Slow-and-Wide Won. microLED Did Not: The Company That Bought a microLED Start-Up Shipped Its First Product With VCSELs
microLED Optical Interconnects: The Complete Picture | November 2026
Two things happened at the OCP Global Summit in October 2026.
First, OCP founded a standards workstream called SRMO (Short-Reach Micro-Emitter Optics): nine founding members plus an unnamed cloud operator, with the spec targeted for 2027.
Second, Credo launched the 1.6T ZeroFlap active light cable: up to 50 m, about 12 W per end, no DSP, shipping in 2027.
Credo is the company that acquired microLED start-up Hyperlume in 2026. And its CEO said it plainly on stage: this product uses micro VCSELs, not microLED.
STT's read: the architecture won, not the light source. Slow-and-wide has gone from a questioned approach to an architecture with a standards workstream and a shipping schedule, but the first component filling that architecture is a microVCSEL. The spec itself deliberately groups LEDs and VCSELs together as “micro-emitters,” which demotes the light source from a roadmap fight to an application choice. And Credo's own platform table from the same session puts red microLED at 2 m or less, serving in-package scale-in only, while microVCSEL reaches 50 m and covers all three layers — the first product had to cross a rack, and microLED was never in that distance slot to begin with. This is not “microLED got replaced”; it is “this generation's slot is not its slot.”
▲ The architecture won, not the light source: of the five stages in the link, the two candidate emitters differ only in the first, and the other four are shared. (Image source: Simple Tech Trend original)
What the Paid Section Covers
Architecture and light source are separate questions: the five stages of the link differ only in the first, and why the one that shipped first is not the one carrying the bigger bet
Copper's reach collapse and the iso-bandwidth line: how 1.2 m is derived, and whether the same 1.6T takes eight lanes or five hundred
The chain reaction of removing the cavity: the seven parts that disappear, one by one, and the bill that comes with them
The fibre that came from endoscopes: four design requirements, four unknowns, and “sixty thousand fibres on a single floor”
Yield is three different numbers, plus the unsettled bill of the reverse GearBox: two charts that say the opposite, with neither side lying
The eleven-item comparison: VCSEL wins six, microLED wins one, and why that one item is the whole bet
Seven bottlenecks in two classes, the six events of Q4, and the three bindings between scale-in and HBM
Ten Taiwanese suppliers, three layers of parts, two windows, plus ten counterarguments and eight signals to track
That is the summary of this piece.
The full read — copper's reach collapse and the arithmetic of slow-and-wide, the two classes of bottleneck, what each of the six Q4 2026 events actually moved, the one slot microLED still owns, and the three layers and two windows for Taiwanese suppliers — is in the paid section: roughly 23,000 Chinese characters and 17 original charts.
Subscribe to STT Premium to read the full analysis on vocus (in Chinese): microLED Optical Interconnects, full edition
Comments