400G reigns supreme in the data center

When it comes to data center optics and switches, 400G technology is king of the castle. But a series of industry experts told Silverlinings that while demand for more bandwidth – 800G and beyond – is on the horizon, hyperscale cloud players are looking to for more than just faster speeds in their next upgrade cycle. Energy efficiency is also a key priority, they said.

The data center market is comprised primarily of two segments: switches and optics. Dell’Oro Group VPs Sameh Boujelbene and Jimmy Yu explained that you can think of a switch as a box inside a data center that is responsible for networking and moving packets around. The optics, meanwhile, connect the switches to the network.

There are optics both within and outside the data center. Optics in the latter bucket (commonly referred to as Data Center Interconnect, or DCI) fall into two categories, including metro, which serves localized data center clusters, and long haul, which connects data centers further apart.

While that all may sound like hum-drum technical speak, there’s good reason for all the buzz around this market. The DCI segment alone was worth $2.5 billion in 2022, while the data center switch market totaled $17 billion. And cumulative spending on data center switches between 2022 and 2027 is forecast to top $100 billion, with cloud service providers tipped to account for 60% of switch sales by the end of the forecast period.

Put a different way, vendors in this space are chasing a lot of green.

Data center switch market

According to Boujelbene, 400-gig technology currently reigns supreme in the data center switch market. However, she noted all of the hyperscalers are at different stages of adoption.

Google was the first to take up 400G, starting its rollout in late 2018, early 2019. For them, Boujelbene said, the technology is now mature and Google is beginning to tackle 800G. They’re one of the only ones doing so, but Boujelbene said its 800G progress to date has been slow and it’s not clear whether it’s the 800G switches or optics that are the problem.

Amazon’s 400G switch rollout, which began in 2019, is winding down. Interestingly, Boujelben said the company seems to have decided not to move on 800G for now. That could be in part because cost premiums for the technology remain high and also because it views the 800G market as immature as of yet (see also, Google’s problems above).

Microsoft and Meta, meanwhile, are in the middle of their 400G switch upgrade cycle. They spent big in 2022 and are expected to continue doing so in 2023, Boujelbene said.

That said, 400G’s reign won’t last. Dell’Oro Group forecast 800G will overtake 400G as the dominant data center switch technology by 2025.

The optical realm

Within the optical realm, 400G is also the technology of choice, for now. Yu highlighted a trend toward 400G ZR coherent optics, which he said are beginning to replace older WDM systems in metro networks. The key benefit of these new optics is that they simplify the network by incorporating the transponder (the part that sends and receives optical signals) into a pluggable module. Older and long-haul systems still use external transponders.

Optical vendors Ciena, Infinera and Juniper Networks each confirmed this trend is very real and all three said it ties in to another trend in the data center: a focus on energy efficiency.

“By converging IP and optical networks and embedding optical transponders in routers and switches, pluggable optics are reducing costs and energy consumption compared to traditional optical networks,” Julius Francis, Senior Director of Product Marketing & Strategy for Security, Automation and Networking at Juniper Networks told Silverlinings.

Infinera SVP of Marketing Rob Shore and Helen Xenos, Senior Director of Portfolio Marketing at Ciena, noted it was actually hyperscalers who drove development of 400G ZR in the first place. Now there is a fairly large ecosystem supporting it.

“There’s such a focus on low-power for these devices, so around 15 watts or so for a 400 Gbps transponder, so that’s a big advantage of that type of pluggable,” Xenos said.

Shore added that in order for hyperscalers to “use pluggables the way they want, you need the smaller form factor and you need that into a much, much smaller power profile. So, they don’t perform that well compared to the embedded engines, but they’re ideal in certain applications.”

800G and beyond

But it doesn’t stop at 400G ZR optics. Francis noted said some hyperscalers are already driving the development of 800G and 1.6 Tb ZR pluggable optics “with the intention of pushing the envelope even further.” And as demand for artificial intelligence clusters grows, they’re also looking at “Linear Optics technology to simplify optics by eliminating DSPs,” Francis said.

Both Ciena and Infinera both already offer 800G optical products (though not necessarily ZR pluggables) and each recently crossed into the terabit realm. Infinera debuted optics capable of reaching 1.2 Tbps on a single wavelength, while Ciena rolled out a 1.6 Tbps product. Cisco and Nokia also play in the optical space and have 1.2 Tbps solutions.

However, going forward Shore said increases in speed may play less and less of a starring role as the industry reaches Shannon’s limit, or the maximum data rate achievable on fiber. In its place, energy efficiency could become the standout feature in future generations of optical products.

“Power costs have gone up substantially, so one of their key elements is helping to reduce power,” Shore said of optical engines. “It’s a recognition that really the current trajectory of power consumption is not sustainable.”

Want to learn more about data center and optics? Don't miss our upcoming Cloud Data Center Strategies virtual event set for May 31- June 1. Register here.