Beyond Optical Modules: The Hidden Materials Powering AI

The AI Era's Real Bottleneck Isn't Chips — It's Materials

While the world obsesses over NVIDIA's latest GPUs and 800G optical modules, a far more fundamental battle is unfolding across the periodic table. According to the latest data from SEMI (Semiconductor Equipment and Materials International), the global semiconductor materials market has surged past $72.03 billion in 2025 — and the companies controlling these critical materials may hold the true 'entry tickets' to the AI era.

Here is a brutal reality the industry rarely discusses: no matter how fast an optical module operates, without an indium phosphide (InP) substrate, the signal cannot travel even 10 centimeters. Without ABF (Ajinomoto Build-up Film) substrates, the most powerful GPU on the planet is nothing more than a leaky silicon wafer.

Key Takeaways

• The global semiconductor materials market has reached $72.03 billion in 2025, per SEMI data
• Indium phosphide (InP) substrates are the irreplaceable backbone of AI-era optical communications
• ABF substrates from Japan's Ajinomoto remain a critical chokepoint for advanced GPU packaging
• China's share of advanced semiconductor materials has doubled from 10.2% to 20.4% over the past decade
• The materials supply chain — not chip design — may be the ultimate constraint on AI scaling
• Control over rare compound semiconductors is becoming a geopolitical flashpoint rivaling chip fabrication

Indium Phosphide: The 'Hard Currency' of Global Semiconductors

Indium phosphide (InP) has emerged as arguably the most strategically important compound semiconductor material in the AI age. In optical communications — the nervous system connecting data centers running large language models like GPT-4 and Claude — InP is absolutely irreplaceable.

Unlike silicon, which dominates traditional computing, InP possesses unique electron mobility properties that make it ideal for high-frequency, high-speed photonic devices. Every laser diode, every photodetector, and every modulator inside those celebrated 800G and emerging 1.6T optical transceivers depends on InP substrates.

The supply situation is alarmingly tight. Global InP wafer production capacity is concentrated among a handful of suppliers, primarily in Japan, the United States, and Europe. Companies like Sumitomo Electric, AXT Inc., and Wafer Technology (a subsidiary of IQE) dominate the market. Demand has skyrocketed as hyperscalers — Microsoft, Google, Amazon, and Meta — pour hundreds of billions into AI data center infrastructure.

Compared to silicon wafers, which are produced at massive scale and relatively low cost, InP substrates remain expensive and difficult to manufacture. A 4-inch InP wafer can cost upward of $500, while a comparable silicon wafer costs under $10. Scaling production is not simply a matter of investment — the crystal growth process for InP is inherently more complex and defect-prone.

ABF Substrates: Japan's Quiet Chokehold on AI Hardware

If InP is the bottleneck for AI's optical interconnects, ABF (Ajinomoto Build-up Film) substrates represent an equally critical chokepoint for AI processors themselves. Every advanced GPU, CPU, and AI accelerator using 2.5D or 3D packaging requires ABF substrates — and one company controls nearly the entire global supply.

Ajinomoto, the Japanese food and chemical conglomerate, holds an estimated 90%+ market share in ABF substrate film. This single-source dependency has caused recurring supply panics across the semiconductor industry.

• NVIDIA's H100 and B200 GPUs require ABF substrates for their advanced packaging
• AMD's MI300X accelerators face the same dependency
• Intel's Gaudi series processors also rely on ABF-based packaging
• Even emerging AI chip startups cannot escape this supply chain reality

The situation highlights a paradox in the AI industry: billions flow into chip design and AI model training, yet a chemical film produced by a food company in Japan remains a binding constraint. Ajinomoto has expanded capacity multiple times, but demand continues to outstrip supply as AI chip production scales exponentially.

China's Strategic Push Into Advanced Materials

Perhaps the most significant geopolitical dimension of this materials race is China's rapid ascent. Over the past decade, China's mainland share of the global semiconductor materials market has climbed from 10.2% to 20.4% — effectively doubling its footprint in a market long dominated by the United States, Japan, South Korea, and Taiwan.

