NextFin News - The global semiconductor landscape is bracing for a tectonic shift as Nvidia accelerates its transition to silicon photonics (SiPh) for its 2026 product roadmap. According to Digitimes, the upcoming Rubin Ultra platform is set to become the primary catalyst for the mass adoption of 1.6T high-speed interconnects, forcing the global optical supply chain into a state of rapid mobilization. This strategic pivot comes at a critical time for U.S. industrial policy; under the administration of U.S. President Trump, there is a heightened focus on maintaining American dominance in AI infrastructure, where silicon photonics is increasingly viewed as the "nervous system" that will prevent the industry from hitting a physical performance ceiling.
The move toward silicon photonics is not merely a choice but a necessity born of physics. As AI models grow exponentially, data centers are encountering what engineers call the "copper wall." Traditional copper-based electrical communication generates excessive heat and resistance when pushing data at the speeds required by next-generation GPUs. By replacing electrons with photons, Nvidia aims to drastically reduce power consumption while increasing bandwidth density. According to Nomura Securities, shipments of 1.6T optical modules are expected to surge from 2.5 million units in 2025 to 20 million units by 2026, with silicon photonics penetration in the 800G/1.6T market projected to reach between 50% and 70%.
This transition is reshaping the competitive dynamics of the supply chain. Industry heavyweights such as Broadcom and Marvell are already locked in a race to provide the underlying silicon for these optical engines. Broadcom has taken an early lead with its "Bailly" 51.2T Co-Packaged Optics (CPO) switch, which integrates optical engines directly onto the switch chip to eliminate copper traces. Meanwhile, Marvell is ramping up its "3D Silicon Photonics" engine to cater to hyperscalers like Google and Amazon who are seeking custom silicon solutions to rival Nvidia’s vertical integration.
The mobilization extends deep into the manufacturing and assembly layers. Specialized foundries and assembly partners are seeing unprecedented demand for high-precision optical alignment. Fabrinet, a key assembly partner for Nvidia, is positioning itself as the primary provider for the first wave of CPO deployments. On the component side, Lumentum and Coherent are expanding capacity for EML (Electro-absorption Modulated Laser) chips, which are essential for maintaining signal integrity at 1.6T speeds. According to industry estimates, the supply of advanced photonic chips may still fall short of demand by 5-15% in 2026, despite an 80% year-on-year increase in production capacity.
From a geopolitical and economic perspective, the acceleration of silicon photonics aligns with the broader "America First" tech strategy of U.S. President Trump. By fostering a domestic ecosystem for advanced optical packaging and manufacturing—led by companies like Applied Optoelectronics and GlobalFoundries—the U.S. aims to secure the AI supply chain against external disruptions. GlobalFoundries’ recent acquisition of Singapore’s Advanced Micro Foundry (AMF) is a clear signal of this intent, as it seeks to become the world’s largest pure-play silicon photonics foundry, offering a "China-free" supply chain for U.S. hyperscalers.
Looking ahead, the year 2026 will likely be remembered as the point where optical communication moved from the periphery to the core of compute architecture. The successful launch of the Rubin Ultra platform will validate the commercial viability of CPO and SiPh at scale. As the industry moves toward 3.2T and beyond, the companies that master the integration of light and silicon today will hold the keys to the AI infrastructure of the next decade. For investors and analysts, the focus has shifted from who builds the best "brain" (the GPU) to who builds the most efficient "nervous system" (the optical interconnects).
Explore more exclusive insights at nextfin.ai.
