Up 115%, Short 70%: Goldman Lifts the 2028 Optical Module Market to $148.5B, but No One Is Doing the Math on the Denominator - 2026 W38 Industry Weekly (09/07-09/13)
1. On 9/7 Goldman Sachs raised its optical module forecast for the third time: 2026-28 market size up 33% / 81% / 115% to $67.7B / $131.4B / $148.5B; yet shipments over the same years were raised only 21% / 31% / 31%.
2. Divide one set of numbers by the other and implied ASPs are raised by about 10% / 38% / 64% - nearly two-thirds of this upgrade isn't "selling more units" but "each unit being pricier and higher-end".
3. The same week, DigiTimes reported that the InP substrate supply-demand gap exceeds 70%, and Lumentum said AI laser demand exceeds supply by more than 30%.
4. So the question isn't the ceiling - it's the denominator.
1. This Week, the Demand Side and Supply Side Turned In Two Charts Pointing in Opposite Directions
This week's information density isn't in any single event but in "two things happening at the same time".
On the demand side, Goldman Sachs made its third upgrade of the year to optical communications on 9/7, lifting the ceiling for the 2028 market to $148.5 billion. The same week, Broadcom's earnings call laid out AI semiconductor visibility of $115 billion for FY27 and $230 billion for FY28, Tomahawk 7 (the industry's first 200 Tb/s Ethernet switch) taped out, and Oracle's RPO backlog surged to $664 billion.
On the supply side, DigiTimes gave a very hard number on 9/9: effective global capacity of optical-grade Indium Phosphide (InP) substrates in 2026 is estimated at only 600-750 thousand wafers/year against demand of 2.6-3.0 million wafers/year - a gap of more than 70%. Lumentum said AI laser demand exceeds supply by more than 30%. Prices of 2-inch optical-grade InP substrates rose from $800/wafer in early 2025 to $2,300-2,500/wafer in April 2026, up nearly 200% in two years.
One side is raising the ceiling, the other is crying that it can't get materials. This isn't a contradiction - it's two facets of the same thing.
2. Demand Side: The Star of This Upgrade Isn't Volume, It's Price
Item | 2026 | 2027 | 2028 |
Market size ($100M) | 677 | 1,314 | 1,485 |
Market size revision | +33% | +81% | +115% |
Total shipment revision | +21% | +31% | +31% |
800G+ shipment revision | +31% | +39% | +36% |
1.6T+ shipment revision | +29% | +61% | +50% |
Most coverage only copied the "up 115%" line. But divide the market-size revision by the shipment revision, and you see what the report is really saying:
Implied ASPs are raised by about 10% for 2026, about 38% for 2027 and about 64% for 2028.
(Method: taking 2028 as an example, revised market size is 2.15x the original estimate and shipments 1.31x; 2.15 / 1.31 ≈ 1.64.)
Goldman didn't materially change its view of "how many modules will ship"; what it changed is "what those modules will look like". 1.6T and above was raised 61% for 2027, the largest of all line items.
The distinction matters because volume problems and spec problems get stuck in completely different places: volume is gated by packaging lines and contract-manufacturing capacity, which can be built out with money and time; specs are gated by epitaxy, substrates and yield, where you have to wait in line.
The same week, Broadcom's Tomahawk 7 (100T -> 200T) doubled per-rack optical-layer bandwidth demand again, and that extra doubling won't be filled with 800G; Ciena's FY26 Q3 revenue was $1.67 billion, up 37% YoY, with backlog up $800 million QoQ to $8.5 billion. Several lines are all pushing demand toward higher data rates.
3. Supply Side: The Scarcest Material Happens to Be the Main Cost of Moving Upmarket
InP is the base material for all EMLs (electro-absorption modulated lasers) and CW lasers. Pluggables need it, and silicon photonics needs it even more - silicon itself doesn't emit light, so SiPh solutions must attach an external CW laser, extending InP demand from the modulator side to the light-source side as well. The higher the data rate and the higher the SiPh penetration, the more InP per unit - not flat, but rising.
This week's supply-side facts were highly concentrated:
• More than 90% of global InP substrate capacity is concentrated in three companies: Sumitomo Electric, AXT and JX Advanced Metals.
• Effective capacity of 600-750 thousand wafers/year in 2026 against demand of 2.6-3.0 million wafers/year - a gap of more than 70%.
• 2-inch substrates are up nearly 200% in two years; high-end 6-inch prices are up more than 250%.
• Lumentum's fifth InP fab (Greensboro, converted from GaAs to InP) will contribute revenue in about six quarters - end of 2027 at the earliest.
• Coherent is advancing volume production of 6-inch InP wafers, with roughly four times as many dies per wafer as today.
• LandMark Optoelectronics (3081) is in volume production of 1.6T SiPh CW lasers, with 3.2T sampling to customers, and plans to increase 2027 capacity by about 1.5x over 2026.
We walked through the context of this section once in You Can Ban the Modules, Not the Substrates: Optical Communications' Decisive Move Has Shifted Down to InP; this week's 70% gap figure from DigiTimes is the quantified version of that piece. The part about the bottleneck shifting from materials to packaging precision is in last week's TSMC Places the Bottleneck Outside Itself: CPO's Decisive Move Shifts to Coupling and Test.
That concludes the key takeaways of this article.
STT's full analysis - why this upgrade is really a "supply tightness report", why 1H27 InP supply is already locked up, how 6-inch yields and the timing of Lumentum's fifth fab decide 1.6T penetration, the real reach of the FCC "component carry-along" clause, and who gets the biggest share of this denominator - is available in full in the premium section.
Subscribe to STT Premium - read the full analysis on vocus (in Chinese): https://vocus.cc/article/6aa74be1fd897800013c1faf
This article is for technology and industry trend analysis only and does not constitute investment advice.
Comments