This growth did not happen by accident. Beijing has designated semiconductor materials as a strategic priority, channeling billions in subsidies and research funding into domestic production capabilities.

• Yunnan Germanium and China Germanium have become major global suppliers of germanium, a critical material for infrared optics and certain semiconductor applications
• SICC (Shandong Tianyue) has scaled silicon carbide substrate production, challenging Wolfspeed and Coherent
• Chinese firms are investing heavily in gallium nitride (GaN) epitaxy for power electronics and RF applications
• Domestic InP substrate development programs are accelerating, though quality gaps remain versus Japanese and American suppliers

China's 2023 export restrictions on gallium and germanium — both critical semiconductor raw materials — demonstrated that Beijing understands the leverage these materials provide. The move sent shockwaves through Western supply chains and forced companies to scramble for alternative sources.

The Periodic Table as Geopolitical Battlefield

The semiconductor materials competition mirrors the broader 'tech cold war' between the United States and China, but with a crucial difference. While chip fabrication equipment and EDA software have clear chokepoints controlled by Western and allied nations (ASML, Applied Materials, Synopsys, Cadence), the raw materials picture is far more complex.

China controls approximately 80% of global gallium production and 60% of germanium production. These elements are not rare in geological terms, but refining them to semiconductor grade requires specialized infrastructure that has been concentrated in China over decades of industrial policy.

Conversely, Japan maintains dominance in high-purity chemicals, photoresists, and specialty films critical to advanced semiconductor manufacturing. JSR Corporation, Tokyo Ohka Kogyo (TOK), and Shin-Etsu Chemical collectively control enormous portions of these supply chains.

The U.S. has responded with initiatives under the CHIPS and Science Act, which allocates funding not just for chip fabrication facilities but also for upstream materials research and domestic production. The Department of Defense has separately identified compound semiconductors as a national security priority.

What This Means for the AI Industry

For companies building AI infrastructure, the materials bottleneck carries several practical implications that extend far beyond component pricing.

First, supply chain diversification is no longer optional. Any company relying on a single source for critical materials — whether ABF substrates from Ajinomoto or InP wafers from a single supplier — faces existential risk. The COVID-era supply chain disruptions provided a preview; geopolitical tensions could trigger far worse scenarios.

Second, materials costs are becoming a larger share of total AI infrastructure spending. As chip designs push toward more advanced packaging (CoWoS, chiplets, silicon photonics), the bill of materials for substrates, specialty chemicals, and compound semiconductors grows disproportionately.

Third, investment is shifting upstream. Venture capital and corporate R&D spending are increasingly flowing into materials science startups and advanced substrate manufacturers. Companies like Coherent (formerly II-VI), Wolfspeed, and IQE have seen renewed investor interest as the market recognizes that materials constraints could limit AI scaling more than Moore's Law.

• Data center operators should audit their supply chains for single-source material dependencies
• AI chip designers must factor in materials availability when selecting packaging technologies
• Investors should look beyond GPU makers to the materials companies enabling AI hardware
• Policymakers need to include materials security in national semiconductor strategies

Looking Ahead: Materials Will Define AI's Next Chapter

The next 3 to 5 years will be decisive. As the industry transitions from 800G to 1.6T and eventually 3.2T optical modules, InP substrate demand will multiply. As AI chips move toward 3nm and 2nm process nodes with increasingly complex packaging, ABF and other advanced substrate materials will face unprecedented demand pressure.

Several developments bear watching. Silicon photonics integration could potentially reduce InP dependency for some applications, though experts remain divided on timelines. Glass substrates, being developed by Intel and others, could eventually complement or partially replace organic ABF substrates for certain high-performance applications.

Meanwhile, the geopolitical dimension will only intensify. Export controls on semiconductor materials are likely to expand, and nations will accelerate efforts to 'reshore' critical materials production. The $72 billion semiconductor materials market is poised to grow significantly — SEMI projects continued expansion through 2027.

The bottom line is clear: in the AI era, controlling the materials is as important as designing the chips. The companies and nations that secure reliable access to InP substrates, ABF films, high-purity chemicals, and compound semiconductors will hold the true 'entry tickets' to AI's future. Everyone else will be waiting in line.