A structural supply shortfall in 1.6T optical module digital signal processors (DSPs) is expected to persist at least through the second half of 2027, according to a September 22 expert call with AI infrastructure service providers organized by Nomura. The imbalance stems primarily from insufficient 3-nanometer process capacity and low yield rates, rather than transient demand fluctuations.

Per trading desk sources, the Nomura optical communications chip expert call notes indicate that 2026 1.6T DSP demand is projected to exceed 20 million units, yet actual shipments will reach only 16 to 17 million units, leaving a gap of roughly 3 to 4 million units. In 2027, the supply-demand mismatch widens further, with demand climbing to 60 to 70 million units against supply of just 40 to 50 million units, expanding the shortfall to over 20 million units. The expert stressed that meaningful capacity release will not occur until the second half of 2027.

In the competitive landscape, exorbitant advanced process tape-out costs, customer qualification cycles lasting approximately two years, and complex signal processing algorithm barriers collectively form a moat in the high-end DSP market. This concentrates share among incumbent suppliers while prices decline moderately rather than triggering aggressive price warfare. Concurrently, the evolution of co-packaged optics (NPO/CPO) routes is reshaping value distribution across the supply chain—DSPs will be eliminated, with equalization and error correction functions shifting to transimpedance amplifiers (TIAs) and driver chips, imposing higher linearity requirements on those components.

Shortfall Trajectory: Widening from 2026 to 2027

The supply-demand trajectories for 800G and 1.6T DSPs diverge sharply. According to the meeting minutes, the expert noted that 800G DSP shipments are approximately 25 to 26 million units in 2025, expected to approach 50 million units in 2026, and further surge to 120 to 150 million units in 2027. Growth drivers include not only AI data centers but also telecom markets entering upgrade cycles. Since 800G DSPs utilize 7-nanometer and 5-nanometer processes, supply remains relatively ample.

The 1.6T situation differs markedly. Shipments exceeded just 3 million units in 2025, and while demand surpasses 20 million units in 2026, actual shipments will reach only 16 to 17 million units. In 2027, demand expands further to 60 to 70 million units against supply of merely 40 to 50 million units. The expert concluded that substantive capacity release awaits the second half of 2027, implying the supply-demand gap remains structural rather than cyclical for an extended period.

The direct causes of this shortfall are process bottlenecks: 1.6T DSPs require 3-nanometer technology, which currently suffers from insufficient capacity and low yields. Additionally, certain forward error correction (FEC) algorithms contain bugs pending fixes, and compatibility issues exist between high-speed DSP interfaces and network switches. These overlapping technical obstacles further suppress the pace of effective supply release.

Competitive Barriers: New Entrants Struggle to Break In via Low Prices

Competition in the high-end DSP market has not followed conventional price-war logic. Per the expert, DSP technical barriers are exceptionally high across three dimensions: first, manufacturing requires advanced processes below 7 nanometers with prohibitively expensive tape-out costs; second, signal processing algorithm complexity encompasses core capabilities like FEC, bit error rate control, equalization, and chromatic dispersion compensation; third, power management and thermal control present significant hurdles.

Customer qualification cycles further entrench incumbent suppliers. DSPs require joint testing by optical module manufacturers and end customers, with individual test cycles potentially lasting around two years. Switching costs for end customers are extremely high, driving continued partnerships with existing suppliers. While some second-tier manufacturers possess 800G DSP production capabilities, they lack robust customer ecosystems and cannot meaningfully penetrate the market. In China, firms like Sitrus Technology and Joywell Semi are developing 800G DSPs but have yet to achieve volume shipments.

Price Trends: Moderate Declines, Smaller Drops for 1.6T

Tight supply-demand conditions underpin relative DSP price stability. The expert noted that 800G DSP long-term agreement (LTA) prices currently range from $45 to $70, while 1.6T DSP LTA prices sit at $120 to $150, with spot prices exceeding contract levels by 10% to 20%. Looking ahead to 2027, DSP prices will decline modestly as capacity gradually comes online. The expert estimates 800G DSP contract prices will fall approximately 10%, while 1.6T DSPs see smaller declines due to larger supply gaps. This pricing path aligns closely with competitive dynamics—incumbent suppliers leverage technical barriers and customer loyalty to maintain relatively stable pricing power rather than engaging in intense price competition.

TIA/Driver Chips: Partial Value Transfer, Chinese Firms Still Lag

The evolution of NPO/CPO packaging routes is reshaping value distribution across the supply chain. Per the expert, NPO/CPO architectures eliminate DSPs to reduce power consumption, with equalization, bit error, and distortion correction functions partially assumed by TIA/driver chips and partially by switch chips. This imposes higher linearity requirements on TIA and driver chip technology. On current technical progress, leading TIA/driver manufacturers have achieved mass production at 200G per channel, with 400G still in R&D, primarily because simultaneous high linearity, low noise, and low power consumption present major challenges. Since TIA and driver chips can utilize mature 28-nanometer processes, Chinese manufacturers have achieved 100G mass production, with 200G in sampling stages. However, domestic firms still face notable gaps versus overseas competitors in product stability and customer qualification cycles.

Google has adopted 2.4T coherent-lite transceivers in its tensor processing unit (TPU) and optical circuit switch (OCS) networks, utilizing a single fixed-wavelength O-band covering 2 to 20 kilometer transmission distances. The expert indicated that coherent-lite DSP technical barriers are lower than traditional coherent DSPs but higher than standard PAM4 DSPs.

Verification Window: First Half of 2027 as Key Observation Period

Several conditions require ongoing tracking to validate the above supply-demand assessments. At the outcome level, actual 1.6T optical module and electrical chip shipments, along with 3-nanometer DSP supply release progress in the second half of 2027, serve as the most direct verification metrics. At the mechanism level, DSP yield improvement and FEC algorithm bug fix progress, 400G TIA/driver sampling and customer qualification developments, and NPO/CPO adoption rates at the switch side all warrant close monitoring.

The expert believes the current shortfall assessment warrants downward revision under the following scenarios: 2027 1.6T supply significantly exceeding 40 to 50 million units; DSP price declines markedly surpassing the assumed ~10% contract price drop; or NPO/CPO accelerating substitution that compresses DSP demand earlier than expected. The first half of 2027, the critical observation window before capacity release, represents the core verification period, with major manufacturers' quarterly earnings releases and optical module shipment data serving as important monitoring points.