UBS Semiconductor Forum: CPO trend is "especially clear" this year; Broadcom says "power consumption will be fully optimized by 2028, replacing copper connections"
```
After attending a recent semiconductor forum, UBS found that under the context of skyrocketing AI data center power consumption and increasingly apparent copper cable transmission bottlenecks, co-packaged optics (CPO) is becoming the key path for the next stage of AI infrastructure upgrades.
On September 13, according to Tracking Trader, UBS stated in its latest research report that the trend of co-packaged optics (CPO) has become "particularly clear" this year, and industry experts predict that power consumption optimization will reach a critical point in 2028.
The report pointed out that the speeches by three experts from Broadcom, Ayar Labs, and Lumentum at the semiconductor forum showed that CPO (co-packaged optics) is gradually overcoming bottlenecks in power and integration, and may enter large-scale deployment for AI server interconnects between 2028 and 2029.
Broadcom stated that the key factor preventing CPO from fully replacing copper cables is that its energy consumption is still higher than 10 picojoules/bit (pJ/bit), but it is expected that solutions in mid-2028 will bring it below 10 pJ/bit, and by 2029 advanced solutions could reach as low as 5 pJ/bit. By then, CPO will truly be capable of large-scale deployment.
Ayar Labs emphasized that the sharp rise in power consumption in AI data centers is driving demand for CPO. They plan to use optical I/O solutions to keep the rack power consumption of 576-GPU systems below 100kW. Lumentum stated that by leveraging its technological advantage 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 are at their limit
Tzu Hao Chow, Chief Technology Expert at Broadcom, said that “horizontal scaling” in AI server networks has widely adopted pluggable optical modules and is gradually moving toward CPO. Meanwhile, “vertical scaling” still relies on copper cables but faces significant physical limits—when single-channel speed increases from 100G to 200G, the effective transmission distance of copper cables shrinks from 4 meters to 2 meters. Even if you add relay chips, the improvement is minimal.
Chow pointed out, the key preventing CPO from fully replacing copper cables is currently that its energy consumption is still greater than 10 pJ/bit, but solutions are expected to bring this below 10 by mid-2028, and advanced solutions could lower it to 5 pJ/bit by 2029. At that point, CPO will truly be capable of large-scale deployment.
He also compared Broadcom’s two proprietary optical module technologies: the silicon photonics (SiPh) solution and the VCSEL solution.
SiPh has advantages in bandwidth, transmission distance (up to 2 km), and power consumption, making it suitable for data center interconnects; VCSEL still has a place in cost and short-range transmissions within 30 meters.
Ayar Labs: AI data center power consumption off the charts, CPO is the only path
Ayar Labs founder Mark Wade tackled the pain point directly: “Meta’s new Hyperion data center will consume 2GW of power by 2030, rising to 5GW by 2035, equivalent to 1.5 times the total residential power usage of the entire state of Louisiana.”
He stated that the exponential growth of AI training tasks makes rack power consumption the bottleneck restricting 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 limited to 100kW (by 2027)Even when expanded to 2048 GPUs, single rack power can still be kept within 100kW (by 2029)
Furthermore, Wade believes AI hardware is moving from the “storytelling” stage to the “plumbing” stage, meaning foundational interconnect and power systems. He emphasized that optical I/O technology will simultaneously boost AI computing efficiency (tokens/sec) and cost-performance ratio (throughput/$), truly achieving a win-win for performance and profitability.
Currently, Ayar has received investment from AMD, Nvidia, Intel and TSMC, and is partnering with Alchip to advance optical integration on the TSMC COUPE platform.
Lumentum: VCSEL technology is mature, paving the way for 3.2T optical interfaces
At the device level, Matt Sysak, Chief Photonics Expert at Lumentum, stressed that the company has deep experience in compound semiconductor materials such as gallium arsenide (GaAs) and indium phosphide (InP), and can provide key devices such as MEMS optical switches and high-speed lasers.
Nvidia is using Lumentum’s ultra-high-power lasers as the optical engine for its Spectrum-X CPO switches, marking the comprehensive introduction of optical technology into network infrastructure by traditional GPU giants.
Lumentum’s newly released PAM4 modulated lasers (400G/448G/450G) and distributed feedback modulators (DFB-MZI) are laying the foundation for the next-generation 3.2T optical interfaces.
Meanwhile, the company’s VCSEL lasers, shipped in quantities exceeding tens of billions for FaceID applications, have demonstrated high reliability and low cost advantages in server clusters, making them suitable for short-distance, large-scale interconnects.
~~~~~~~~~~~~~~~~~~~~~~~~
The above excellent content is from Tracking Trader.
For more detailed interpretation, including real-time analysis and firsthand research, please join [Tracking Trader ▪ Annual Membership]
Risk Warning and DisclaimerThe market has risks, investment needs to be cautious. This article does not constitute personal investment advice and does not take into account the individual user’s investment objectives, financial situation or needs. Users should consider whether any opinions, views or conclusions in this article are suitable for their particular circumstances. Investment based on this article is at one’s own risk. ```