A couple of earnings reports last week may be telling us a lot about the markets’ direction. IBM had a good quarter, and Juniper’s was a bit less than that. That the two are in different spaces is surely true, but if the HPE deal for Juniper goes through, will the combined company be more like IBM? Will networking, which HPE will have more of, be a boost or a drag? Will HPE become an AI power like IBM? Let’s see.
As I’ve pointed out in past blogs, IBM has consistently led the industry in terms of strategic influence on customers. That means that IBM has been able to set the parameters of major projects, the sort of thing that typically drives “digital transformation” and the adoption of new technologies. They’ve done this by focusing their account teams on the things that make customers’ business cases, not just on pushing products. In the last decade, even though enterprise buyers have been struggling to make new business cases that would expand tech spending, IBM has been able to help its own customers put things together.
To be fair, IBM has an advantage in that a new business case in tech is almost surely driven by a new application, meaning software and systems, which IBM offers. The only thing that ever threatened IBM’s strategic dominance (for a short time) was the network space. IBM has network technology but it was expensive, and Cisco came along to offer connectivity on the cheap. But IBM adapted and resumed its strategic influence supremacy quickly, which meant it still had control of the applications that drive new spending.
Juniper is a network vendor. It was launched during the period when IBM’s network technology was giving way to IP, and when server resource pools were springing up to consume data center switches. The problem Juniper had, and still has, is that the boost that the shift to IP gave network vendors didn’t last, and network vendors aren’t the players enterprises run to when they want insight on digital transformation and new productivity business cases.
Ah, but what happens when (if?) HPE acquires Juniper? HPE like IBM is a software-and-systems provider. They understand digital transformation and making productivity business cases. The question is whether, the IP transformation period having come and gone, there’s a strategic hook to which networking can be tied, one that is also tied to productivity business cases and transformation. There may well be, and it’s AI.
Generative AI, in the form of massive public services hosted on thousands of GPUs and providing writing assistance to many workers, isn’t moving the ball much, based on what enterprises tell me. Yes, workers often like the tools, and yes, they may generate better emails or documents, but so far enterprises aren’t reporting real savings very often. Where AI is creating real business benefits is largely in the area of business intelligence and analytics, which happens to be the place where IBM has been focusing.
Companies are not going to cede BI to the cloud or to an “AI-as-a-service” technology because of the often-cited issues with data security/sovereignty. Most companies don’t and won’t store business-critical data in the cloud, so giving it to an AIaaS tool is a non-starter. That means that BI applications of AI would have to be run in-house. It’s the implications of that truth that create a potential opportunity for Juniper, HPE, and the union of the two.
To host, and to train, an LLM in-house means at least 200 GPU cores installed in servers and networked effectively. Juniper’s “NOW” announcement of networking for AI targets this. It also requires an LLM to host, and enterprises are increasingly interested in Meta’s Llama series, which is open source and getting increased attention among developers who are creating mission-specific AI models for business use. Juniper highlights Llama in its announcement, recall, and it created an AI lab that customers can use to prove out applications, which also uses Llama (but customers can bring their own model). What Juniper did in their announcement was a strong step toward offering critical guidance on in-house AI hosting.
It doesn’t address the strategic influence issue, though. Juniper has between a quarter and a fifth of IBM’s influence, which isn’t enough to give it a seat at the application planning table. However, HPE is (narrowly) in second place for strategic influence, which could well rise with the addition of Juniper and it’s networking-for-AI NOW strategy.
The problem is that the HPE/Juniper deal is on regulatory hold, at least for the moment and possibly into early 2025, though recent news says that at least the EU approval may be coming shortly. The deal, of course, will take time to gel once it’s final, and all that is likely to capture management attention on both sides. Then sales/marketing/positioning material needs to be done. All of this puts the combination of the two companies for in-house AI hosting at risk, at the time when it’s most needed. Too much delay here risks having someone else (Dell, Supermicro, NVIDIA, Broadcom, even Cisco) step up, now that Juniper has made its position clear.
There is, I believe, a significant opportunity here. In-house AI could be a major enterprise priority, not only for BI but also for things like pre-sale/support chatbots, because even that data is considered confidential by 45%/68% (respectively) enterprises. The opportunity would be for AI hosting and networking first, but there would also be a growing opportunity for AI training resources, since training a model would typically require what enterprises estimate as double to quadruple the resources. Think sovereign-training-as-a-service and you get the idea. Some of this could be done generically, with augmentation with company-private data coming later and in house. There is work being done in this area already, I’m told.
Users believe that a sovereign version of GPUaaS would be possible, and their candidate provider is not the current cloud or AiaaS providers, but rather the same firms who have significant strategic influence. IBM tops the list, and in fact is mentioned as a desirable GPUaaS provider by more enterprises who aren’t IBM accounts than I’d expected. IBM is the leader here, so it’s up to HPE to change that.
AI isn’t the only opportunity for HPE and Juniper to demonstrate symbiosis, but it may be the most important one. First, Juniper has just announced this and done a decent job with PR and positioning. Second, it’s an opportunity that involves products and initiatives from both companies. Third, it’s a strategic area for enterprises, making it an opportunity to exploit strategic influence, which means it could be a good way to increase it. Will HPE and Juniper take the leap? We’ll see.
