In the AI optical communication industry chain, indium phosphide (InP) and thin-film lithium niobate (TFLN) play very different — yet equally indispensable — roles.

One is the material that “creates the heartbeat” of optical communication, while the other “controls the bloodstream.”
The former determines whether light signals can be generated at all; the latter determines whether those signals can be modulated fast enough, transmitted far enough, and controlled precisely enough.

Many people mistakenly see these two materials as competitors, assuming that thin-film lithium niobate will eventually “replace” indium phosphide. In reality, this reflects a misunderstanding of how optical communication systems actually work.

Today, let’s break down their roles in the clearest possible way: who does what, why this division of labor exists, and which technology is currently closer to large-scale commercialization.

1. Understanding the Division of Labor: Emission and Modulation Are Never the Same Job

If optical communication were a relay race, indium phosphide would be the starting runner — the one responsible for launching the signal. Thin-film lithium niobate would be the middle-distance accelerator — pushing transmission speed higher, extending distance, and maximizing efficiency. Silicon, meanwhile, acts more like the system coordinator on the sidelines: not generating light itself, but integrating all components into one platform.

Indium phosphide is essentially the “engine of light.”

In 800G and 1.6T optical modules, EML (Electro-Absorption Modulated Laser) chips must be fabricated on InP substrates because indium phosphide can efficiently emit light while naturally covering the two key low-loss optical fiber windows: 1310nm and 1550nm. Without InP, the fundamental optical source inside a module simply would not exist.

Thin-film lithium niobate, by contrast, is the “transmission gearbox of light.”

Its role begins after light is generated. TFLN modulators perform ultra-high-speed, low-power electro-optic modulation — encoding electrical signals onto optical waves by altering light intensity and phase. The modulator itself does not emit light, but it determines how fast signals can travel, how far they can reach, and how much power the system consumes.

In April 2026, Huatai Securities published a research report systematically comparing the growth logic of the InP substrate industry and the TFLN industry. The report emphasized that the two are complementary rather than substitutive inside optical modules. The next-generation optical module upgrade is not a matter of “either-or,” but rather a question of “who handles which function.”

2. Indium Phosphide: The “Light Engine” at the Core of AI Infrastructure

In the BOM (Bill of Materials) of 800G and 1.6T optical modules, optical chips account for more than half of total costs — and InP substrates are among the most critical foundational materials within those chips.

According to reports from Omdia and Yole, global demand for indium phosphide substrates (measured in 2-inch equivalents) is expected to reach approximately 2.0–2.1 million wafers in 2025, while effective global production capacity remains only around 600,000–700,000 wafers. This leaves a supply gap exceeding 70%.

By 2026, global demand is projected to surge to 2.6–3.0 million wafers, while production capacity may only increase to around 750,000 wafers. The shortage ratio is therefore expected to remain above 70%.

Pricing reflects this imbalance even more directly.

The price of 2-inch InP substrates rose from roughly USD 800 per wafer in early 2025 to around USD 2,300–2,500 per wafer, nearly tripling in a short period. Spot pricing for urgent orders has reportedly exceeded USD 3,000 per wafer.

NVIDIA predicts that overall demand for indium phosphide wafers may increase by nearly 20 times between 2026 and 2030. Huatai Securities also noted in its report that upstream core optical materials are entering a strong growth cycle, with InP substrates experiencing severe supply-demand tightness driven by rapidly expanding optical chip demand.

On the supply side, the industry remains highly concentrated. Japan’s Sumitomo Electric, the United States’ AXT, and Japan’s JX Metals collectively control over 90% of global production capacity. Meanwhile, expansion cycles typically require two to three years.

In February 2025, China officially added indium and indium phosphide-related materials to its export control list, further reinforcing the strategic importance of upstream InP resources.

3. Thin-Film Lithium Niobate: The “Optical Transmission Gearbox” Catching Up Rapidly

Thin-film lithium niobate does not generate light — but it solves precisely the problems where traditional modulation materials are beginning to hit physical limitations: bandwidth and power consumption.

Current mainstream TFLN modulators generally still operate with half-wave voltages above 1.8V. These relatively high driving voltages limit further increases in modulation bandwidth while also contributing to higher system power consumption.

