Why Silicon Photonics Is Becoming Critical in the AI Era — And What GF’s Acquisition of AMF Really Means

GlobalFoundries’ (GF) announcement that it will acquire Singapore-based silicon photonics foundry Advanced Micro Foundry (AMF) has quickly drawn broad industry attention. In recent years, AI systems have expanded dramatically in scale, driving data centers to demand higher bandwidth, lower latency, and improved energy efficiency. Meanwhile, traditional electrical interconnects are increasingly constrained by physical limits at high frequencies. At the same time, high-speed optical modules and photonic integration technologies are moving rapidly toward mainstream deployment, positioning silicon photonics as a crucial foundation for future data center interconnects.

Against this backdrop, GF’s move to reinforce its silicon photonics manufacturing capability appears to be a strategically timed decision. It has triggered deeper industry discussions about next-generation interconnect architectures, component demand, and the long-term evolution of the electronics supply chain. This article analyzes the technical logic, industry drivers, and supply-chain implications behind the acquisition through a Q&A format, and incorporates commonly referenced high-speed component models to illustrate how these changes manifest at the product level.

Q1: Why has GF’s acquisition of AMF become such a widely discussed event? What exactly does it impact?

The acquisition stands out because it directly addresses the most critical pain point in today’s AI infrastructure: data exchange inside large GPU clusters has fallen far behind the growth of compute power. For example, high-end data-center switches like Broadcom’s Tomahawk 5 (51.2 Tbps) already operate at the limits of electrical signaling, while 800G optical modules such as QSFP-DD 800G DR8 are becoming standard elements in AI training fabrics. These trends underscore a simple fact: communication capacity—not compute—is now the system bottleneck, and electrical interconnects struggle to support higher speeds without unsustainable power consumption.

GF’s move signals to the market that silicon photonics is shifting from a research topic to an engineering-ready technology for mass deployment. This has significant implications for upstream foundries, downstream optical-module manufacturers, and the broader component ecosystem.

The company’s decision to anchor its silicon photonics build-out in Singapore further reflects the region’s growing role as a hub for photonics manufacturing and packaging, with potential consequences for regional supply-chain distribution.

Q2: AI compute keeps getting stronger, so why has the bottleneck shifted to “data movement”? Are electrical interconnects truly hitting their limits?

Training large-scale AI models requires hundreds or even thousands of GPUs to work collaboratively, exchanging vast amounts of data at every iteration. While GPU compute scales exponentially, interconnect bandwidth requirements are rising even faster. Under these conditions, electrical interconnects are approaching theoretical limits.

A clear example is the widespread use of 112G PAM4 SerDes in high-end networking devices. As their signaling rates increase, power consumption, heat dissipation and signal-integrity challenges escalate sharply. Even with advanced PCB materials such as LCP or PTFE laminates, high-frequency copper traces still suffer from attenuation, crosstalk and increasing design complexity—making further scaling economically and physically unsustainable.

At the same time, 800G optical modules have become the basic units of AI cluster connectivity. Their deployment is fundamentally shifting the energy profile of data centers, with communication power now approaching or surpassing the power used by compute nodes. Simply adding more GPUs can no longer improve system-level performance; interconnect efficiency has become the new ceiling.

These realities are pushing the industry to seek solutions that bypass the fundamental limitations of copper, and silicon photonics is currently the most mature and scalable alternative.

Q3: Why is silicon photonics considered the breakthrough technology? What exact problems does it solve?

The core value of silicon photonics lies in its ability to shift high-speed data transmission from electrical to optical signaling. Optical signals experience dramatically lower loss at high frequencies, reducing the need for equalization, re-timing, and other energy-hungry techniques that electrical interconnects rely on. This makes silicon photonics particularly attractive for next-generation systems that will rely on 800G and eventually 1.6T optical modules.

Just as importantly, silicon photonics is compatible with mainstream CMOS manufacturing. Key devices—such as Mach-Zehnder interferometer (MZI) modulators and photonic integrated circuits (PICs)—can be produced on 200mm or 300mm wafers, enabling scalable and cost-efficient production. This compatibility is already driving adoption: high-density PICs are being incorporated into early 800G CPO (Co-Packaged Optics) prototypes, allowing optical paths to be placed directly adjacent to switch ASICs or accelerators and thereby avoiding the electrical bottleneck entirely.

However, this shift also reshapes supply-chain requirements. Packaging must achieve tighter optical alignment; materials must maintain optical and thermal stability; and testing workflows must now cover optical loss, wavelength characteristics, and coupling efficiency in addition to electrical performance. These changes are fundamentally redefining the electronics component value chain.

Q4: Why did GF acquire AMF? The motivation goes well beyond adding manufacturing capacity.

