Indium Phosphide Shortage: Why Optical Module Bottlenecks Reach the PCB
On 17 August 2026, supply chain reporting quoted the chief executive of Lumentum describing the current indium phosphide supply-demand imbalance as more severe than that seen in DRAM and NAND memory, with some order schedules extending to 2028. The supporting figures are stark: global demand for indium phosphide substrates in 2026 is estimated at 2.6 to 3.0 million wafers in two-inch equivalent terms, against effective capacity of roughly 750,000 wafers. That is a shortfall of more than seventy percent.
The consumption numbers explain the gap. A single 800G optical module requires approximately four to eight indium phosphide based laser chips, and a 1.6T module consumes nearly three times that quantity. As optical interconnect becomes the standard way to move data between accelerators and racks, the demand for these devices scales faster than module count alone would suggest.
The detail that matters for board manufacturing is the next sentence in that reporting: alongside the laser chips, peripheral optical module materials including printed circuit boards and capacitors are also entering a tighter supply situation.
When the Bottleneck Leaves the Chip
The indium phosphide shortage is fundamentally a demand story about AI computing, but its structure is instructive. Supply constraints are no longer concentrated in the highest-value components such as lasers and digital signal processors.
As optical modules move from 800G to 1.6T, the data density, component count, and power consumption inside each module rise together. Items that previously represented a modest share of the bill of materials, including the PCB, capacitors, connectors, and thermal materials, can become delivery constraints of their own, because their technical specifications have escalated while their effective capacity has not.
This changes how the optical module supply chain competes. In previous cycles, the priority was to secure the most expensive and scarcest chips. Now module makers must manage materials, boards, and other critical components simultaneously, because the constraint can appear anywhere in the bill of materials. The reporting that one module manufacturer raised its prepayments from roughly 134 million yuan to 1.488 billion yuan to lock in capacity illustrates the shift from conventional purchasing toward capacity reservation. When demand growth outpaces the expansion of high-end manufacturing capacity, the quantity a company can ship may be limited not by its most expensive component, but by whichever element of its bill of materials has the least supply elasticity.
What 1.6T Does to the Board
The PCB becomes a constrained item because of what higher data rates require from it, not because of volume alone.
As channels move from 800G toward 1.6T and then toward 3.2T, per-lane rates continue to rise. That makes dielectric loss, conductor loss, via stub effects, and impedance variation progressively more significant. In response, the material system has to move toward lower dielectric constant and lower dissipation factor, which is what drives adoption of M8 and M9 class high-speed laminates and low-profile, low-roughness copper foil such as HVLP grades.
At the same time, the module’s internal geometry becomes more demanding. High-speed interfaces and densely packed devices push high-density interconnect, any-layer structures, and finer lines into optical communication boards. Fine-line processes producing traces at 0.075 mm and below free routing space around high pin count ball grid arrays and high-speed differential channels, which is exactly the space that a 1.6T module does not have.
Manufacturing tolerance then tightens. Differential impedance control has to be held to a narrow band, in some high-speed applications toward plus or minus five percent, because at higher rates a deviation in material properties, copper thickness, line width, or a localised process variation converts directly into insertion loss, reflection, or crosstalk. The margin that a design could absorb at lower rates no longer exists.
Because these parameters are set during fabrication and assembly rather than corrected afterwards, the practical question for a supplier is whether the process can hold them in volume. Confirming achievable microvia size, line width, aspect ratio, and impedance tolerance against demonstrated capability, rather than against a target specification, is the purpose of a capability review. For optical communication programmes, that conversation belongs in telecommunications PCB planning and should happen before the layout is released.
A Shared Material Base Across Several Markets
There is a structural reason to expect optical module boards to remain tight even if indium phosphide supply improves. The materials they require are the same materials several other growing markets are competing for.
