Soitec Locks In AI Optics Demand With Deposits As Photonics Wafer Orders Surge

French chip materials maker Soitec is using a surge in demand for wafers used in AI data-center optics to secure multi-year customer commitments, with deposits and fixed pricing designed to give the company greater visibility over future demand and protect its margins.

The strategy comes as hyperscalers race to expand AI computing infrastructure and increasingly turn to optical connections to move data between processors. Copper connections are becoming less attractive for some high-speed applications because of their power consumption and performance limitations, increasing demand for silicon photonics.

"We are using the current situation to find the right balance between the value we bring and the price we can ask," Soitec CEO Laurent Remont told Reuters.

Soitec told investors last month that revenue from photonics-SOI, the silicon substrate used to manufacture silicon photonics chips, would more than double in the current financial year from slightly above $100 million. Remont said that forecast now represents "absolutely a floor," implying revenue of more than $200 million.

The acceleration is of the essence to Soitec because silicon photonics is becoming an increasingly important component of AI infrastructure. As AI systems require ever greater volumes of data to move between computing and networking components, optical technology offers advantages in speed, distance and energy efficiency.

Soitec supplies the substrate used by almost all silicon photonics chips, according to UBS, which estimates the French company controls about 95% of the market. Its shares have almost quadrupled this year as investors have bet on the company benefiting from the expansion of AI-related optical networking.

Rather than simply expanding capacity immediately, Soitec is seeking to make customers commit capital alongside their orders.

About 80% of the company's capacity reservation agreements with more than 10 photonics customers are expected to be signed within the next one to two weeks, with the remainder expected within a month, Remont said.

The agreements will lock in prices and require customers to put down deposits against committed volumes. Customers that take the agreed quantities will have their deposits returned, while those that fall short will forfeit them. Orders above contracted volumes will be subject to fresh pricing negotiations.

"That's a way for us to have our customer with skin in the game," Remont said.

The arrangement gives Soitec greater confidence when allocating scarce manufacturing capacity while limiting the risk that customers reserve more wafers than they ultimately need. Customers will also be required to share inventory information, which Soitec says will help prevent companies from accumulating excess capacity simply to keep wafers away from competitors.

The approach could prove important as the AI supply chain moves from short-term capacity concerns toward longer-term commitments. For Soitec, securing demand before committing billions of dollars to new manufacturing facilities reduces the risk of expanding too aggressively if the current AI investment cycle eventually moderates.

The company does not expect to require a new fabrication plant until around 2029. Instead, it plans to increase production through existing assets.

One option is to shift output between businesses where facilities are underutilized. Another is to install additional manufacturing equipment in existing cleanroom space.

"With that we will cover easily this year and next year," Remont said.

Soitec can also repurpose part of a French facility originally built for silicon carbide production. The company wrote down €41 million ($47.7 million) of that facility last year.

Singapore provides another potential source of expansion. Soitec produced photonics-SOI exclusively in France until five months ago, but has since qualified customers at a Singapore facility. The company also has an unequipped building there that could be fitted with manufacturing equipment instead of constructing an entirely new plant.

A decision on whether to equip that building is expected within six to 12 months, Remont said.

"We can increase quickly without building a completely new fab, just equipping a building," he said.

The strategy also means Soitec currently sees little need to establish manufacturing capacity in the United States, even though much of the AI infrastructure boom is being driven by U.S. technology companies.

"We don't need a U.S. plant at this stage," Remont said, adding that customers are "more desperate to get wafers than being too picky about where the location for production is."

The comments indicate that in the AI semiconductor supply chain, demand is no longer concentrated only in the processors that train and run AI models. Supporting technologies such as high-bandwidth memory, advanced packaging and optical networking are becoming critical bottlenecks as data-center operators build increasingly powerful systems.

That creates an opportunity for Soitec to translate its dominant position in photonics substrates into longer-term contracts, better pricing visibility, and potentially stronger returns on existing manufacturing assets before it commits to the much larger expense of building a new fab.