Lumentum Management: CPO Mass Production Inflection Point Locked in 2028
N.R. Finch
Lumentum told Bank of America's 2026 Global Tech Conference that EML laser supply trails demand by roughly 30% with no relief for two years — and that 2028 marks the start of CPO's mass-production cycle, turning optical components into AI infrastructure's hardest bottleneck.
EML lasers face a 30% shortfall — why can't the gap close?
EML lasers — high-speed chips that convert electrical signals directly into light — currently face a supply-demand gap of roughly 30%. Management sees no narrowing within two years.
This means → AI compute is expanding faster than optical capacity can ramp. Supply is chasing demand, not the other way around.
A single optical transceiver sells for hundreds of dollars. A few percentage points of yield difference translates into large cost gaps. In plain terms = whoever wastes fewer chips sets the price — this is a yield barrier, not a capacity game.
Why is 2028 the critical inflection point for CPO?
Management laid out a clear timeline: the next two years stay in scale-out mode (adding racks side by side), where pluggable transceivers suffice. By 2028, inference clusters expand to thousand-class XPU scale, and pluggable solutions hit a wall on power and density.
This means → CPO — co-packaged optics, placing optical components directly beside the chip — shifts from an option to a necessity. Per-rack optical demand rises at least 3×.
This reflects a deeper rhythm: AI infrastructure is moving from "lay down volume first" to "raise density per node" — and optics is the physical bottleneck for that density leap.
Cloud providers are diverging — who is betting on CPO, and who is still watching?
Roadmaps have visibly split: providers on the Nvidia ecosystem are going all-in on CPO; others stick with in-house architectures using OCS plus pluggable transceivers; a third group leans toward NPO — near-package optics, a socket-based bridge solution — as a short-term compromise.
NPO beats traditional pluggables on power but falls short of CPO. Management sees it as a transition product. After 2028, the industry will converge on silicon-photonics CPO built on the OCI MSA standard.
In plain terms = everyone is taking a different road right now, but the destination is likely the same — silicon-photonics CPO. The only question is who arrives first.
Nvidia takes a stake and signs take-or-pay — what does this lock-up mean?
Nvidia invested in Lumentum as a shareholder and separately signed a multi-year take-or-pay contract — locking both volume and price for ultra-high-power lasers paired with CPO.
Neither side agreed to exclusivity. This means → Nvidia is locking supply certainty, not an exclusive right. Lumentum retains the freedom to ship to other customers.
This reflects the current power map in optics: when a downstream giant is willing to prepay to secure supply, it confirms the supply side is the scarce side.
The pricing playbook has flipped — why has the price war stopped?
For thirty years the optical-component industry grew by undercutting on price. Now AI demand is surging while quality supply remains scarce, and structural pricing power is consolidating with top-tier suppliers.
Ultra-high-power UHP lasers must marry special modulators with silicon-photonics processes, demanding extreme bandwidth and noise control. High-speed EMLs need structural optimization to cut channel crosstalk — manufacturing difficulty rises in lockstep.
In plain terms = the old game was "overcapacity → cheapest wins." The new game is "scarcity → highest yield wins." The rules have flipped.
Supply-chain control and contract strategy — how is Lumentum deploying its resources?
Management named supply-chain control one of three top operational priorities. AI demand is tightening PCBs, lasers, photodetectors, and DSPs across the board — clearing supply-chain bottlenecks is a precondition for any capacity ramp.
Legacy products are being outsourced to free up top engineering talent for OCS and transceiver mass production, redirecting in-house wafer fab capacity to the highest-value lines.
Long-term agreements are predominantly take-or-pay, running two to three years with large orders reaching five. But the company will not lock all capacity into contracts — it reserves a portion to capture spot-market premiums. This means → the playbook is "contracts for the base, spot for the upside" — management is playing both ends.
Content is for reference only, not financial advice.