GF’s strategic rationale is rooted in deeper architectural shifts underway in data centers. Although silicon photonics has been studied for over a decade, the challenge has always been manufacturability. AMF, after years of process development, now offers a commercially viable 200mm silicon-photonics platform, along with a proven library of optical devices—including MZI modulators, waveguides, and couplers—and a production-ready PDK.

By acquiring AMF, GF effectively gains a turnkey ecosystem for silicon-photonics commercialization. This enables GF to serve hyperscalers and AI-system developers pursuing CPO and board-level optical engines, particularly as modules like QSFP-DD 800G DR8 and OSFP-XD 1.6T enter data-center qualification cycles. As these interfaces scale, photonic components transition from optional enhancements to required capabilities.

Singapore also provides GF with a strong R&D and manufacturing base supported by local research institutions and industry infrastructure. The acquisition positions the region as a future photonics hub and may catalyze the formation of a broader optical-electronics supply-chain cluster across Southeast Asia.

Q5: What does the GF–AMF acquisition mean for the electronics component supply chain?

From a supply-chain perspective, this acquisition represents a long-term structural shift rather than a short-term demand fluctuation.

First, demand for high-speed optical modules, optical engines, and CPO-related components will become clearer and more predictable. As AI clusters adopt 800G DR8, 800G FR4, and potential 1.6T optical interfaces, the need for photonic chips, couplers, modulators and detectors will expand. These components sit upstream in the supply chain and require advanced process control, creating new challenges and opportunities for suppliers.

Second, packaging is evolving toward hybrid optical-electrical integration. Many optical-module vendors are exploring PIC-on-board architectures to reduce electrical drive losses. This shift increases requirements for thermally stable materials, low-expansion substrates, and precision optical-alignment adhesives.

Third, test and validation systems will be substantially reshaped. Silicon-photonics devices require not just electrical verification but also optical testing—including wavelength, loss, coupling, and thermal stability measurements. Multi-channel optical-electrical probe cards for CPO validation are already entering qualification flows, altering the traditional ATE-centric test model.

Finally, regional supply-chain dynamics will shift. With GF strengthening its Singapore base, photonic manufacturing and portions of advanced packaging are likely to concentrate further in Southeast Asia. This may influence where optical materials, module manufacturing, and high-speed test capabilities develop in the future.

Q6: Will the impact of this acquisition be visible in the short term, or is it driven by long-term trends?

In the short term, the industry should not expect immediate changes in pricing or availability. Silicon photonics is still progressing through its engineering-maturity cycle, and both module vendors and system providers need time to validate new interconnect architectures. GF and AMF must also align processes, IP and customer development. Therefore, near-term effects will appear mainly at the roadmap and ecosystem-preparation level.

However, over the medium to long term, the significance of the deal will become increasingly visible. As AI clusters scale, switch bandwidths approach the 100-Tbps tier, and 1.6T modules approach manufacturability, silicon photonics is poised to become a foundational technology for mass deployment. This will reshape the electronics supply chain: packaging methods will evolve, test workflows will diversify, materials will be upgraded and regional production patterns will shift.

In other words, this acquisition affects the next 3–5 years of technology and supply-chain evolution, rather than near-term market fluctuations.

Q7: Will silicon photonics completely replace electrical interconnects? What does the future path look like?

Silicon photonics will not fully replace electrical interconnects in the short term. Electrical signaling remains advantageous for short-reach, low-speed and cost-sensitive scenarios. However, in high-bandwidth, long-reach and energy-efficient applications, silicon photonics offers fundamental advantages that electrical interconnects cannot match.

The likely progression is a layered coexistence:

• Inter-rack optical fiber connections are already mature
• Rack-to-board and board-to-board optical links are accelerating
• Chip-to-chip optical interconnects will become a focal point of next-generation architectures
• Chiplet-level optical integration may eventually emerge, reshaping processor and system designs

GF’s acquisition of AMF positions the company at the manufacturing entry point for this long-term transition, rather than addressing only today’s optical-module market.

Conclusion: What does this acquisition truly change?

The GF–AMF deal signals a shift in how AI infrastructure will evolve: system performance will no longer be determined solely by GPUs, but by the combined optimization of compute × communication × energy efficiency. Silicon photonics is one of the few technologies capable of improving all three simultaneously.

For the component supply chain, this implies:

• Optical and opto-electronic components will move from peripheral roles to core system elements
• Packaging, materials and test workflows will undergo significant upgrades
• Regional production footprints will shift toward Asia as photonics manufacturing scales
• Key components such as QSFP-DD 800G DR8, 112G PAM4 SerDes, 800G CPO optical engines, and 6T optical moduleswill become essential indicators of industry progress

This is not a short-term trend but a foundational shift that will define how electronic systems and supply chains evolve over the next decade.