Hardware for the Edge

Does a brand new market need new hardware? We were recently asked what kind of new technology do we need for Edge Computing.

As we wrote back in December, Edge Computing is the idea that we put high powered computer servers somewhere in between cloud data centers and users. Our conclusion in that piece is that Edge Computing matters a lot but to only certain people and industries.

Many of the most avid promoters of Edge Computing are the hardware and chip makers. Some of this is wishful thinking. These companies are increasingly squeezed selling more to a shrinking pool of customers. The top seven cloud service providers consume 50%-70% of high end compute and networking gear. Amazon alone probably buys 20%-30% of the industries’ output. It would be nice for these customers to have a new growth segment, especially one that is less concentrated.

In fairness, this is going to be real market and it is worth considering what technology is needed to serve it. Do we need new products for Edge Computing?

The short answer is no. But it is worth walking through the stack to uncover what opportunities there are.

Semiconductors – The category is not big enough nor distinct enough to merit a new class of semiconductors. These boxes will be filled with CPUs and GPUs just like severs in data centers. The one important distinction is that these boxes will likely need to pay retail for electricity, meaning they will be much more expensive to operate on a per unit basis than data center servers. In theory, this will favor Arm CPUs which offer much better performance per watt. And we have seen some of the nascent edge compute providers like Cloudflare dip their toe in that pool.

The good news for the chip makers is that this could be a fairly sizable market. Rough math, there is probably sufficient demand for 100 edge data centers in 200 cities around the world. Let’s say each one of those has 50 racks of servers, with 20 CPUs in each rack. That works out out to 20 million CPUs, divided by 5-year deployments is 4 million chips a year. That is roughly equivalent to two cloud providers, so decent business but not game changing.

Networking Gear – Networking may need a slightly different set of equipment than traditional data centers. Not so much brand new products, just a nice mid range product line. For networking equipment makers this is a decidely mixed bag. There is decent volume here, but only for a relatively unexciting portion of their portfolio. Top of rack switches in all those racks and some kind of middling switch to connect back up to the core Internet. Making matters worse, the kinds of companies that are going to build edge data centers likely operate a lot of other network gear and thus more comfortable using commodity off the shelf hardware running some fancy orchestration software.

Software – The one area which might be more interesting is software, or some kind of clever bundle of software and networking hardware. The whole networking industry from Cisco on down is generally moving in this direction already.

In particular, we think edge data centers are going to favor particular industry verticals which leaves room for some vendor to deliver a specialized solution. We revert to our favorite use case here, gaming. Gaming networks need to balance users across the full range of distances and last-mile connections. Latency matters a lot. The traffic is also widely scattered, one server connecting to thousands of end points. This will require a fair bit of load balancing, possibly good for a company like F5. Alternatively, some company will create some specialized software and then stitch together all the required underlying hardware pieces.

Which leads to the ultimate question hanging over every company looking at edge data centers. Will this become just another feature from a cloud service provider? The answer for many is probably yes. AWS, GCP and Azure already have widespread distribution in metro areas sufficient to roll out a credible edge solution today. True, they are not as close to the end user as they could be, but they are already a large portion of the way there.

That being said, one of the interesting things about edge computing is that for many use cases to make commercial sense they need to have global availability. Back to our gaming example, there are dozens of important gaming communities that are far away from Amazon regions (Brazil, South East Asia, Eastern Europe). We can think of many other industries that need solutions beyond what the hyperscale cloud providers can offer today or really any time soon.

Our sense is that there is a sizable market opportunity in Edge Compute. The key will not be shiny new technology, but instead a clever solution tailored for specific industries.

<a href="http://<span>Photo by <a href="https://unsplash.com/@scottwebb?utm_source=unsplash&utm_medium=referral&utm_content=creditCopyText">Scott Webb</a> on <a href="https://unsplash.com/s/photos/infinity-pool?utm_source=unsplash&utm_medium=referral&utm_content=creditCopyText">Unsplash</a&gt;Photo Credit – Scott Walsh via Unsplash

Leave a Reply