cignal AI’s 2026 optical-components research tracks 800G growth alongside the beginning of volume shipments of 1.6TbE optics. This creates demand for optical ICs capable of supporting higher lane rates, compact packaging and lower power consumption.
The competitive landscape extends across silicon photonics, optical modules, networking silicon and accelerator fabrics, with companies including NVIDIA, Broadcom and major optical-component suppliers investing heavily in next-generation AI connectivity.
Top Insights
- Semtech has joined the XPO MSA with 200G-per-lane TIA and modulator-driver silicon for higher-density AI optical interconnects.
- XPO targets bandwidth density and power efficiency as AI clusters move toward increasingly demanding networking architectures.
- Semtech’s portfolio offers multiple TIA and MZM-driver configurations designed to accommodate different transceiver and photonic-integrated-circuit designs.
- IDC data shows data-center Ethernet spending accelerating rapidly as hyperscalers expand infrastructure for AI training and inference.
- XPO forms part of a broader industry transition toward 800G, 1.6T and eventually higher-speed optical connectivity for AI fabrics.
Power Tomorrow’s Intelligence — Build It with TechEdgeAI
