EML laser TAM is growing with AI cluster optical interconnect demand — every 800G and 1.6T link inside a hyperscale AI data center requires EML-based transceivers. The base case reflects continued AI infrastructure buildout through 2030 with EML demand growing as average transceiver speed doubles every 2 years. Bear case assumes silicon photonics CPO achieves volume production by 2028, partially displacing discrete EML transceivers inside the data center and compressing TAM growth. Bull case assumes AI cluster sizes continue to expand (1M+ GPU facilities), 3.2T transceiver transitions begin by 2029, and EML suppliers cannot ramp fast enough — keeping pricing elevated and TAM above the base case.
“NVIDIA committed $2B each to Lumentum and Coherent in March 2026 — a $4B lockup of EML laser supply — alongside additional investments in Scintil Photonics and Ayar Labs. The move secured priority capacity and future access rights for NVIDIA through 2027+, leaving rival AI cluster builders unable to source EML supply at any price.”
“Fewer than five companies manufacture EML lasers at commercial scale: Lumentum, Coherent, Mitsubishi Electric, Sumitomo Electric, and Broadcom. EMLs combine a continuous light source and high-speed modulator on a single chip, encoding data onto optical fiber at up to 200 Gb/s per lane. Every 800G and 1.6T transceiver inside an AI cluster requires them — there is no qualified substitute at this speed.”
“AI data center optical component shortage has pushed 800G transceiver production 40–60% below demand through 2027, according to McKinsey. TrendForce and LightCounting both flag EML demand 30%+ above supply. Monthly EML capacity across all suppliers reached 50.7 million units in 2026 and is still insufficient — Broadcom, Lumentum, and Sumitomo collectively control ~55% of that output.”
“AI data centers require 16–36× more fiber and optical transceivers than conventional cloud facilities. At 1.6T switch speeds, every rack-to-rack link needs high-speed EML-based transceivers. The transition from 400G to 800G to 1.6T is compounding EML demand per data center — each speed doubling roughly doubles EML unit count per facility.”
“Silicon photonics co-packaged optics (CPO) could eventually eliminate discrete EML transceivers inside data centers — integrating the laser directly onto the switch ASIC package. Intel, Broadcom, and Marvell are all pursuing CPO. If CPO achieves volume production at 1.6T by 2028–2029, it reduces EML demand inside the data center while creating new demand for EML sources integrated into CPO modules.”
NVIDIA's $2B investment each in Lumentum and Coherent is the clearest signal of EML scarcity: NVIDIA did not pay $4B for components it could buy on the open market. The lockup secures priority allocation through 2027, functionally removing two of five global EML suppliers from the merchant market. Rivals building AI clusters — xAI, Meta, Google, Amazon — are competing for Sumitomo, Mitsubishi, and Broadcom output. The market already can't clear at any price.
Data centers are transitioning from 400G to 800G switches now, with 1.6T on the 2027 roadmap. Each speed doubling requires EMLs that modulate at double the frequency — not all existing EML manufacturing lines can produce the higher-speed variants. The transition creates a qualification bottleneck on top of the capacity shortage: even if suppliers had more units, not all units qualify for 1.6T applications.
Co-packaged optics integrates the laser source directly onto the switch ASIC package, eliminating pluggable transceivers inside the data center. If CPO achieves volume production at 1.6T by 2028, it restructures EML demand — fewer discrete transceivers, but new demand for EML sources embedded in CPO modules. Net effect on EML TAM is uncertain: CPO could expand the addressable market (more lasers per switch) or compress it (fewer modules per facility). NVIDIA's investment in Ayar Labs (a CPO startup) alongside Lumentum and Coherent suggests they are hedging both paths.