AI optical bottleneck: U.S.-China rift sparks supply chain shakeup

The indium phosphide (InP) shortage has emerged as a critical bottleneck in AI optical interconnect production, with supply-demand imbalances exceeding 30%, surpassing even the memory shortage crisis. Industry leaders like Lumentum CEO Michael Hurlston warn that shipments are falling significantly short of demand as AI data centers ramp from thousands to hundreds of millions of units, underscoring the unprecedented scale-up challenges for non-silicon materials.

This shortage is deeply rooted in the complex upstream supply chain, as indium—a key raw material for InP—is predominantly co-mined rather than sourced via dedicated channels, creating bottlenecks that ripple through wafer and device manufacturing. Compounding these constraints are export controls imposed by China, which dominates roughly 70% of global refined indium production and has sharply curtailed exports since early 2025, driving wafer prices up by 250% and threatening the scalability of Western manufacturers reliant on Chinese indium.

In response to these material and production challenges, major AI and data center players like NVIDIA have made massive investments—totaling billions of dollars—in dominant InP producers such as Lumentum and Coherent, who are aggressively expanding capacity through new fabs and technology upgrades. For example, Coherent is scaling its 6-inch InP production line in Texas, while Japan’s JX Advanced Metals plans a 7–10x substrate capacity increase, and Chinese firms like Sanan Optoelectronics are localizing wafer manufacturing to meet soaring demand amid a global capacity arms race.

Despite robust revenue growth and pricing power among InP-related companies, market skepticism persists as reflected in recent sharp stock sell-offs, signaling uncertainty about how quickly supply constraints can be resolved. Meanwhile, the indispensable role of InP-based lasers—especially electro-absorption modulated lasers (EMLs) critical for AI data center optical interconnects—remains unchanged despite shifts toward silicon photonics, as no copper or alternative solutions can match the required bandwidth and energy efficiency at scale, prolonging the urgency of overcoming these production and material hurdles through 2028.

In a landmark move to secure U.S. leadership in AI hardware, the government awarded GlobalFoundries a $300 million CHIPS Act grant tied to a 1% equity stake, signaling a strategic intertwining of public funding and corporate growth. This investment targets critical advancements in silicon photonics, including next-generation wafer technologies and advanced packaging techniques like 3D hybrid bonding, aimed at overcoming copper interconnect bottlenecks and enabling scalable co-packaged optics essential for AI and high-performance computing infrastructure. Industry heavyweights such as Nvidia, AMD, and Marvell have endorsed this initiative, emphasizing that connectivity innovations are now as vital as compute performance for future AI systems.

Beyond GlobalFoundries, the U.S. Department of Commerce has strategically invested up to $874 million across seven disruptive semiconductor ventures, including companies pioneering co-packaged optics, ferroelectric memory, and novel optoelectronic substrates, to lead the emerging 'post-GPU era.' This diversified portfolio reflects a deliberate shift from laissez-faire market policies toward active technological encirclement, with the government securing minority equity stakes without control rights to align national strategic interests with capital deployment. Such investments aim to accelerate innovation timelines by two to three years, reducing reliance on foreign suppliers and reshaping the semiconductor landscape.

GlobalFoundries is aggressively building an end-to-end silicon photonics manufacturing ecosystem in the U.S., exemplified by its acquisitions of Advanced Micro Foundry and Infinilink and the launch of its SCALE™ co-packaged optics platform. This platform targets 400 Gbps data transmission speeds with up to fivefold improvements in energy efficiency, positioning the U.S. to overcome critical AI data-movement bottlenecks. Complementing these upstream efforts, companies like Aeluma are advancing large-substrate optoelectronic device manufacturing by integrating compound semiconductors on 12-inch wafers, a significant scale-up from traditional 2- to 4-inch indium phosphide substrates, further bolstering domestic supply chain resilience.

The proposed U.S. ban on Chinese-made optical transceivers, which dominate over 60% of the global data center market through suppliers like Innolight and Eoptolink, threatens to severely disrupt AI infrastructure supply chains by tightening critical networking equipment availability and inflating costs. This ban not only jeopardizes Chinese manufacturers but also imperils U.S. chipmakers such as Broadcom, Marvell, and Nvidia, who depend heavily on Chinese transceiver demand for their digital signal processors and lasers, underscoring the deep interdependence within the supply chain.

Replacing Chinese manufacturing capacity for advanced optical transceivers, especially next-generation 800G and 1.6-terabit modules critical for hyperscale AI deployments, is a formidable challenge. U.S. suppliers like Lumentum Holdings and Coherent Corp. currently lack the production scale and face material bottlenecks due to China's control of approximately 70% of global refined indium phosphide, a key substrate for laser chips, whose price has surged 250% since export controls tightened in early 2025. This material dependency risks undermining the effectiveness of the ban and could delay U.S. AI infrastructure expansion by 12 to 24 months.

