Citi China Conference: Three Optical Transceiver Companies Disclose 2027 Capacity and DSP Supply Highlights
nashnova research
At Citi's 2026 China conference, Eoptolink, DongShan Precision, and HGTECH all pinpointed 2027 as the year 1.6T optical modules hit commercial scale — but all three flagged DSP chip shortages as the binding constraint, meaning capacity plans hinge on foundry expansion timelines.
How big is the 2027 ramp for 1.6T modules?
Eoptolink's CFO said 2027 is the first year of 1.6T commercialization in China. Combined with 800G, the total market could reach 40–50 million units.
DongShan Precision is more specific: planned capacity for 800G and 1.6T combined is at least 35 million units in 2027, with required DSP supply largely locked in.
HGTECH expects 800G and 1.6T combined shipments to grow over 50% year-on-year in 2027, while 400G stays flat. This means → nearly all incremental volume comes from high-end products; the industry's center of gravity is shifting fast.
Where exactly is the DSP bottleneck?
DongShan Precision stated that 2026 deliveries of 1.6T modules are constrained by shortages of 3nm DSPs — digital signal processors, the core chips inside optical modules that handle signal conversion. Relief is expected once foundry capacity expands in Q2 2027.
HGTECH's response: a long-term DSP agreement worth over $1 billion with major overseas suppliers, covering most of its projected 2027 demand.
HGTECH ranked component shortages by severity: DSP > high-end PCB > high-end MLCC > InP substrates. In plain terms = DSP is the number-one chokepoint, but several other links in the supply chain are also tight.
DongShan Precision said it has seen no customer order cancellations. This reflects that downstream demand remains firm — the problem is on the supply side, not the demand side.
How far along is domestic DSP substitution?
HGTECH's target: near-100% domestic DSP adoption for 400G products by 2027, with 800G localization also accelerating.
This means → 400G, as the mature product, is leading the domestic-substitution push. 800G and 1.6T still rely heavily on overseas suppliers — which is precisely why the billion-dollar long-term agreement was signed.
NPO, CPO, or pluggable — which technology wins?
Eoptolink said three paths coexist today: pluggable, NPO — near-package optics, moving the module closer to the chip to shorten signal distance — and CPO — co-packaged optics, integrating the module directly onto the chip.
NPO is seen as the mainstream direction for vertical interconnect scaling; CPO is a longer-term play driven by global leaders. HGTECH expects China NPO orders to begin in 2027, with volume production by year-end.
HGTECH provided a key pricing signal: 3.2T NPO will cost roughly 1.7–1.8× the price of a 1.6T pluggable module, with gross margins expected to exceed 30%. In plain terms = NPO is not just a technology upgrade — it is a margin upgrade, with higher ASPs and thicker profits.
DongShan Precision disclosed that silicon-photonics modules are taking a larger share of its shipment mix. The original 2027 plan was a 7-to-3 split between EML and silicon photonics; silicon photonics is now rising.
What is happening with the Thailand lines and overseas expansion?
DongShan Precision's Thailand factory began trial production in September 2026, entered volume production in November, with Phase 1 designed monthly capacity of 1.5 million units. The factory posted a loss of over RMB 100 million in Q3 (mainly from PCB operations); Q4 losses are expected to narrow slightly.
HGTECH is expanding overseas faster: optical-communications revenue outside China grew roughly 400% year-on-year in the first nine months of 2026. Thailand monthly capacity is expected to exceed 1 million units by Q3 2027. Management expects overseas revenue could surpass China revenue in 2027.
HGTECH's H-share listing is on track, targeting completion in H1 2027. This means → overseas capacity and capital-market positioning are advancing in parallel — a two-track hedge against geopolitical risk.
What is optical computing, and what is Lightelligence doing?
Lightelligence builds matrix multiply-accumulate chips using silicon-photonics technology — performing AI's most resource-intensive calculations with light signals instead of electrical ones. Compared with electronic chips, these are faster, lower-latency, and lower-power, and do not require advanced process nodes.
Two products are already commercialized. The core target customers are token factories — large-scale compute centers running AI inference workloads.
Lightelligence's CFO described the field as a blue ocean, with key players including Lightmatter overseas and Huawei's optical technology team.
Both Eoptolink and Lightelligence operate dual supply chains: a China chain serving Chinese customers and an existing overseas chain serving international customers. In plain terms = two chains run independently, so a geopolitical disruption cannot break both at once.
市场有风险,内容仅供研究参考,不构成投资建议。
