CPO Fears, Credo, and Role of AEC in 1.6T Networking
A rational, engineering-centric approach to understanding the future of active electrical cables in the era of CPO and 1.6T datacenter networking.
There is a lot of fear in the markets that the adoption of co-packaged optics (CPO) in the future will take away noticeable market share from active electrical cables, leaving companies like Credo (CRDO 0.00%↑) with a lower share of the copper interconnects market. In this post, we will take a rational, unbiased, data-driven approach to exploring the relevance of AEC in the dawning era of CPO.
If you need a primer on interconnects in datacenters, check out the post below. Thanks to Collyer Bridge for suggesting the topic for this post: you should check out their rapidly growing Substack publication.
First, let’s look at some recent statements from industry that provides an outlook for copper cables as CPO looms on the horizon. From Credo’s CEO Bill Brennan from Barclay’s GTC:
I don’t necessarily think that people will say, I need CPO because of the inability to do it the same way we’ve been doing it for 10 years. There’s nothing that stands out. You can’t point to cost for sure, it’s going to be more expensive. You can’t say reliability because it’s going to be far less reliable, given the fact that it’s laser-based.
At first look, this seems broadly dismissive of the fact that CPO is becoming increasingly important as interconnect speeds get faster, and sweeping assumptions that CPO is inherently unreliable because “lasers”, especially when solutions exist to this problem.
Instead, Brennan sees other approaches. He goes on to say that technologies such as near package optics with microLEDs offer a lower cost, practical approach to scale-up/scale-out interconnects compared to CPO. Credo is betting on active LED cables (ALCs) given their recent acquisition of Hyperlume.
The micro LED effort that we’re putting into ALC, there’s a direct application for near package optics that’s 1/3 the power of CPO. And it doesn’t come with such exotic switch design as you see today, when you see at conferences (sic). And so we think we’ve got a path that’s much better when the industry finally gets there.
LEDs are not without their challenges. Credo is keenly aware of the problems with GaN microLEDs as seen from their recent job posting for a “Device Analysis Manager” to “lead the development of physics based accelerated lifetime models for core optical components in Credo LED cable products …” I’ll do a deeper dive into microLEDs on this newsletter in the future.
What “exotic switch design” refers to is unclear but high throughput network switches is what hyperscalars want, according to Hock Tan at Broadcom’s recent earnings call:
Our current order backlog for AI switches exceeds $10 billion as our latest 102-terabit per second Tomahawk 6 switch, the first and only one of its capability out there continues to book at record rates.
Tomahawk 6 (TH6), code-named Davisson, is the fastest CPO-based network switch available in the market today at 102.4 Tbps aggregate switch bandwidth, and is primarily aimed at scale-out applications. When Hock was asked what he thinks of Marvell’s acquisition of Celestial AI for the photonic interconnect fabric, here is what he had to say:
So when you talk about your next question of silicon photonics as a means to create basically much better, more efficient, lower power interconnects in not just scale-out, but hopefully, scale-up. Yes, I could see a point in time in the future when silicon photonics method is the only way to do it. We’re not quite there yet.
… try to do scale-up within a rack in copper as long as possible and in scale-out, non-pluggable optics.
The picture is starting to come together: copper for scale up, optics for scale out. This is pretty much how it always has been: copper when you can, optics when you must. The need to switch to optics for scale up is not imminent. MicroLED-based approaches are still largely an unknown at scale; we will have to wait and see how it plays out.