2026 is widely regarded by the industry as the dividing line for 1.6T optical modules moving from “exhibition samples” to “volume orders.” LightCounting has already raised its 1.6T shipment forecast from a small base in 2025 to the level of tens of millions of ports, and 800G shipments are also expected to more than double. For overseas buyers, these figures represent both an opportunity and a hidden anxiety: if they do not secure an early position in 1.6T, the next-generation AI cluster expansion window may be seized by competitors. But the question truly worth asking is another one: when all leading suppliers claim to have 1.6T capabilities, on what basis should procurement decisions be made?
One easily overlooked fact is that the “deliverability” of 1.6T depends to a large extent on the yield of silicon photonics chips, not on capacity on the packaging line. Silicon photonics solutions already proved their power-consumption advantage in the 800G era. The silicon photonics chip yield of leading domestic manufacturers has exceeded 85%, with power consumption reduced by about 25% compared with traditional discrete solutions. However, 1.6T pushes the per-lane rate to 200G, and the tolerance windows for silicon photonics modulators, detectors, and coupling processes narrow correspondingly. The yield ramp curve will directly determine a supplier’s actual shipments in the first two volume-delivery quarters. If procurement teams only look at news headlines saying “already in mass production” without asking about “parameter consistency across production batches” and “the slope of the yield ramp,” they can easily encounter delivery slippage after orders are placed.
Power tiering is the second variable that needs to be treated precisely. For the same 800G, the module power of traditional DSP solutions can be above 14W, while LPO solutions can be below 8.5W, and silicon photonics solutions offer a more balanced compromise between the two. At 1.6T, absolute power levels move up overall, but the relative gaps between solutions do not disappear. For an AI data center planning to deploy tens of thousands of modules, a difference of a few watts per module multiplied by scale and annual operating hours will become a non-negligible operating cost item in electricity bills and cooling load. What buyers need to do is translate the power specification from “a number on a datasheet” into “a variable in a three-year TCO model,” rather than simply choosing the lowest value—because lower power often corresponds to a narrower applicable distance or stricter host-side supporting conditions.
The issue of delivery cadence also needs to be viewed in specific deployment scenarios. First-wave 1.6T demand is highly concentrated among the AI training clusters of hyperscale cloud providers. These customers usually lock up upstream capacity through long-term agreements and prepayments, leaving a not particularly ample “spot availability window” for small and medium-sized buyers. Major domestic manufacturers are already locking up incremental capacity for optical chips and DSPs through long-term agreements, while strengthening supply-chain ties through prepayments. This means that a buyer that has not entered a supplier’s early validation list in the first half of 2026 will find it difficult to obtain stable 1.6T supply through ad hoc orders in the second half. A more pragmatic strategy may be to maintain a posture of “technically ready but not overbetting” on 1.6T, while making silicon photonics and LPO 800G solutions the mainstay for current AI cluster scale-out, using an already mature supply chain to absorb immediate bandwidth pressure.
Ultimately, competition in 1.6T has moved from “who lights up first” to “who can first give customers confidence to put production traffic on it.” Speed numbers will converge quickly, but yield discipline, honesty in power tiering, and transparency in delivery cadence are the keys that distinguish whether a supplier can become a long-term partner. For buyers, the best bargaining chip is not obtaining a lower initial quote, but securing a place that is taken seriously on the supplier’s capacity allocation table.
