Why is silicon photonics expected to gain share?

  • Nomura expects 800G and 1.6T high-speed optical modules, as well as near-packaged and co-packaged optics solutions, to continue supporting demand for optical chips over the next 3–5 years. The bank believes that solutions combining continuous-wave lasers with silicon photonics, which have more diversified supply chains and more competitive costs, are gaining share from electro-absorption modulated lasers (EML).
  • For network connections between data-center servers, Nomura expects silicon photonics modules based on continuous-wave lasers to reach 60%–70% penetration in 800G and 1.6T pluggable modules by 2027, with the remainder consisting mainly of EML solutions.
  • The shift in share does not mean demand for indium phosphide (InP) will disappear: silicon photonics still needs InP lasers as light sources. Nomura expects shipments of InP laser chips, including EML, VCSEL and continuous-wave lasers, to rise from 480 million units in 2026 to 1.542 billion units in 2028.

Why will supply still diverge despite growing demand?

  • The upgrade from 800G to 1.6T is driving EML single-channel speeds from 100G to 200G and continuous-wave laser power from 70 milliwatts to 100 milliwatts; near-packaged and co-packaged optics are adding demand for lasers above 120 milliwatts.
  • Nomura expects supplies of low- and mid-range chips, such as 70/100 milliwatt continuous-wave lasers and 100G EML, to become more plentiful over the next 1–2 years as Chinese manufacturers enter the global market. However, continuous-wave lasers above 200 milliwatts, EML above 200G and high-speed VCSELs remain largely in the hands of global manufacturers. Slow capacity expansion, long equipment lead times and yield issues will keep the supply gap in place through 2026–2027.
  • Upstream InP substrates are also a bottleneck: high-end supply is concentrated among Sumitomo Electric, JX Advanced Metals and AXT. Nomura believes China's export controls on InP substrates and raw materials could also affect effective global supply.

How do VCSEL and MicroLED challenge existing approaches?

  • Silicon photonics currently mainly uses a “fast and narrow” architecture with a small number of high-speed channels; VCSEL and MicroLED use a “slow and wide” architecture that transmits in parallel over more low- and mid-speed channels. Nomura positions silicon photonics for longer distances, VCSEL for roughly 10–100 meters, and MicroLED for ultra-short distances within 10 meters.
  • Microsoft's MOSAIC solution illustrates the trade-off between channel count and speed: a conventional 800Gbps link typically uses 8 channels at 100Gbps each, while MOSAIC transmits in parallel over 400 optical channels at 2Gbps each.
  • The new approaches still face engineering hurdles: MicroLED must address yield and consistency in the mass transfer of tiny light-emitting devices; VCSEL must deal with array coupling, alignment and high-density packaging.

What trade-offs do modulator and module suppliers face?

  • Modulators encode electrical signals onto light waves. Broadcom's co-packaged products use the more mature MZM approach, which occupies more chip area but is less sensitive to temperature; the microring modulator (MRM) used by Nvidia is smaller and requires a lower drive voltage, but needs thermal control circuitry to suppress temperature-induced wavelength drift.
  • Nomura believes Chinese manufacturers' strength in pluggable optical modules helps domestic optical chip companies gain share. Multiple optical solutions will coexist over the next 2–3 years, and cloud service providers and AI customers will also need customized products. At the same time, global customers are seeking supply diversification, and Chinese optical chip manufacturers still face regulatory risks in expanding overseas.