Key Points:
- Rosenblatt predicts that despite a 12-fold increase in production capacity, the optical component supply will remain 50% below demand through 2030.
- The severe shortage centers on Indium Phosphide (InP) lasers, which require rare, expensive wafers that cost thousands of dollars per unit.
- Optical components are essential to create high-performance xPU clusters that power advanced artificial intelligence workloads.
- Rosenblatt highlighted Coherent and Applied Optoelectronics as key near-term beneficiaries of this persistent networking bottleneck.
The explosive growth of global artificial intelligence (AI) infrastructure is running into another major physical bottleneck. According to a new research report from Wall Street firm Rosenblatt Securities published on Thursday, May 28, 2026, the global optical component supply will fail to keep pace with soaring data center demand through 2030. While optical component manufacturers expect to expand their production capacity by approximately 12 times between 2025 and 2030, Rosenblatt’s proprietary supply-and-demand model projects that total supply will remain roughly 50% below market demand through the end of the decade.
The severe technological shortage centers on specialized lasers that transmit data using light rather than electrical currents. Specifically, the analysis focuses on companies manufacturing Electro-Absorption Modulated Lasers (EMLs) and Continuous Wave (CW) lasers for high-speed optical transceivers. These mission-critical components require Indium Phosphide (InP) wafers, which cost thousands of dollars per 6-inch wafer due to the extreme rarity of indium and the high complexity of the chemical manufacturing process. In contrast to typical 12-inch silicon wafers used for computer processors, InP wafers are orders of magnitude more expensive to produce.
This optical hardware is absolutely vital because it enables the distributed architecture of modern AI computing systems. To train massive generative AI models, hyperscale data centers must connect thousands of individual graphics processing units and accelerators—collectively known as xPUs—into a single, unified supercomputer. Traditional copper wiring cannot handle the immense bandwidth required for these high-speed clusters over long distances without generating excessive heat and signal lag. Optical transceivers solve this by converting electrical data into ultra-fast, low-latency light signals that link xPU clusters both within and across massive data center campuses.
The structural supply deficit is not a new phenomenon but has steadily worsened as the AI infrastructure boom accelerates. Rosenblatt’s model shows that the supply of InP-based Datacom components was already approximately 50% behind market demand at the end of 2025. Although manufacturers are racing to migrate their production lines from smaller 2-inch, 3-inch, and 4-inch wafers to more efficient 6-inch wafers, this transition requires massive capital investments and years of technical recalibration. This slow-moving supply response has left data center operators facing prolonged procurement lead times.
To bypass some of these packaging bottlenecks, the tech industry is actively developing Co-Packaged Optics (CPO), which mounts optical transceivers directly on the same substrate as the main processor. However, Rosenblatt’s base-case scenario assumes that high-volume CPO production will only begin to build in the second half of 2027, with first commercial shipments delayed until 2028. The firm noted potential timing challenges across the entire supply chain, including bottlenecks in raw wafer production, test and measurement equipment, digital signal processors (DSPs), and photonic integrated circuits (PICs).
Despite the painful shortages for data center operators, Rosenblatt believes that this persistent supply-demand mismatch creates highly lucrative opportunities for select public tech companies. The firm highlighted Coherent Corp. (COHR) as a prime near-term beneficiary, predicting rapid revenue acceleration and gross margin expansion. This growth will rely heavily on Coherent’s ability to scale up its laser production on 6-inch wafers, directly feeding the high-margin market for 800-gigabit (800G) and 1.6-terabit (1.6T) optical transceivers.
Additionally, Rosenblatt identified Applied Optoelectronics Inc. (AAOI) as another high-potential pick for investors. The firm projects that Applied Optoelectronics is strongly positioned to double its market share, climbing to nearly 10% from its historical level of under 5%. The company’s focus on high-speed, virtual transceivers has made it an attractive alternative supplier for major hyperscalers—such as Microsoft, Google, and Meta—who are desperately trying to diversify their component supply lines to avoid a total dependence on industry giants like Broadcom and Lumentum.
The severe optical hardware bottleneck emerges as the global market for data center networking technologies experiences an unprecedented influx of capital. Driven by the AI boom, analysts project the optical networking market will exceed $154 billion by 2034, representing a massive shift from traditional CPU-based server architectures. As tech giants continue to spend billions of dollars on data center buildouts, the physical limits of our materials, factories, and wafer packaging systems will continue to dictate the speed of technological progress.
Ultimately, Rosenblatt’s detailed analysis proves that the artificial intelligence revolution is fundamentally a physical, resource-constrained challenge. Facing these physical limitations, developers cannot rely solely on software optimization. Until optical component manufacturers can successfully scale up their 6-inch Indium Phosphide production lines and implement co-packaged optics, the global optical component supply deficit will remain a persistent headwind for the tech sector, forcing developers to optimize their existing networks to survive the decade.
