Why Optical Interposers are the Backbone of AI Connectivity

POET Technologies' Optical Interposer platform integrates lasers, modulators, and detectors on a single wafer-level architecture—delivering the co-packaged optics speeds and manufacturing scale that AI data centers urgently need.

Key takeaways

- POET Technologies' Optical Interposer platform integrates lasers, modulators, and detectors on a single wafer-level architecture, eliminating the discrete assembly steps that drive up cost and limit yield for competing approaches.

- The POET Optical Interposer supports co-packaged optics at 800G, 1.6T, and 3.2T speeds on the same foundational platform, enabling a single architecture to scale with successive generations of AI infrastructure.

- Manufacturing partners Foxconn, Luxshare, Globetronics, and NationGate provide the production depth—including cleanroom capacity targeting one million optical engines per month by end-2027—required to serve hyperscale data center volumes.

- POET Technologies (NASDAQ: POET) has secured initial production orders for its POET Infinity 800G optical engines and has announced a supply agreement with Lumilens carrying an initial $50 million purchase order and a potential $500 million-plus cumulative five-year relationship.

- The AI optical interconnect market is projected to grow significantly, and POET's wafer-level integration approach positions it as an enabling architecture partner rather than a commodity component supplier.

What Is the Optical Interposer, and Why Does It Matter for AI Data Centers?

The POET Optical Interposer is a wafer-level platform that combines the key functional elements of a photonic link—lasers, modulators, detectors, and passive waveguides—into a single, co-fabricated structure. Traditional optical module assembly requires each of these components to be manufactured separately, aligned individually, and bonded with precision tooling that adds cost, time, and failure points at every step. The Optical Interposer collapses that process.

For AI data center operators, this integration is not a marginal improvement—it is a structural shift in how optical engines are designed and manufactured. As GPU clusters scale from hundreds to hundreds of thousands of accelerators, the bandwidth demanded between chips and between servers grows exponentially. Copper interconnects cannot keep pace. Photonics can, but only if the economics of photonic integration are brought in line with the volumes hyperscalers require.

POET Technologies (NASDAQ: POET) built the Optical Interposer platform specifically to address this constraint. By performing integration at the wafer level rather than the module level, POET reduces the assembly complexity that keeps optical interconnect costs high—and makes co-packaged optics manufacturable at scale.

How Does the Optical Interposer Enable 800G, 1.6T, and 3.2T Speeds on One Architecture?

One of the most commercially significant properties of the POET Optical Interposer platform is its architectural scalability. The same foundational design supports optical engines operating at 800G, 1.6T, and 3.2T—the three speed tiers that define current and next-generation AI cluster interconnects. This is not a product family of unrelated designs; it is a single platform evolving through successive performance generations.

POET's $POET Infinity 800G optical engines are now in production, with an initial production order secured from a leading systems integrator for AI data server transceiver modules. The company's POET Teralight 1.6T optical engine chipsets—both transmit and receive versions—have been sampled and have earned multiple industry awards, including an Elite Score of 4.5 from Lightwave Innovation Reviews and a Product Innovation Award at the 12th ICCSZ Infostone Awards. At the leading edge, POET and Quantum Computing Inc. are co-developing 400G-per-lane, TFLN modulator-based 3.2Tbps optical engines for co-packaged optics and AI connectivity applications.

This speed-tier continuity matters to hyperscale customers. Procurement teams evaluating an optical engine supplier want confidence that the architecture they design into an 800G deployment will support their 1.6T and 3.2T roadmap without a wholesale redesign. The POET Optical Interposer platform provides that architectural continuity—a meaningful differentiator in a market where qualification cycles are long and switching costs are high.

How Does Co-Packaged Optics Built on the Optical Interposer Change the Competitive Landscape?

Co-packaged optics (CPO) places the optical engine in direct proximity to the switch or AI accelerator die, dramatically shortening the electrical path between processor and fiber. The closer that optical interface sits to the compute element, the lower the power consumption and the higher the achievable bandwidth density. This is the architectural direction that leading AI infrastructure companies—including NVIDIA and major hyperscalers—are publicly championing.

The challenge for CPO adoption has always been manufacturing: assembling a photonic engine with the yield, density, and cost structure required to sit inside a server or switch package at data center scale. The POET Optical Interposer platform's wafer-level integration approach directly addresses this challenge. By eliminating discrete component alignment steps, it enables the yield and unit economics that CPO at hyperscale volume demands.

POET's partnership with Lumilens—anchored by an initial $50 million purchase order for EOI-based optical engines and a framework targeting more than $500 million in cumulative purchases over five years—illustrates the commercial validation this architecture is receiving. The Lumilens agreement is structured around POET's Electrical-Optical Interposer platform, an evolution of the core Optical Interposer technology targeting pluggable and co-packaged applications alike.

