Indium phosphide (InP)

Indium phosphide (InP) is a III-V compound semiconductor that, unlike silicon, can efficiently generate and detect infrared laser light. It is the core material for the 1310/1550 nm lasers, photodiodes and modulators inside fiber-optic transceivers, making it a critical and supply-constrained input for AI data center networking.

What it is, in plain terms

Indium phosphide (InP) is a **compound semiconductor** made from indium and phosphorus, two elements drawn from groups III and V of the periodic table (hence the label "III-V"). Ordinary chips are made from silicon, a single element. InP is grown as small, expensive crystal wafers and used not to do computation, but to do something silicon physically cannot: turn electricity into light efficiently. It sits in a family of exotic materials alongside gallium arsenide (GaAs) and gallium nitride (GaN).

How it works: the direct-bandgap trick

The reason InP matters comes down to physics. Silicon has an *indirect* bandgap, so when an electron drops to a lower energy state it mostly releases heat, not light. InP has a **direct bandgap**: that same electron transition releases a photon cleanly and efficiently. That makes InP a natural laser and light-detector material. Crucially, its bandgap lines up with the **1310 nm and 1550 nm** infrared wavelengths where glass optical fiber carries light with the least loss (around 0.2 dB/km at 1550 nm). InP also has high electron mobility, which suits high-frequency electronics.

Why it matters for AI and data centers

AI training clusters move enormous amounts of data between thousands of GPUs, and copper wiring runs out of reach and power budget quickly. The fix is **optics**: convert the data to light, send it over fiber, convert back. Every one of those conversions needs a laser to make the light and a photodiode to detect it — and those are made of InP. As the industry moves to **800G and 1.6T optical transceivers** and **co-packaged optics (CPO)**, the number of InP lasers per system rises sharply. In short, InP is a small material with outsized leverage over how fast and how cheaply AI infrastructure can scale.

Where it sits in the supply chain

The chain runs: refined **indium metal** → **InP crystal/substrate wafers** → **epitaxy** (growing precise active layers on the wafer) → **laser and photodiode chips** → packaged **transceivers and optical engines**. Substrate makers sell bare wafers; epi houses and device makers turn them into working lasers. Modern designs often **bond InP laser dies onto silicon photonic chips**, combining InP's light-making ability with silicon's cheap, dense circuitry — so InP and silicon photonics are complements, not rivals.

Who the key players are

On **substrates**, Japan's **Sumitomo Electric** and US-based **AXT, Inc. (NASDAQ: AXTI)** together hold roughly 80% of global InP wafer capacity; the top five (adding Germany's Freiberger, JX Nippon Mining & Metals, and Taiwan's Visual Photonics Epitaxy) made up about 70% of 2024 revenue. On **epitaxy and devices**, compound-semiconductor foundries such as Taiwan's **WIN Semiconductors (TPEX: 3105)** — the largest GaAs/GaN foundry, with optical and InP exposure — and the transceiver makers **Coherent** and **Lumentum** convert those wafers into shipping product. NVIDIA underscored the stakes in late 2025 by committing about \$2 billion each to Coherent and Lumentum.

What's changing now

The story of 2025 has been **supply scarcity**. In February 2025 China — the dominant source of indium — imposed **export licensing on indium phosphide**, and because much of AXT's substrate capacity sits in China, shipments stalled until the first licenses arrived mid-year. Reported prices for 6-inch InP wafers jumped roughly **250% (to about \$5,000/wafer)**, and transceiver makers saw backlogs stretch into 2027-2028. The InP wafer market is small but fast-growing — estimated around \$211 million in 2025 and projected near \$628 million by 2035 (~11-12% CAGR) — which is why supply, not demand, is the binding constraint.

Frequently asked

Why can't silicon do the same job as indium phosphide?

Silicon is an indirect-bandgap material, so it releases energy mostly as heat rather than light and cannot make an efficient laser. InP has a direct bandgap and emits photons cleanly, which is why the light-generating parts of optical chips use InP even when the surrounding circuitry is silicon.

What wavelengths and devices use InP?

InP targets the 1310 nm and 1550 nm infrared bands where optical fiber loses the least signal. It is used for laser diodes, high-speed photodetectors, and optical modulators inside fiber-optic transceivers and photonic integrated circuits.

Why did indium phosphide prices spike in 2025?

China, the largest producer of indium, introduced export licensing on InP in February 2025. With a major share of substrate capacity located in China, shipments were disrupted and reported 6-inch wafer prices rose about 250%, tightening supply for AI data center optics.

Which companies are most exposed to InP?

Substrate suppliers Sumitomo Electric and AXT (AXTI) dominate wafers; Freiberger, JX Nippon and Visual Photonics Epitaxy round out the top five. Device and foundry players include WIN Semiconductors (3105), Coherent and Lumentum, the latter two backed by large NVIDIA investments in 2025.

Is InP a replacement for silicon photonics?

No. They are complementary. Silicon photonics provides dense, low-cost passive circuitry but cannot generate light efficiently, so InP laser chips are bonded to silicon photonic platforms. The two technologies are increasingly used together in the same package.

Related companies

$AXTI $3105