Skip to content

Search the site

Bare Metal-as-a-Service: Is this co-lo niche a democratising force in a cloud oligopoly?

We spoke with Packet founder Jacob Smith.

“Physical infrastructure and networks are increasingly important to everything that you do up the stack” notes Jacob Smith. “This foundational layer of infrastructure is evolving really quickly. Silicon's changing; people are getting into machine learning; there's fancy offloads; new kinds of security. But the only companies who can really consume this and ship it and do it their way are the top seven or eight companies in the world.”

If you're not in this tier, you run the risk of getting outspent and stomped on commercially in the long run, is the subtext. Miss a hardware upgrade cycle in certain financial services categories for example, and you are likely to get outperformed vary fast indeed: if not by a deeper-pocketed rival with their own data centres, then by a cloud-provider muscling into your space -- with its own extensive suite of IaaS and SaaS tools at its disposal.

Smith's talking to The Stack as the co-founder of Packet, a "bare metal automation platform" that gives users the ability to deploy and manage hardware on-premises or at edge sites/in global co-location data centers within minutes. It's a fascinating category -- somewhere between cloud and traditional co-location -- that Smith thinks is going to become increasingly important as enterprises think through their infrastructure strategy.

Equinix Metal: "Particularly interesting for those who are trying to win"

Open source-powered Packet was snapped up by data centre giant Equinix for $335 million in early 2020 and now represents a way of deploying bare metal servers on an opex basis across Equinix's huge global footprint).

The tool (which lets enterprises click-and-configure, and provision physical infrastructure via API and common libraries) represents a fundamentally democratising force in a concentrated cloud oligopoly, Smith argues.

As he puts it: "Equinix Metal, which is the product we launched, looks a lot like the benefits of the cloud. It's opex-based. It's consumption-based. It's automation-friendly, so developers are happy, but it's also single tenant; it's physical; you can run your VMware on it; it looks like on-prem. So that for that bucket of customers who are transforming and trying to get out of their own data centers it's a good fit -- it's particularly interesting for those who are trying win: working on projects where they need to put infrastructure globally; do it at lower cost, higher performance, or in regulated environments, or with special hardware for streaming, etc."

See also: Introducing Apache AGE

For those pushing into new markets globally the platform can be particularly useful, he suggests: letting central development teams spin up dedicated servers at the level they want in one of Equinix's 220+ data centre locations, without having to go knocking on doors and opening racks.

He says: "Traditionally to spin up bare metal of your own in a co-location environment you've got to get a rack, get a cage, get everything together in the traditional kind of co-location enterprise model. That works for some people are really good at it, like, say, Akamai. But it doesn't work great for, Deutsche Telekom. You know, 'I'm trying to do this globally and I don't have the stuff; I don't have the people there; I need it done faster...'"

Equinix claims to have got most installs down to 60 seconds; from gaming servers to full Kubernetes clusters for "bare metal at software speeds" across one of the world's largest network of interconnected data centres.

Is there a tension between trying to wrestle people away from managing their own data centres/racks in co-location data centres and them -- once making the decision to do something at arms-length -- simply opting for cloud bare metal instances instead? How does Equinix Packet see itself nestling in that sweet spot between these two extremes?

A curated selection of bare metal server options via Equinix Metal: customers can opt for on-demand, reserved, or spot capacity, and heavily customise the above as required.

The Packet co-founder, who brought his team to him with Equinix, says: "There's use cases like someone in financial services saying 'I have to be in NY4 not NY5 because I'm a high frequency trader'. Well great: co-lo! Please buy the thing and do it yourself in the place that you need it with the things that you need and the specialised stuff that you need. There's always going to be a space for that and Equinix does co-lo very well.

"I think that just gets more niche over time however. Those doing it end up becoming service providers: internal service providers like Bloombergs and Nasdaqs; even Netflixs and Akamais; as you know, they run their network nodes... I think that gets very specialised. I think the issue is, is can we make it more consumer friendly? Right now though, it's actually very hard to do this stuff. But if we can get taxis in five minutes on our mobile phone, we can probably help people get hardware in Hong Kong, you know?

Pick and provision your servers...

More personally, how's it been going from open source-hearted scaleup, to being owned by a leviathon, The Stack wonders?

"They are a company in digital transformation that is also trying to help companies in digital transformation. I often turn to our team and say 'look at Milind [Equinix's global CIO]. Could he buy our product? What would he do with it?' Being inside of that is really instructive."

He adds: "You know, I think this year is going to be big for appliances, and for getting on-prem cool again; whether it's your factory floor, your co-location. The ecosystem has had to solve that problem form a capex, flexibility, deployment, operations perspective. But it's just like in the software world: you know, three years ago you had to be a rocket scientist to use Kubernetes. Now's it's increasingly straightforward. I think that's what we need to do and are going to do for physical infrastructure."

See also: Slack points finger at AWS bottleneck after major outage in early January triggered firefighting.