Optical Component Supply: The Critical Bottleneck in AI Infrastructure

Every data center fabric is, at its core, a chain of relay stations. The signal must propagate from source to destination through a series of optical interconnects, each imposing a latency floor and a bandwidth ceiling. For Alphabet, whose TPU-based AI accelerators demand ever-higher throughput and lower per-hop delay, the optical network is not peripheral—it is the central nervous system. The industry now stands at a inflection point: co-packaged optics (CPO), external laser sources (ELS), and photonic integration are moving from laboratory curiosities to commercial deployments. This report examines the shifting supply dynamics that will determine whether the relay chain holds firm or buckles under the load of AI-scale workloads.

The Architecture of Density: CPO, ELS, and the Next Relay

The fundamental constraint is propagation. As switch ASIC speeds climb, the distance between the electrical interface and the optical port becomes the dominant source of signal impairment. CPO collapses that distance by integrating optical engines directly onto the switch substrate, reducing the latency and power penalties of pluggable modules. Marvell’s demonstration of a 51.2 Tbps CPO switch ⁹ confirms that the technology can sustain line-rate performance. For the TPU ecosystem, Credo supplies active electrical cables and retimers ⁵, while Corning provides the active optical cables and fiber ⁵ that carry these dense signals between nodes. The shift is not speculative: Lumentum expects CPO/ELS revenue to become “meaningful” as early as the December quarter ³, and Coherent’s order books now extend into 2028 ⁶.

External laser sources represent an architectural choice of consequence. By separating the laser from the optical engine, ELS simplifies thermal management and enables higher port densities. But this design places the reliability burden on the laser supply itself—a single point of failure in the relay chain that requires mechanical rather than negotiated guarantees. The market is now contesting whether external lasers will be supplied by independent vendors or captured within internal supply chains ¹⁰. The answer will shape pricing power and technology lock-in for years.

The Weakest Tower: Component Shortages and the Collapse Margin

A relay chain is only as strong as its most constrained component. Today, that constraint is the continuous-wave (CW) laser. The device is in tight supply ³, forcing Lumentum to prioritize its use for internal transceiver production over external sales ³. The same scarcity afflicts laser diodes and electrical components ³, creating a cascading effect: module vendors who lack internal supply of critical building blocks face higher costs and delayed shipments ³. This is not a software bug—it is a physical shortage of fabricated devices, and its resolution depends on wafer capacity and process yield, not protocol negotiation.

The industry is responding by scaling the manufacturing towers. Six-inch indium phosphide (InP) production lines are cutting costs and improving yield ^1,6, offering a path to higher volume at lower cost per good photon. But the benefits accrue disproportionately to those with captive or contracted capacity. For Alphabet, the message is clear: the relay towers that produce these lasers must be secured, or the entire data center fabric risks operating below its engineered potential.

Locking the Towers: Long-Term Agreements as Structural Guarantees

In an era of component famine, long-term agreements (LTAs) function as architectural guarantees—contractual redundancy that ensures signal continuity. Coherent’s LTAs now span hyperscalers and system providers ⁶, often with upfront customer investments and supply commitments ⁶, and their durations stretch to the end of the decade ⁶. Lumentum has multi-hundred-million-dollar purchase orders for CPO/ELS delivery in the first half of 2027 ³ and is engaged with multiple customers on pluggable ELS modules ³. These agreements are the modern equivalent of fortified relay stations: they ensure that a tower will not go dark because a supplier reallocated capacity.

Yet such contracts also introduce rigidity. The longer the commitment, the greater the risk of locking in a technology path that may be superseded. The arrival of 3.2T-capable electroabsorption modulated lasers (EMLs) ⁶ and 1.6T coherent sampling at 2 nm ⁹ suggests that the performance ceiling is still rising. Alphabet’s infrastructure roadmap must align with these optical benchmarks, but the timing of LTA renewals and the pace of internal development (such as custom photonic integrated circuits) will determine whether the company can upgrade its relay chain without stranded investment.

Competing Semaphore Lines: Supplier Diversity and Risk

The optical component supply base is no longer a monopoly of Western firms. Chinese players like Zhongji Innolight and Eoptolink demonstrated optical circuit switching (OCS) hardware at OFC ³ and benefit from Veeco equipment orders ⁴. Their emergence mirrors the proliferation of semaphore lines in the 19th century: more operators can reduce cost and improve resilience, but they also introduce variability in signal quality and compliance. For a hyperscaler like Alphabet, this dualism creates strategic optionality—the ability to source from multiple vendors—but also the burden of testing and qualifying each new link in the chain.

The desirability of dual-sourcing is heightened by the risk that incumbent suppliers may become bottlenecks themselves. If the external laser market consolidates around a few providers, the pricing power shifts away from the data center operator. Alphabet’s scale grants it negotiating leverage, but the structural answer is architectural: ensure that any laser module can be replaced with a functional equivalent without redesigning the switch. That interchangeability reduces dependency and aligns with the principle that a relay system should tolerate the failure of any single tower.

Implications for Alphabet: Engineering the Next Chain

The optical interconnect landscape imposes three clear imperatives on Alphabet’s infrastructure strategy:

1. Supply chain resilience must be engineered, not assumed. The tightness in CW lasers and other building blocks indicates that component availability will be a gating factor for data center builds. Alphabet should treat laser supply contracts with the same rigor it applies to chip tape-outs, with multi-year visibility and strategic co-investment in InP wafer capacity.

2. Architectural flexibility should be designed into the physical layer. As CPO and ELS technologies evolve, the ability to swap between internal and external laser sources, or among vendor implementations, will prevent lock-in and allow Alphabet to ride the cost-reduction curve of 6-inch InP production.

3. Broader signal paths must be monitored. The presence of tangential claims—lidar deals for Ouster ^7,8, construction software acquisitions ¹¹, or healthcare imaging expansions ²—hints at Alphabet’s expanding footprint across AI-driven industries. Optical interconnects may soon be needed not only in data centers but in autonomous vehicles and edge devices. The same principles of low-latency, high-bandwidth propagation will apply, and the same supply chain vulnerabilities will recur.

Conclusion: The Towers Must Not Go Dark

We are witnessing the molecular reorganization of the optical relay network that underpins modern AI infrastructure. The shift to CPO and ELS is a genuine architectural advance—it reduces the number of conversion stages between the electrical and optical domains, thereby cutting latency and power just as the optical telegraph replaced the multi-hop mechanical relay. But the physical reality of laser shortages introduces a fragility that must be managed with the same deliberate planning that Alfred Vail and Samuel Morse applied to their line construction. Alphabet’s challenge is not merely to procure enough lasers; it is to build a relay system that can scale in density, withstand supply disruptions, and adapt to an evolving competitive landscape. The tools have changed. The principles have not.