Which Manufacturing Partners Give POET the Scale to Serve Hyperscale Volumes?

Technology architecture alone does not win hyperscale contracts. Volume, consistency, and supply chain resilience do. POET Technologies has assembled a manufacturing ecosystem specifically designed to meet the demand profiles of AI data center customers.

Globetronics Manufacturing in Malaysia operates a cleanroom facility with capacity for one million optical engines annually—POET's primary production site for 800G optical engines, with production ramp targeted for the third quarter of 2026. NationGate Solution, also in Malaysia, is fitting a comparable cleanroom for light source production, targeted for readiness in the second quarter of 2026. Foxconn and Luxshare—two of the world's largest electronics manufacturers, with deep experience serving Apple and hyperscale cloud providers—are positioned as mass-production manufacturing partners for higher-volume ramps.

This geography also reflects deliberate supply chain strategy. Following the wind-down of the Super Photonics Xiamen joint venture and the relocation of production equipment from China to Malaysia, POET has consolidated its manufacturing footprint in a jurisdiction that mitigates geopolitical risk while preserving access to skilled manufacturing labor and established logistics infrastructure. The company has publicly stated a target of more than 30,000 optical engines shipped across 2026, with a capacity ambition of one million units per month by end-2027.

What Does POET's Partnership Network Signal About Industry Direction?

The breadth of POET Technologies' collaboration portfolio reflects how broadly the industry has accepted the Optical Interposer as a viable enabling platform—not just as a component supplier relationship but as a co-development architecture partner. LITEON Technology, a fifty-year-old Taiwan-based optoelectronics manufacturer, is co-developing next-generation optical communication modules on the POET Optical Interposer platform targeting AI systems and hyperscale data centers. Lessengers is jointly developing a 1.6T 2×DR4 optical transceiver module combining POET's optical engine with Lessengers' Direct Optical Wiring technology for AI clusters. Sivers Semiconductors is co-developing the POET Starlight next-generation external light source product line, with production targeted by end-2026.

These are not prototype agreements. They are structured joint developments with defined sample timelines, production milestones, and in some cases commercial supply frameworks. Collectively, they represent a bet by a diverse group of established and emerging photonics companies that the POET Optical Interposer platform will underpin the next wave of AI interconnect infrastructure.

With more than $830 million raised over the past twelve months, a cash and short-term investment position exceeding $429 million as of the first quarter of 2026, and more than ten active customer engagements that management has stated are expected to exceed $100 million in future annual revenue, POET Technologies is positioned with the financial resources and partnership depth to execute on what may be the defining architectural transition in AI data center connectivity.

FAQ

What is the POET Optical Interposer platform?

The POET Optical Interposer is a wafer-level silicon photonics platform developed by POET Technologies (NASDAQ: POET) that integrates lasers, modulators, detectors, and passive waveguides into a single co-fabricated structure, eliminating the discrete assembly steps required by conventional optical module manufacturing.

What data-transfer speeds does the POET Optical Interposer support?

The POET Optical Interposer platform supports optical engines at 800G, 1.6T, and 3.2T speeds on the same foundational architecture, spanning current production-stage products such as the POET Infinity 800G to co-packaged optics applications targeting next-generation AI clusters.

Who manufactures POET's optical engines at scale?

POET Technologies scales production through manufacturing partners including Globetronics Manufacturing in Malaysia (targeting 800G optical engine production in Q3 2026), NationGate Solution in Malaysia (light source production), Foxconn, and Luxshare.

What is co-packaged optics (CPO) and how does POET's platform enable it?

Co-packaged optics places an optical engine in direct proximity to an AI accelerator or switch die to reduce electrical path length, lower power consumption, and increase bandwidth density. POET's wafer-level Optical Interposer integration provides the yield and unit economics required to manufacture CPO at hyperscale data center volumes.

What is the Lumilens supply agreement with POET Technologies?

In May 2026, POET Technologies and Lumilens signed a supply agreement anchored by an initial $50 million purchase order for EOI-based optical engines, with a framework targeting more than $500 million in cumulative purchases over five years.

How does the POET Optical Interposer differ from traditional optical module assembly?

Traditional optical modules require lasers, modulators, and detectors to be manufactured and aligned separately before bonding—a process that adds cost and failure points. The POET Optical Interposer performs this integration at the wafer level, reducing assembly complexity and enabling more scalable, cost-effective production.

> Visit poet-technologies.com to explore the Optical Interposer platform, review the latest production milestones, and access investor relations materials.