The Physical Supply Chain Behind 800G and 1.6T AI Networking

A 1.6T port is a chain of specialized components

An 800G or 1.6T link begins electrically. A switch ASIC sends extremely high-speed signals through its package substrate and system board toward a pluggable module such as OSFP. Those signals must survive connectors, traces, cages, and thermal constraints before reaching the electronics inside the transceiver.

Inside the module, an optical DSP performs signal processing and lane conversion. Broadcom's Taurus, for example, combines a 1.6T PAM4 DSP with an integrated laser driver and is designed to work with 400G electro-absorption modulated lasers and photodiodes.[1] Marvell is attacking the same layer with its Ara family, including 1.6T DSPs built around 200 Gbps electrical and optical lanes.[2]

The DSP is only one part of the bill of materials. The signal must then be converted into light. Depending on reach and architecture, that can require continuous-wave lasers, electro-absorption modulated lasers, VCSELs, silicon photonics, modulators, photodiodes, transimpedance amplifiers, fiber arrays, and associated packaging.

Coherent's 2026 demonstrations illustrate how broad this layer has become. Its 1.6T platforms included silicon photonics PICs, high-power indium phosphide continuous-wave lasers, 200G indium phosphide EMLs, and 200G gallium arsenide VCSELs. The company paired those optical technologies with DSPs from three different suppliers.[3]

That is the central supply-chain point: there is no single "1.6T chip." A 1.6T port is a system assembled from semiconductor, photonic, mechanical, thermal, and optical components that must all operate at increasingly difficult signal speeds.

The less obvious public exposures sit below the switch

Coherent is one of the clearest public examples. It participates not only in finished optical modules but also in the underlying lasers, photodiodes, silicon photonics, and compound-semiconductor components. Faster networking therefore expands the value of the optical component stack even if the switch ASIC remains the most visible piece of silicon.

The same logic applies to connector suppliers. TE Connectivity is already demonstrating 1.6T linear-receive optics alongside ultra-high-density connectors, fiber-array units, external-laser connectivity, blind-mate optical interfaces, and optical backplanes.[4] At 112G and 224G signaling speeds, the connector and cable stop being passive commodities in the investment sense. Signal integrity, insertion loss, thermal performance, and manufacturing tolerances become part of whether the system works.

Package substrates and module boards form another largely hidden layer. Advanced DSPs and switch ASICs still need high-density semiconductor packaging, while optical modules require increasingly complex electrical and photonic integration. Public companies such as Ibiden, Shinko Electric, Unimicron, Kinsus, and Nan Ya PCB participate in the broader high-performance substrate ecosystem, although public disclosures generally do not identify which substrate vendor sits inside a specific 1.6T transceiver. They are therefore category exposures, not verified supplier links to the modules discussed here.

The optical-module manufacturers themselves are also important. Broadcom's Taurus announcement specifically included Eoptolink, whose management described working with Taurus-based high-speed transceivers.[1] That places a Chinese optical-module supplier directly in the technology transition being driven by American switch and DSP companies.

This is where economic dependence can become very different from technical importance. A hyperscaler or switch vendor may source optics from several module suppliers, while a smaller optical manufacturer can have a large portion of its production tied to a limited number of high-volume AI programs. The component may represent only a small share of the customer's total infrastructure spending, yet the customer or program can matter enormously to the supplier.

That asymmetry is often more relevant to investors than the headline relationship itself.

1.6T shifts the bottleneck toward manufacturing

The transition from 800G to 1.6T effectively doubles the amount of information moving through a port, but it does not double the tolerance for loss, heat, errors, or manufacturing variation. Higher lane rates make optical alignment, laser performance, connector quality, thermal design, and packaging yield more demanding.

This means the second-order beneficiaries of AI networking do not have to design GPUs or Ethernet switches. They can manufacture a laser that survives higher modulation speeds, a photodiode that receives the signal, a connector that preserves signal integrity, a substrate that supports the DSP, or a module that integrates those components at acceptable yield.

It also creates a different risk profile. If 1.6T deployment grows faster than expected, demand can propagate simultaneously into DSPs, compound-semiconductor lasers, silicon photonics, optical packaging, connectors, and module assembly. If qualification schedules slip or yields disappoint, the same chain can delay deployment even when switch silicon is available.

Conclusion

The physical supply chain behind 800G and 1.6T AI networking is broader than the companies selling AI switches. Broadcom and Nvidia sit near the visible top of the stack, but the bandwidth increase pushes technical and economic importance into optical DSPs, lasers, photodiodes, silicon photonics, connectors, substrates, and transceiver manufacturing.

For investors, the more interesting question is therefore not only who sells the next AI switch. It is which smaller suppliers become harder to replace as every optical port becomes faster, hotter, denser, and more difficult to manufacture.

Sources

1. "Broadcom Delivers Industry's First 400G/lane Optical DSP for Next-Generation AI Networks," Broadcom Inc., March 11, 2026. Broadcom Investor Relations.

2. "Marvell Ushers In the 1.6T Era with Expanded Optical DSP Platform Portfolio, Redefining AI Data Center End-to-End Connectivity," Marvell Technology, March 12, 2026. Marvell Newsroom.

3. "Coherent Demonstrates Technologies for Next-Generation Pluggable Transceiver at OFC 2026," Coherent Corp., March 17, 2026. Coherent Newsroom.

4. "Advancing End-to-End Optical Infrastructure for Next-Generation AI Data Centers at OFC 2026," TE Connectivity, March 13, 2026. TE Connectivity News Center.

How to Cite This

For research inquiries or data access: press@altsets.com

Go Deeper