Networking is changing, and in every changing market the biggest pressure is applied to the vendors. You can’t sell the same old stuff when the buyer is moving on, and they surely are. There are some pressures that we could classify as systemic, meaning they don’t originate from a single thing but rather from gradual tectonic movements, and others that are reactive, meaning they come about from a single factor. We have both, and that will impact network buyers and, of course, vendors.
Both our forces apply, but the most fundamental source of pressure is the lack of differentiation. Absent differentiation, any market tends to commoditize as vendors hunker down on the same feature set, the only features buyers value. As that happens, it creates its own set of pressures on vendor profits, but it also opens the door to third-party products, new vendors, and new means of gaining account control. That’s what’s happening in networking today.
What’s the most powerful force in the networking market today? Incumbency. Among the over-400 enterprises who have commented to Andover Intel, well over 85% say that they are “unlikely” to buy a product from a vendor other than the one they’re currently committed to. The more committed they are, the less chance a competitor can get in, not so much because a broad commitment means buyers are fixated on their main vendor but because having a few foreign products in a network, say buyers, creates a constant finger-pointing problem.
There are two vendor strategies to create or preserve incumbency. One is to create a valuable ecosystem around the network products, which can be done by sales means (discounts, delivery commitments, on-site support), by creating a management framework to ease operations, or both. The latter is the normal case, of course. The other way is to jump on developments that create a “green field” in the network. Anything that opens a new and relatively loosely coupled addition to network requirements invites a competitor to get their foot in the new door. AI is such a development.
Juniper, now a part of HPE, has done that, or at least has tried. I don’t think that it’s easy for networking to create such a green field on its own, nor will “AI” in a simple sense of using the cloud AI tools from Google, Microsoft, or OpenAI do that. It requires self-hosting AI, and self-hosting LLMs in particular. If an enterprise has to build, train, and use an AI cluster then that cluster will demand high-performance internal networking. HPE, though, is not known as a leader in that space. IBM has the most enterprise strategic influence in AI, but HPE is virtually tied with Nvidia, Broadcom, Dell, Oracle, and Supermicro. Can HPE stand out? Will Juniper help that?
Enterprises tend not to think about AI networking until they’ve planned an AI hosting cluster, so I can’t validate the chances that Juniper’s AI positioning would favor HPE. I also can’t validate the notion that, given a cluster is planned, Juniper’s AI-centric network positioning would overcome the influence of whoever was selected to provide the hardware. Thus, the big factor in establishing the value of the HPE deal for Juniper will be whether HPE realizes that to make it truly valuable, they have to establish strategic influence in AI hosting.
Cisco, meanwhile, has hired away an intel software and AI leader, and it’s already been working hard to expand its QoE credentials. However, the company has very little recognition as a key AI player, even among its customers. While most (over 80%) say that Cisco’s network products support AI networking, none list Cisco as a key AI influencer, and non-Cisco enterprises don’t cite them at any statistically significant level.
In many ways, Cisco has a harder nut to crack, AI-wise, than HPE. Cisco’s hosting capability is real, but it hasn’t contributed in any meaningful way to an AI position. Can they, at this point, do something transformational with regard to their AI credentials? Maybe, and one option for them would be some sort of alliance with Cisco, who you may recall bought IBM’s networking business long ago. IBM may see HPE/Juniper as a threat to their own commanding AI position, a position that’s driven their stock price up sharply. Might they see some sort of Cisco alliance as a solution to both company’s new HPE risk?
Computer/server competitor Dell has a similar problem. Right now, they’re at least in the game with AI hosting, but there’s no question that their own networking portfolio is limited. If HPE makes a strong move to gain AI influence, then HPE could overwhelm Dell. But who does Dell buy, or what do they do? One possible answer is to commoditize what you can’t win. Broadcom and Nvidia have networking chips, and Broadcom also has some AI stuff. Suppose Dell went behind the switch to the chip, fostering the notion of supporting white-box tools for networking and perhaps an open-source switch software and management strategy?
Perhaps the best example of this in the current market is DriveNets, who use Broadcom chips in their “Network Cloud” switch-cluster configuration. While they’ve been focused on the telecom market so far, they recently announced an AI cluster win with NeoCloud WhiteFiber, and there’s no reason why they couldn’t sell into any AI cluster mission, even one in the enterprise. Since their Network Cloud can perform as a switch, router, or both, they have a unique ability to play as the entire cluster/data-center infrastructure including connection to the Internet or a VPN. This is the sort of thing that could give all the major vendors headaches.
Smaller network vendor competitors don’t have it easy, either. The big risk they all face is a shift of strategic influence away from networking toward hosting, servers, and software. If that happens, they’re almost certain to lose market share. Can Arista. Extreme, or Palo Alto counter that? Only with difficulty. Hybrid cloud, or just cloud, networking is too widespread at this point. Management tools alone will help you keep your customers, but won’t open new green fields to harvest. And behind it all is the problem of commoditization, white boxes, and Broadcom.
Broadcom is in an interesting position. Because of VMware, they have a significant amount of enterprise strategic influence and a good cloud position. Because of their switch chips they have a good data center position, at least as a provider of white boxes. Like Dell, they could in theory hope to commoditize networking and by doing that invalidate HPE’s ability to leverage Juniper. They could, perhaps, create a kind of portable AI hosting model using VMware. They don’t need to be immediately proactive on either; they can wait to see whether players like Cisco, Dell, and HPE do anything smart.
The HPE deal for Juniper is likely a response to the systemic commoditization change in networking rather than a unique causal force, but even so it’s a validation of change that other vendors will surely respond to, because they have to. HPE or a competitor might just come up with a smart response, but it’s too soon to be sure.