However, rapid technological progress is changing the landscape.

In January 2026, Nature Communications published breakthrough research on ultra-broadband electro-optic modulators based on thin-film lithium niobate. The work demonstrated a record-breaking 800nm optical bandwidth covering the entire optical communication spectrum.

The modulator achieved electro-optic bandwidths exceeding 67GHz across the O-U telecom bands, with approximately 100GHz performance in the O/S/C/L bands and over 50GHz performance at the 2μm wavelength region. The device also demonstrated PAM-4 transmission exceeding 240Gbps per wavelength — setting a new performance benchmark for TFLN devices.

At OFC 2026, companies such as HyperLight and other TFLN vendors showcased thin-film lithium niobate chips and devices targeting ultra-high-speed optical modules, ultra-wide-bandwidth photonic chips, and next-generation modulators.

At the same event, Coherent presented 400G-per-channel solutions based on InP EML architectures, along with 3.2T transceivers and future-oriented architectures targeting beyond 12.8T systems.

The simultaneous presence of both technologies at OFC clearly illustrated two parallel technological pathways for future ultra-high-speed optical modules.

Huatai Securities explicitly categorized both InP substrates and TFLN as major long-term upstream opportunities in optical communication. Their relationship is expected to remain one of coexistence and complementarity rather than replacement.

Industry discussions and search analyses also indicate that although most TFLN modulators still maintain half-wave voltages above 1.8V, several engineering optimization strategies have already pushed some devices below 1.6V.

This suggests that future flagship devices — combining larger bandwidth, lower power consumption, and higher integration — are steadily moving from laboratory research toward real-world commercialization. TFLN technology remains in a rapid iteration phase, with manufacturing processes continuing to improve year after year.

4. The 1.6T and 3.2T Era: Division of Labor Will Become Even Clearer

As optical modules move from 1.6T toward 3.2T and beyond, the technological roadmap is becoming increasingly defined.

OFC 2026 already sent a strong signal: iteration cycles are accelerating rapidly.

1.6T optical modules are transitioning from limited-volume deployment toward large-scale commercialization, while the technical direction for 3.2T architectures has largely taken shape.

At the same time, silicon photonics penetration continues to rise quickly.

Industry forecasts suggest that silicon photonics solutions may account for more than 50% of 800G optical modules by 2026. In 1.6T modules, silicon photonics penetration could even reach 70–80%.

Yet silicon photonics itself does not provide a light source. It still relies on external continuous-wave (CW) lasers based on indium phosphide.

The higher the adoption of silicon photonics, the stronger the demand becomes for high-performance modulators such as TFLN.

As a result, optical modules are evolving away from “single-material dominance” and toward a collaborative ecosystem built around:

• Indium phosphide as the laser foundation
• Silicon photonics as the integration platform
• Thin-film lithium niobate as the ultra-high-speed modulation accelerator

This multi-material collaboration is becoming the true foundation for large-scale AI optical communication infrastructure.

Final Thoughts

Perhaps the biggest misconception in optical communication today is the idea that these two materials are rivals.

In reality, the opposite is true.

Indium phosphide generates the light source. Thin-film lithium niobate controls the speed and modulation. In many mainstream optical module architectures today, both technologies coexist inside the same packaged module, operating simultaneously along the same optical fiber and electronic system.

Whether in EML architectures, silicon photonics architectures, or future TFLN-based platforms, InP and TFLN each perform distinct functions within different stages of the same communication chain.

Their shared objective is clear: pushing the interconnect speed of AI computing clusters to its physical limits.

Indium phosphide creates the heartbeat. Thin-film lithium niobate enables the circulation.

Neither can replace the other.

In 2026, the InP market is facing supply shortages exceeding 70%, rapidly rising prices, and order backlogs extending into 2027. Meanwhile, TFLN breakthroughs are opening the door toward near-3.2T modulation capability across ultra-wide optical bands.

These technologies are not mutually exclusive. Their combined evolution is what is truly driving the next era of AI optical communication.

The future of optical communication is not a “replacement war” between materials — it is a highly specialized collaboration between complementary functions.