Bandwidth is per-lane speed times lane count. Push speed and you hit the SerDes energy floor; push lane count and you hit fiber and connector cost. The new fact this year: the second wall arrived first.

Scale-up optical interconnect looks like a seven-way melee, but there are really only two strategies, Fast-and-Narrow and Slow-and-Wide, each hitting a symmetric wall. Learn the yardstick first, then see how the seven approaches layer and coexist.

Tackling both the power and capacity bottlenecks of GPU scale-up, Lumentum taps the mature 3D sensing and LiDAR supply chain to redefine high-density interconnect with 1060nm VCSEL arrays.

Three scale-up MSAs launched at OFC 2026, and the loudest, XPO, was championed by Arista co-founder Andy Bechtolsheim. But Meta's 90 million hours of data took apart its core selling point. Is XPO a bet on extending pluggables' life, or the real answer?

Corning's quarterly optical communications revenue topped $2 billion for the first time, with Enterprise up 65% YoY. The real story: management has written scale-up into the business plan — fiber content per GPU going from 16 to 160 fibers, plus a new $10 billion photonics platform by 2030. Why does Corning dare say it will "grow faster than GPUs"?

850 nm is being pushed off the AI scale-up shortlist not because of speed but because of junction temperature. Moving to 1060 nm isn't swapping a die — fiber, PD, packaging and standards all have to change together.

Celestica grew revenue 62% and EPS 83% YoY with a record operating margin, raising full-year guidance for the second time. The real signal: OpenAI custom rack orders surfaced, AMD Helios scale-up became a multi-billion-dollar pipeline, and 2027 growth is set to be even faster.

With 104% YoY growth and Scorpio reaching the top a quarter early, Astera declared its promotion from connectivity sidekick to AI Fabric supplier. The real signal isn't one quarter's numbers but dollar content per XPU heading toward "thousands of dollars." STT flags the three numbers to watch next quarter.

The value of optical interconnect isn't saving power; it's unlocking the physical boundary of the pod. Astera Labs laid out the scale-up fabric bottleneck plainly at OCP APAC, and its candid Q&A reservation on CPO serviceability is where Taiwan's supply chain has a real opening.

Token demand wants 10x a year while supply delivers only 2x, and the compounding 5x gap can only be closed by architecture. Ayar Labs CEO Mark Wade's key line at OCP APAC wasn't about CPO but this: the unit of adoption shrinks from a whole rack to a single chassis.

Optical circuit switching has stayed out of scale-up not because it fails to save power, but because it is two to three orders of magnitude slower than AI communication bursts. At OCP APAC, Linque showed a 64-channel silicon photonics OCS reconfiguring in 15 µs, clearing the “switch is too slow” hurdle. The other half—cutting transceiver lock time from milliseconds to nanoseconds—is the real battleground.

Broadcom's CPO session skipped the theory and laid out field failure data from three product generations, because what holds CPO back was never power, but how to fix it when it breaks. With reliability now the selling point, the OCI spec and a 2027 fourth generation are opening three orders for Taiwan's supply chain.

Broadcom's "Open Ethernet" keynote talked standards on the surface, but really declared that Ethernet will take over in-rack scale-up, NVLink's deepest moat. Open specs grow the ecosystem while in-house silicon collects rent, and behind this two-layer play sits a whole new layer of demand for Taiwan's optical interconnect and CPO supply chain.

NVIDIA posted $96.2B in quarterly revenue, up 106% YoY, but the real signal is that its take per 1 GW of data center jumps from $25B to $40B, with the increment all in CPUs and networking. Non-hyperscale customers grew 138% YoY and NeoCloud capacity is heading from 3 GW to 8 GW in a year — that's the order pool for the optical supply chain. The cost: gross margin troughs at 71–72%.

The CPO technology debate is over; the race is now about yield, test and delivery. With SCALE, GlobalFoundries turns the optical engine into a fully tested semiconductor part built to last a decade, targeting 20–30% of pluggable cost. The real variable isn't optics — it's interoperability standards that have yet to converge.

AMD has officially shifted its center of gravity from "PC + server" to the AI data center, with Data Center alone taking 58% of revenue. The real signal is the 2027 guidance - Data Center more than doubling, server CPU up over 70% - and Helios pins down the timeline for optics entering the rack. For the optical supply chain, demand visibility now extends into 2027.

Compute grows 75% a year; the copper feeding it grows only 20%. That scissors gap is the root of AI infrastructure pain. Winzer uses six figures to complete the causal chain for linear optics, and lands on a sharp conclusion: this generation's ceiling is the 1 meter of scale-up copper.

AMD's OCP keynote was framed around open standards, but the line worth underlining is the CPU-to-GPU ratio converging from 1:4 toward 1:1. Agentic workflows hand orchestration, retrieval and tool execution back to the CPU — and that matters far more to Taiwan's rack-integration and scale-up optical interconnect supply chain than a slide full of open acronyms.

On the surface, Microsoft's keynote showcased in-house silicon; underneath, it publicly moved the AI infrastructure bottleneck from accelerators to power, cooling and networking. For Taiwan's supply chain, Microsoft's own "70% of components come from Taiwan" is an entry ticket, not a moat: the money is flowing to system-level co-design.

The market loves to frame scale-up as a copper-vs-optical death match; SemiAnalysis calls it a false choice. The real inflection is the HBM downgrade: the savings flow into scale-up networking, memory's share of TCO drops from 41% to 28%, and scale-up network cost triples. Copper holds inside the rack, optics bridges racks, and both ramp together.