Network spending overwhelmingly favors incumbent network vendors, as any enterprise will tell you. Network modernization spending, which currently accounts for about 90% of what enterprises spend on network gear, doesn’t come about as a big fork-lift of an entire network, but rather a gradual replacement of boxes that have been fully written off and are getting old, limited in capacity, and expensive to repair and operate. The easiest replacement to make is to just switch to the newer product of the same vendor, and hey, who doesn’t want easy? This sort of purchase is also less risky; there are no new relationships to learn and to manage, and a no-risk decision is good, too, right?
How does a network startup fit in, then? Consider what offsets ease and lack of risk: Money. If new network equipment can save a lot of money, or if it’s needed to support a new application or service that improves productivity or lowers overall costs a lot, then it’s worth looking into. Of 321 enterprise IT professionals who have offered me comments on network startup vendors, 303 said that money is the only reason to run to a startup, and the rest said it is the most compelling reason.
Startups, and the venture capitalists that back them, aren’t stupid either, so we can already see examples of how network startups address this money focus. There are two primary paths: lower costs radically with a revolutionary design and focus on network impacts of new computing strategies. A great example of both can be seen in the software-defined wide-area network (SD-WAN).
SD-WAN vendors push into security
When SD-WAN first came along, it was directed purely at network service costs. Business networks used specialized technology (digital trunks, carrier Ethernet) for their access connections and MPLS VPNs for their WANs, when consumer-dominated Internet connections were (according to the enterprises) an average of one-eighth the cost while offering double the capacity. SD-WAN built an entire business network, or added “thin location” sites to traditional ones, using the Internet, so startups in the SD-WAN space succeeded in making their cost-savings business case to enterprises.
When cloud computing and multi-cloud came along, many of the SD-WAN startups recognized that the cloud needed a new networking approach to be optimally efficient, and they jumped on that mission. At the same time, startups saw that security concerns (everyone who thinks the Internet is secure, raise your hands!) so SD-WAN started pushing security. For the cloud this resulted in the secure-access service edge (SASE), which aggregates SD-WAN and access security tools into a cloud-hosted form.
The hottest area in SD-WAN today is multi-cloud, and that’s an important indicator of how enterprises look at network startups and also a demonstration that every enterprise network startup opportunity is a flash in the pan. Only 44 of the 321 enterprises I’ve chatted with on network startups said they were multi-cloud users, while every one of them was a VPN/carrier-Ethernet user and 302 were users of cloud computing. Why jump onto such a small target? Because incumbents, having bought up SD-WAN startups, were now dominating the space.
So, what are the areas where enterprises think a network startup could win their business today? That depends on how enterprises view risk and reward.
The space where enterprises today are least satisfied with incumbent vendors is security.
Of the 321 enterprises, 295 said they’d examine a strong security offering from a network startup, by far the highest level of acceptance. But that group stressed the point that “strong” here meant compelling, because enterprises have a very low risk tolerance in the security space, and startups’ business risk is considered very high. A security comment made to me in 2006 holds true today: “We want a vendor who we can sue if they fail to deliver on their promises.” So, while 295 enterprises said they’d explore network startups in the security space, only 23 said they believed it likely that any security-justified startup would pass muster with them.
The SD-WAN space, it turns out, can easily expand into security, and into accommodating new network services, so there’s not much chance for a startup there. Enterprises say they’d need an area where there is more capital equipment budget on the table, and they agree that their budget for network gear is primarily aimed at the data center.
That’s an area where, if you recall, we’ve already had two waves of new-technology promotion that have swept startups into the game—but they failed. There were software-defined networks (SDN) and then white-box switching. The problem with both, from a startup-potential perspective, was the lack of a compelling benefit, a transformational challenge to be met.
We might have one of those now, in AI.
Startup opportunities: AI networking and IoT networking
The enterprises I talk with tell me that hosted public generative AI isn’t likely to transform their business. To do that they need AI to analyze the core business information whose governance requirements have already ruled out public cloud hosting. They say that also rules out public AI, and thus they’d need to host AI in their own data centers. AI depends on clusters of GPU-equipped servers, and a prompt (a query, in AI-speak) of a hundred characters might generate gigabits of data movement within the cluster, just to return a couple hundred characters of result. “Horizontal” traffic is what this sort of server-to-server stuff is called, and we already see data center switches adding features to accommodate it. Could the AI traffic model be different enough from traditional LAN switching to justify a new approach? Of the 321 who had startup comments, 202 said they were exploring or already deploying self-hosting of AI, and 144 of them said they would consider a startup to support the networking there.
Could anything perhaps accelerate startup interest in AI networking? One thing is clear, which is that enterprises are only now starting to explore hosting their own AI models. Their network professionals say that today, the question of how to network their deployment is way behind the question of what model to host and how to host and manage it. They say that what’s needed is a fairly universal way of approaching the building and connecting of AI clusters, and they’re not hearing that yet.
There is one interesting point: Many enterprises (105 of the 144 enterprises who’d consider startup AI network vendors) think that AI networking, container virtualization in the data center, and virtual networking in the WAN via SD-WAN should all be elements of a new network and security model. If they were, all of the 202 who said they were doing or considering AI self-hosting said they’d be eager to review the offering. But they also said that the winner among startups in this new combined space would likely be an independent startup “for weeks” before an incumbent vendor snapped them up. But hey, what are startups for if not to flip? And, of course, to innovate to make that flip happen.
Another interesting area, enterprises say, is IoT networking.
Enterprises believe that AI and the concept of “digital twins” or computer models of real-world systems will combine to finally create a framework for IoT to shine. When that happens, they believe, there will be many more public sensors deployed, and this will mean coming up with a secure and efficient mechanism for connecting those sensors. But most of all, IoT networking needs a model that links it to things like digital twins and AI insights. The old days of networking, connecting people with people, and the middle-ages of connecting people to applications and content, are passing, say 154 of my enterprises. It’s the Age of Things that will create the network of the future – and make giants out of some happy startups. And the security requirement for this new age might also validate virtual networks, combining these two startup opportunities into something enterprises say they could really jump on.
Ah, but there’s that pesky qualifier “could” again. What could rain on our startup parade? Two things.
First, how VCs see as the optimum road for profit. Why jump in to finance a challenging technology shift when you can get into the social media dance, or ride the AI hype wave? Only 44 enterprises said they believed that VCs would likely fund a useful networking startup.
Second, competition, not from incumbents or other startups, but from the current second-tier network vendors (Extreme comes to mind) or even players like Broadcom. It’s easier to go from the minor leagues to the majors than from sandlot ball. So, while enterprises are ready and maybe even eager for a startup revolution, they’re not holding their breaths, nor am I. You shouldn’t be either, but maybe a startup threat to that second tier will usher in some network transformation next year and bring back some excitement to the space.
Read more from this author
- Why enterprises should care more about net neutrality: Net neutrality policies are the most significant regulatory influence on the Internet and data services, and they’re the reason why end-to-end Internet QoS isn’t available.
- Security platforms offer help to tame product complexity, but skepticism remains: As the number of security tools deployed continues to climb, it gets more difficult for enterprise security teams to know, with confidence, their overall state of security.
- What enterprises say the CrowdStrike outage really teaches: After the CrowdStrike outage, some enterprises IT teams are rethinking their assumptions about how use of the cloud impacts application reliability.
- There’s a new difficulty in tech hiring: how to recognize a good candidate. Hyped technologies create staffing pressures. And as technology gets more complex, it’s harder to understand what skills are needed and identify who has them.
Source:: Network World