Samsung Just Entered Silicon Photonics — Here's What That Tells Us About Where The Real Money Is In Optical Interconnects

Summary

• Samsung Foundry officially entered the silicon photonics market at OFC 2026, joining TSMC and GlobalFoundries in a race to supply the optical interconnect infrastructure powering AI datacenters.

• When the world's largest foundries all converge on the same technology layer, history tells us exactly what happens next: components commoditize, and value concentrates at the integration layer above.

• The integration layer in silicon photonics — co-packaged optics and optical engine platforms — remains largely unsolved by the foundry giants, with Samsung's own CPO roadmap extending to 2029.

• The companies solving that integration problem ahead of the foundry timelines deserve a closer look.

When Samsung Moves, Pay Attention

At the Optical Fiber Communication Conference (OFC) 2026 in Los Angeles, Samsung Electronics (OTCPK: SSNLF) made it official: its foundry division is entering the silicon photonics market.

OFC 2026 Just Confirmed the Optical Supercycle — And the Ent

The announcement was not a concept pitch or a research preview. Samsung arrived at OFC with a completed process design kit (PDK), a validated 224 Gbps per lane modulator benchmarked by Belgium's imec, and a phased commercial roadmap. That is a company that has done its homework and made a decision.

The roadmap runs from photonic integrated circuits (PICs) in 2026 through optical engines in 2027 and 2028, arriving at full co-packaged optics (CPO) by 2029. Samsung also highlighted something TSMC cannot offer: vertical integration of HBM memory, logic foundry, advanced packaging, and now silicon photonics — all under one roof.

This is not a niche announcement. This is one of the three most powerful semiconductor ecosystems on the planet formally committing to optical interconnects as a strategic priority.

The Foundry Convergence Is The Signal

Samsung is not alone in this move. TSMC has already disclosed a co-packaged photonics roadmap developed in collaboration with NVIDIA (NVDA), including the Quantum-X Photonics InfiniBand switch for AI cluster interconnects — unveiled last year — and the Spectrum-X Photonics Ethernet switch targeted for the second half of this year. Intel and STMicroelectronics have operated PIC platforms for years. GlobalFoundries offers related services.

Why POET Technologies Isn't a Silicon Photonics Stock — And

The picture that emerges is not one company making a speculative bet. It is an entire tier of global foundry infrastructure converging on the same technology at the same time.

That kind of convergence has a consistent historical pattern. When foundry capital floods a component market — logic, DRAM, NAND, advanced packaging — two things follow. Manufacturing capacity expands fast. And component costs compress as scale increases and suppliers compete for the same customers.

In silicon photonics, those customers are the optical module companies — Coherent (COHR), Lumentum (LITE), and a growing field of fabless PIC designers building custom architectures for hyperscale datacenters. Samsung, TSMC, and GlobalFoundries all want to supply PICs to this ecosystem. That competition is good for the ecosystem. Better PICs, at better prices, flowing faster into the supply chain.

But here is what foundry competition does not solve.

The Problem That Manufacturing Scale Cannot Fix

PICs are inputs. They are the chips that integrate modulators, waveguides, and photodiodes onto silicon — the raw material of optical interconnects. What the foundry layer produces is necessary but not sufficient.

Tech Risks Rise After Anthropic Ban, Samsung Delay - Mar 3

The harder problem — the one that determines whether AI datacenters can actually run at the bandwidth densities and power envelopes the next generation of accelerators demands — is integration. Specifically, how do you co-package lasers, photodetectors, and PIC chiplets into a single manufacturable optical engine without the expensive, yield-limited assembly steps that make current approaches difficult to scale?

That is the co-packaged optics problem. It is the reason TSMC and NVIDIA are collaborating on photonics switches. It is the reason Samsung's roadmap shows CPO as a 2029 target — not because Samsung lacks resources, but because the integration challenge is genuinely hard.

AI infrastructure demand is not waiting until 2029. The power consumption and bandwidth density constraints that CPO addresses are already present in today's AI clusters, already being priced into hyperscaler capex decisions. The window between now and when the foundry giants reach CPO maturity is not empty. It is the most interesting stretch of the optical interconnect trade.

Where Value Concentrates When Components Commoditize

Semiconductor history offers a reliable guide here. When the manufacturing layer gets competitive, value moves up the stack. The companies that win are not the ones making the components — they are the ones with the platform-level integration that makes those components work together.

Nvidia’s $4B Photonics Bet Sets the Tone — Sector Surge Meet

In optical interconnects, that platform is the optical engine — the solution that takes a PIC from whichever foundry produced it and integrates it into a high-performance, low-power module that can sit inside next-generation AI hardware. Companies solving that problem with a manufacturable, scalable, wafer-level approach hold a structurally differentiated position.

Samsung's entry into PIC manufacturing does not dilute that position. It strengthens it. More foundry competition means better, cheaper PIC inputs flowing into the ecosystem. Companies operating at the integration layer benefit from that input cost pressure — while remaining insulated from the foundry competition itself, because what they build cannot be replicated by adding another PIC production line.

The 2029 Gap

Samsung's roadmap is worth sitting with for a moment. A company with Samsung's resources, vertical integration advantages, and global manufacturing footprint has mapped out a disciplined path to CPO — and it lands in 2029.

That timeline reflects the genuine complexity of the integration problem. The foundry giants are building toward it from the bottom up — starting with PIC manufacturing, layering in optical engine packaging, and working toward full co-packaged optics over multiple product generations. It is a logical, well-resourced approach. It is also a slow one.

The more interesting architectural story is happening at the other end of the stack. Companies like POET Technologies (NASDAQ: POET) have been working the integration problem from the platform level down — building wafer-scale interposer architectures that co-package lasers, photodetectors, and PIC chiplets into unified optical engines without the discrete assembly steps that make current approaches expensive and difficult to scale. The Optical Interposer platform is foundry-agnostic by design — it does not care whether the underlying PIC came from Samsung, TSMC, or GlobalFoundries. Every new foundry entrant expands the ecosystem it can draw from rather than threatening it.

That positioning is worth understanding in the context of what Samsung just announced. As PIC supply scales up and component costs compress — the natural outcome of foundry-level competition — the economics improve for anyone building at the integration layer on top of those inputs. POET's optical engine development, its advancing customer engagements, and the broader commercial momentum building around co-packaged optics all sit squarely in that context. The company has been doing the unglamorous work of platform development while the macro narrative was still forming around it. That narrative is now forming fast.

When the foundry timelines eventually converge with the integration-first platforms, the companies that built customer relationships and design wins during the gap will be in a position that no new roadmap announcement can easily replicate. Samsung just defined that gap as three years wide. That is a long time in AI infrastructure.

The Takeaway

Samsung's silicon photonics announcement is one more data point confirming what the smart money in AI infrastructure has been tracking for two years: optical interconnects are not a secondary theme. They are the critical path for AI scaling beyond current architectures.

The foundry layer is getting built out. PIC supply is going to expand. Component costs are going to come down. And as that happens, the integration layer — the companies solving the co-packaged optics problem with manufacturable platforms — will attract increasing attention from investors who follow the value up the stack.

Samsung just told the market that CPO is a 2029 event via the foundry-native approach. There are companies building toward it on a different timeline, from a different angle.

That gap is worth paying attention to.

This article is for informational purposes only and does not constitute investment advice. The author may hold positions in securities mentioned. All investors should conduct their own due diligence before making investment decisions.