GPU compute boards, switch fabric, midplanes, and high-speed backplanes are all advancing simultaneously, pushing demand for 16 to 40 layer boards and, in the most complex AI server and switching systems, toward 50 and 78 layer structures. Low-loss laminates, low-profile copper foil, high-grade glass fabric, and now the filler systems inside those laminates are shared inputs. Automotive central computing platforms need high-reliability multilayer boards with high-speed interfaces. Robots and semiconductor equipment add their own demand for dense control boards and flexible interconnects.
The consequence is that the market is not competing for optical module boards in isolation. It is competing for the material and process capacity that produces every high-speed board in the industry. When one segment accelerates, the effect is felt in the others through material allocation long before it appears in a quotation.
That also explains why capacity figures can be misleading during such a cycle. A supplier may have square metres available and still be unable to accept a high-speed programme, because the constraint sits in laminate supply, in the number of lines qualified for a particular material, or in the availability of engineers who have already developed the process window for a new grade. This is the distinction between nominal capacity and effective capacity, and it is the distinction that determines whether a programme ships on schedule.
What This Means for Hardware Programmes
Treat board supply as a scheduling constraint, not a purchasing step. If the module family a programme is building requires M8 or M9 class material, the availability of that material and of a qualified process is a schedule input. Engaging board manufacturing during design, while stackup and impedance decisions are still open, converts an unknown into a plan.
Qualify alternatives early. Validating a second laminate or a second supplier takes weeks of build and electrical measurement. Doing it under allocation pressure removes negotiating leverage and forces compromises that would not otherwise be acceptable.
Plan inspection for what cannot be seen. Dense optical module boards place high pin count devices and fine-pitch components in a small area, and the joints that matter most are frequently hidden beneath a package. Paste inspection, optical inspection, X-ray, and electrical verification together cover the failure modes that determine whether a module transmits cleanly, and the resulting records belong in a defined quality system rather than in a project folder.
Verify the assembled result, not just the bare board. A board can meet its loss budget and still fail once the module is assembled, because paste volume, placement accuracy, and joint quality alter the electrical path. A structured test strategy that reaches the finished assembly is what prevents a marginal joint from becoming an intermittent link fault in a data centre.
Reading the Signal Correctly
The indium phosphide shortage will eventually be relieved by capacity additions, as previous component shortages have been. What will not reverse is the underlying pattern: as AI infrastructure expands, the constraint moves along the supply chain rather than disappearing. It has moved from accelerator supply to advanced packaging, then to high bandwidth memory, then to laminates, glass fabric, copper foil, silica filler, and now to the compound semiconductor materials inside optical transceivers.
For PCB manufacturers and their customers, the implication is that resilience comes from breadth rather than from a single advantage. A manufacturer that can cover high-layer-count boards, high-density interconnect, fine lines, and heavy copper power designs, and that maintains qualified material alternatives across several laminate systems, is better able to absorb a shortage in any one of them. For buyers, the corresponding lesson is that supply chain questions deserve a place in the technical review, because in this cycle the answer to “can this be built” increasingly depends on “can the material be obtained”.
Frequently Asked Questions
Why is indium phosphide in short supply? Demand for laser chips in optical modules has grown faster than substrate capacity. Estimated 2026 demand of 2.6 to 3.0 million two-inch equivalent wafers compares with effective capacity of about 750,000 wafers.
How much does a 1.6T module consume compared with 800G? A 1.6T module uses close to three times the indium phosphide based laser chips of an 800G module, so material demand scales faster than module count.
Why does the shortage affect PCB supply? Because higher optical module rates tighten board requirements for loss, via stub, and impedance control, while the low-loss laminates and low-profile copper foil needed are shared with AI servers, switches and other high-speed boards.
What board capabilities do optical modules require? Low-loss material, controlled-impedance high-speed channels, high-density interconnect and fine lines for dense device layouts, and reliable assembly of high pin count packages in a confined space.
What is the practical response for a programme? Engage manufacturing during design, qualify alternative materials before they are needed, and plan inspection and test to cover hidden joints on the assembled module rather than the bare board alone.
