Innolight Clarifies 1.6T Price Cut Rumors, Orders Through 2027 Already Secured
N.R. Finch
Innolight held an investor call to deny rumors of extreme 1.6T transceiver price cuts, stating orders now extend to 2027 with monthly delivery volumes locked in — and some key customers have already issued 2028 guidance for next-generation products. This means the market's fear that AI capex is unsustainable finds no support in this leading supplier's order book.
What actually happened with the price-cut rumors?
Management called the circulating extreme-low 1.6T prices "seriously inaccurate," saying next year's product pricing is far above rumored levels.
This means → extrapolating industry-wide ASPs from a single rock-bottom quote is flawed methodology. Transceiver specs vary widely — multimode, short-reach, long-reach, silicon photonics, EML (chips that use different materials to generate light) — and prices diverge accordingly.
In plain terms = someone took the cheapest variant's price tag and applied it to the whole industry. Management says that is apples-to-oranges.
Annual price adjustments are normal, but with demand strong and materials tight, top-tier suppliers have no incentive to slash prices recklessly.
How far out do orders reach?
Nearly all customers have placed 2026 orders. Some have already committed 2027 orders with monthly delivery volumes specified — not verbal guidance, but firm commitments.
Beyond that, select key customers have issued 2028 new-product demand guidance covering 2.4T, NPO (near-package optics — optical links built right next to the chip), and XPO products, with "very large" dollar amounts.
This means → if major customers did not intend to keep investing in AI infrastructure, there would be no reason to lock in new-product demand two years out. This stretches demand visibility from the usual one or two quarters to over two years.
How is the GPU-to-transceiver ratio changing?
Customer feedback shows the GPU-to-transceiver ratio is rising — from roughly 1:3 as previously understood, it has at least doubled and possibly more.
This reflects the growing scale of AI clusters: larger clusters need exponentially more connections between machines.
In plain terms = a single GPU used to need about three transceivers. Now it may need six or more. Transceiver demand is growing faster than GPU demand itself.
Will 1.6T competition get worse?
Management acknowledged that 800G volumes are large enough that some customers have added new suppliers alongside incumbents.
But 1.6T and higher-end products have long qualification cycles and steep technical barriers. No flood of new suppliers has emerged. The higher the speed, the fewer players can compete.
Even within 800G, a technology split exists: the 4×200G architecture (based on 200G-per-lane) commands higher prices and margins than the mainstream 8×100G approach. Innolight is already shipping it.
What will drive margin improvement?
Management expressed confidence in 2027 gross margins, citing solid pricing and demand in the current order book.
More importantly, as 2027 new products (2.4T, NPO, XPO) begin ramping, their margins will be meaningfully higher than legacy products — pulling up the company-wide gross margin.
This means → new products are not just a revenue add-on; they are a margin lever. The profit-improvement story hinges on product-mix upgrade.
How will optical connectivity's share of capex evolve?
In the traditional data-center era, transceivers accounted for less than 5% of cloud-provider capex.
As speed iterations accelerate and use cases expand from scale-out (rack-to-rack) to scale-up (chip-to-chip) and even intra-cabinet links, that share will keep rising.
In plain terms = even if cloud-provider total capex growth slows, optical connectivity is taking a bigger slice of the pie — giving the sector a growth story independent of overall AI investment growth rates.
The market's ultimate proof point: whether 2027 new products ramp on schedule and margin improvement shows up in reported earnings.
Content is for reference only, not financial advice.