UBS recently discovered at a semiconductor forum that against the backdrop of soaring AI data center power consumption and increasingly evident copper cable transmission bottlenecks, Co-Packaged Optics (CPO) is becoming a critical pathway for the next phase of AI infrastructure upgrades.

On September 13th, UBS stated in its latest research report that the Co-Packaged Optics (CPO) technology trend has become "particularly clear" this year, with industry experts predicting that power consumption optimization will reach a critical juncture in 2028.

The research report noted that statements from three experts at the semiconductor forum - Broadcom, Ayar Labs, and Lumentum - indicate that CPO is gradually breaking through power consumption and integration bottlenecks, potentially enabling large-scale applications in AI server interconnects by 2028-2029.

Broadcom stated that the key factor currently preventing CPO from fully replacing copper cables is that its energy consumption remains above 10 picojoules per bit (pJ/bit). However, they expect mid-2028 solutions to drop below 10 pJ/bit, with advanced solutions in 2029 potentially reaching as low as 5 pJ/bit. At that point, CPO will truly possess large-scale deployment capabilities.

Ayar Labs emphasized that the sharp rise in AI data center power consumption is driving CPO demand, planning to limit rack power consumption of 576 GPU systems to within 100kW through optical I/O solutions. Lumentum indicated that leveraging its technical advantages in compound semiconductor materials and ultra-high power lasers, it is paving the way for next-generation 3.2T optical interfaces.

**Broadcom: CPO Energy Consumption to Drop Below 5pJ/bit by 2029, Copper Cables Approaching Critical Limits**

Broadcom Chief Technology Expert Tzu Hao Chow stated that "scale-out" networks for AI servers have widely adopted pluggable optical modules and are gradually transitioning to CPO. "Scale-up" networks still rely on copper cables but face significant physical limitations - when single-channel rates increase from 100G to 200G, copper cables' effective transmission distance shortens from 4 meters to 2 meters, with extremely limited improvement even when adding repeater chips.

Chow pointed out that the key factor currently preventing CPO from fully replacing copper cables is that its energy consumption remains above 10 pJ/bit. However, they expect mid-2028 solutions to drop below 10 pJ/bit, with advanced solutions in 2029 potentially reaching as low as 5 pJ/bit. At that point, CPO will truly possess large-scale deployment capabilities.

He also compared two optical module technologies developed by Broadcom: Silicon Photonics (SiPh) and VCSEL solutions.

SiPh offers advantages in bandwidth, transmission distance (up to 2 kilometers), and power consumption, making it suitable for data center interconnects, while VCSEL still maintains a position in cost and short-distance transmission (within 30 meters).

**Ayar Labs: AI Data Center Power Consumption Explodes, CPO is the Only Path**

Ayar Labs founder Mark Wade directly addressed the pain point: "Meta's new Hyperion data center will consume 2GW by 2030, rising to 5GW by 2035, equivalent to 1.5 times the residential electricity consumption of the entire state of Louisiana."

He stated that the exponential growth of AI training tasks has made data center rack power consumption a bottleneck constraining expansion. Ayar Labs' CPO solution, including TeraPHY optical I/O chiplets and SuperNova multi-wavelength light sources, will enable:

- Single-rack deployment of 576 GPU systems with power consumption controlled within 100kW (2027) - Even when expanded to 2048 GPUs, maintaining single-rack power consumption stable within 100kW (2029)

Additionally, Wade believes AI hardware is transitioning from the "narrative" to the "plumbing" stage, meaning the infrastructuralization of underlying interconnect and power systems. He emphasized that optical I/O technology will simultaneously improve AI computing efficiency (tokens/sec) and cost ratio (throughput/$), truly achieving a win-win in performance and profitability.

Currently, Ayar has received investments from AMD, NVIDIA, Intel, and TSMC, and is collaborating with Alchip to promote optical integration on TSMC's COUPE platform.

**Lumentum: VCSEL Technology Mature, Paving the Way for 3.2T Optical Interfaces**

At the device level, Lumentum Chief Photonics Expert Matt Sysak emphasized that the company has deep expertise in compound semiconductor materials from gallium arsenide (GaAs) to indium phosphide (InP), and can provide core components including MEMS optical switches and high-speed lasers.

NVIDIA is adopting Lumentum's ultra-high power lasers as the optical engine for its Spectrum-X CPO switches, marking that traditional GPU giants have fully introduced optical technology into network infrastructure.

Lumentum's latest PAM4 modulated lasers (400G/448G/450G) and distributed feedback modulators (DFB-MZI) are laying the foundation for next-generation 3.2T optical interfaces.

Meanwhile, its VCSEL lasers, which have shipped over hundreds of billions of units in FaceID applications, are also demonstrating high reliability and low-cost advantages in server clusters, suitable for short-distance large-scale interconnects.