Citi TMT Conference: NPO Leapfrogs CPO, Lasers & Fiber Positioned as "The Next HBM"
nashnova research
Citi documents key takeaways from a Lightmatter management interview: near-package optics (NPO) has overtaken co-packaged optics (CPO) as the near-term consensus, while lasers and fiber are likened to 'the next HBM' on scarcity logic, reshaping the valuation narrative.
Why is NPO suddenly ahead of CPO?
Before the 2026 OFC conference, the market consensus was CPO first; after OFC, the spotlight shifted squarely to NPO.
NPO platforms are expected to appear in 2027, with high-volume deployment in 2028. CPO's scale-out window stays at 2028–2029.
This means → NPO is not replacing CPO — it is entering the deployment cycle one step earlier, with both tracks running in parallel on staggered timelines.
The catalyst: the OCI (Open Compute Interconnect) industry standard landed — prior architectural uncertainty had held back customer commitment. Once the standard dropped, design activity accelerated.
How does Lightmatter cut costs?
The core pitch is BiDi (bidirectional) fiber: one fiber carries both send and receive signals, eliminating the need for separate transmit and receive strands.
In plain terms = one pipe used to flow water in one direction only; now the same pipe runs both ways — fiber usage is cut in half.
The Passage L20 NPO engine delivers 6.4 Tbps bandwidth. Management estimates savings of roughly $1.6 billion in networking costs per gigawatt of data-center capacity, a steep unit-cost drop versus today's pluggable transceivers (benchmarked at about $0.50/Gbps).
Where is the CPO bottleneck?
Management was explicit: the bottleneck is not technology — it is test capacity.
Current CPO switch production sits at just a few thousand to ~10,000 units per year. The industry needs roughly 100× that, and wafer-level test equipment suppliers remain critically scarce.
This means → the test-equipment segment may be an underappreciated opportunity — it is not a problem of making the switches, but of testing them fast enough.
On industry coordination, Lightmatter's OCP silicon-photonics initiative has grown from 9 to roughly 19 partners and published a 300-page joint spec — positioned as the equivalent of an Nvidia reference design manual.
Why are lasers being compared to 'the next HBM'?
Single-device laser power is approaching physical limits — roughly 400 mW, with a ceiling around 500 mW. Pushing bandwidth further requires more lasers, not more powerful ones.
Lightmatter's answer: GaAs-based laser arrays on 300 mm CMOS wafers, 128 lasers per chip, achieving 16 Tbps per module in the latest product.
In plain terms = a single light bulb has maxed out its brightness; the only way to get more light is to install more bulbs — whoever can manufacture more bulbs controls the bottleneck.
This reflects an HBM-style dynamic: demand is locked in, supply is scarce, and whoever scales first captures the premium.
What do the photonic interposer and micro-ring modulator mean?
Management calls the photonic interposer — a chip-scale optical bridge — a one-of-a-kind product on the market today, removing shoreline constraints on laser placement and delivering roughly 10× the bandwidth density of alternatives.
The OCI MSA has formally adopted the micro-ring modulator — a tiny ring-shaped device that switches light signals on and off — as the industry default architecture, ahead of the electro-absorption modulator.
This means → the technology-path debate is largely settled, and the direction Lightmatter bet on is becoming the industry standard.
Where do the Nvidia partnership and tech validation stand?
In June 2026, Lightmatter and Nvidia struck a deal on NVLink Fusion; fiber interconnects will extend the scale-out domain beyond single-rack copper reach.
Management cited over 1 million hours of error-free data as proof of production readiness.
This means → this is no longer a slide-deck stage — with a top-tier customer endorsement and extended run-time data, the commercialization inflection point is closing in.
市场有风险，内容仅供研究参考，不构成投资建议。