Not all AI-related stocks seem to be rising together. Lumentum went up roughly 10 times more than Nvidia, and Applied Materials also gained far more. Why are these optical component and equipment companies suddenly outperforming so dramatically? And can it last?

The Bottleneck in AI Is Shifting

The extreme performance gap between Nvidia (YTD roughly +12%) and Lumentum (+121%) or Applied Materials (+67%) is not a simple rotation. It is a structural signal that the bottleneck in AI datacenters is migrating from GPU compute to optical interconnects.

Why Optical Interconnects — and Why Now

AI clusters operate with thousands of GPUs working in parallel. When the speed at which these GPUs exchange data is slower than their processing speed, even the best Nvidia chips are bottlenecked by the "slow network" — what the industry calls an interconnect bottleneck.

Worldwide shipments of high-speed optical transceivers above 800G are forecast to jump from 24 million units in 2025 to 63 million in 2026 — a 2.6x surge (TrendForce, April 2026). Yet fewer than five companies globally can manufacture EML (electro-absorption modulated laser) chips at commercial scale — devices that convert data running at 200 gigabits per second into light. The full list is Lumentum, Coherent, Mitsubishi, Sumitomo, and Broadcom.

What Nvidia's $4 Billion Bet Reveals

On March 2, 2026, Nvidia invested $2 billion each in Lumentum and Coherent — a combined $4 billion — primarily to lock in EML supply (Futurum Research, 2026). The fact that Nvidia itself treats optical interconnects as a strategic bottleneck is telling. Competitors trying to source EML from Lumentum or Coherent are pushed back to 2027 or later (TechTimes, May 2026).

Applied Materials' gains reflect a different layer. Scaling AI clusters requires TSMC, SK Hynix, and Samsung to expand capacity — and that semiconductor equipment demand concentrates in AMAT. Samsung's HBM4 ramp and semiconductor exports +205.8% (June 1–10) substantiate this.

How Long Can It Last?

The optical transceiver market is expected to grow 57% from $16.5 billion in 2025 to $26 billion in 2026 (semiconductor-today.com, April 2026). The 800G transceiver shortage is projected to persist through 2027, and the 1.6-terabit class through 2029 according to McKinsey.

Two risks warrant attention.

First, optical networking companies soared in the internet boom — then crashed. Cisco, JDS Uniphase, and Corning surged dozens of times in 2000 before falling more than 90%. The key difference today is that demand is backed by actual datacenter capital expenditure ($200B+, combined across Meta, Amazon, Google, and Microsoft).

Second, supply may expand. EML manufacturing currently resists automation due to extreme optical alignment precision requirements, but investment in new indium phosphide (InP) production lines is growing. Lumentum is expanding its InP facility in North Carolina (Digitimes, March 2026).

The Signal Investors Should Actually Watch

Nvidia is not underperforming because it is "expensive" — it is already fully priced for expectations, while component companies with severely constrained supply relative to surging demand are catching up. The inflection point — when new EML production lines come online or 800G oversupply becomes visible — is likely the peak for optical names. For now, EML lead times remain pinned at "2027+," and no oversupply signal is visible.