The regulatory landscape is complicated by ambiguities in the FCC's draft ban, which may allow Chinese optical transceiver makers to circumvent restrictions by relocating production to Southeast Asia, as seen with Innolight and Eoptolink's capacity shifts to Thailand. Moreover, the ban is expected to be phased, permitting continued use of certified 800G modules while restricting certification of next-generation 1.6T modules, signaling a strategic, long-term pressure tactic rather than an immediate supply cutoff. This nuanced approach reflects the structural dependency where U.S. firms control core chip technologies but rely on China's industrial integration and packaging prowess for mass production.

This geopolitical friction is catalyzing a lucrative capital rotation toward Western optical transceiver manufacturers like Lumentum and Coherent, who are experiencing robust revenue growth—Lumentum's nearly 90% year-over-year surge to $808 million and Coherent's $1.81 billion in revenue—alongside soaring stock valuations with price-to-earnings ratios around 144. However, despite these gains, these companies currently cannot fully compensate for the supply shortfall caused by restricting Chinese imports, placing hyperscalers such as Microsoft and Alphabet in a precarious position that could slow AI infrastructure deployment and reshape the physical networking architecture into a tightly regulated national security domain.

By mid-2026, U.S. AI infrastructure leaders such as Broadcom, Nvidia, and AMD confronted escalating supply chain risks stemming from potential bans on Chinese-made optical transceivers, which dominate over 60% of the global data center market. This regulatory pressure threatens to constrict the availability of critical networking hardware, driving up infrastructure costs and inflationary pressures, while simultaneously disrupting demand for U.S.-based DSP chip suppliers like Broadcom, Marvell, Credo, and MaxLinear, who rely heavily on Chinese transceiver manufacturers as key customers.

Western laser component suppliers such as Lumentum and Coherent face a double-edged sword: while poised to benefit from the U.S. restrictions on Chinese optical modules, their capacity expansions focus predominantly on high-margin indium phosphide (InP) laser chips rather than the fiercely competitive, low-margin module packaging segment, which remains firmly in Chinese hands due to superior industrial integration and scale. This complex interdependence means that even aggressive Western capacity ramp-ups will struggle to replace the volume and yield of Chinese module assembly in the short to medium term, potentially delaying AI infrastructure deployment despite regulatory intent.

The U.S. government's phased approach to restricting Chinese optical module imports—allowing continued use of certified 800G modules while tightening certification for next-generation 1.6T devices—reflects an acknowledgment of the intertwined supply chains and the technological bottleneck shifting from GPUs to networking hardware. This transition is underscored by the rapid growth of the optical transceiver market, projected to surge 41% year-over-year to $31 billion in 2026, driven by AI superclusters from Nvidia, AMD, Google, and Amazon, where high-speed 800G and 1.6T optical modules become indispensable infrastructure components with diminished price sensitivity due to operational necessity.

Investor sentiment vividly mirrors the market’s anticipation of regulatory tailwinds favoring Western optical component manufacturers, with Lumentum’s stock trading at a premium price-to-earnings ratio near 144 and analysts forecasting a surge to $1,400 per share. This bullish outlook is fueled by Lumentum’s nearly 90% year-over-year revenue growth and Coherent’s robust $1.81 billion revenue, highlighting a capital rotation toward firms capable of filling the looming supply void, even as the industry braces for short-term shortages and the complex challenge of realigning global AI optical interconnect supply chains.

Enablence Technologies and ShunYun Technology have successfully established a fully integrated optical chip supply chain outside China, spanning design in Ottawa, wafer fabrication and specialized assembly in Fremont, California, and high-volume packaging and assembly in Vietnam. This end-to-end platform supports advanced optical products critical for AI data centers, including 800G/1.6T optics, Co-Packaged Optics (CPO), and Linear Pluggable Optics (LPO), positioning itself as a robust China-alternative for North American hyperscalers and telecom operators.

This strategic realignment emerges as a direct response to potential U.S. regulatory restrictions on Chinese optical transceivers, aiming to mitigate supply chain risks by reducing reliance on Chinese-origin components. Todd Haugen, CEO of Enablence Technologies, emphasizes that the Vietnam facility not only completes the supply chain but also ensures a continuous, compliant supply without the threat of import blocks or regulatory re-certifications, thereby safeguarding North American AI and telecom infrastructure.

ShunYun Technology’s Vietnam production site has reached full operational readiness, having completed installation and qualification of all production-line equipment. The facility has already produced qualified samples released to key North American customers for evaluation, signaling its preparedness to support the growing demand for high-speed AI data center components and marking a significant milestone in diversifying global optical interconnect manufacturing hubs.