(This is the first part in a two-part column on AI networking. The first will focus on the larger public companies, and the seconde will focus on the startups in the market.)
It’s going to be a fascinating year in large-scale enterprise networks. The networking infrastructure market continues to be fueled by interesting twists and turns—much of it triggered by the explosion of AI applications and data. The stakes have never been higher for the winners and losers.
Demand for AI has opened the spigot for new data and connectivity. Some of the largest datacenters in the world are being built to support AI. Additional demand for training and inference data, as well as distributed cloud applications, will spur the need for automated connectivity among clouds and distributed data platforms.
The excitement about the networking market is rooted in two growth opportunities: 1) Networking for AI, or the connectivity infrastructure needed for AI datacenters and inference; and 2) AI for networking—the AI automation to drive operations, often also referred to as AIOps.
This has served as a catalyst for shakeups happening in all segments of the networking market—enterprise campus, datacenter, hyperscaler, and telecom—and sets the stage for competitive fireworks for many years to come. The emerging AI infrastructure boom presents a strategic turning point in the networking landscape.
AI Drives Juniper-HPE Deal Drama
Market shifts often start with changes at the top. When AI infrastructure came onto the scene, with an explosion of AI buildouts expected to reach $1 trillion over the five years ending in 2030 (according to Futuriom), it presented an immediate opportunity for the leading networking players to see a piece of the action, with networking needed to connect the backend and front-end of AI clouds. Seeing the need for cutting-edge networking, GPU chip leader NVIDIA entered the market quickly in 2019 with the purchase of Mellanox for $7 billion. That established NVIDIA with a dominant position in InfiniBand, a popular supercomputing networking technology that leant itself well to networking some of the largest AI training clusters. It also helped NVIDIA build a full AI systems stack.
But InfiniBand wasn’t enough. NVIDIA knew that Ethernet would play a role in AI networks, so it quickly developed Spectrum-X, its own flavor of Ethernet. Leading networking companies such as Cisco, Juniper Networks, and Arista Networks also built Ethernet-based AI networking portfolios and coalesced around the Ultra Ethernet Forum, an alliance designed to guide Ethernet into the AI era with new feature sets.
But wait—there’s more. Just last month, Cisco teamed with NVIDIA to develop systems based on Spectrum-X. What’s going on here? Cisco appears to be hedging its bets, looking to protect its franchise in proprietary Silicon One chips, while at the same time integrating systems on which it can deliver networking for any flavor of Ethernet.
Cisco is looking to make sure it doesn’t make mistakes in AI that it made in the datacenter category, where it has let rival Arista Networks gain share in recent years. Arista Networks has a strong AI networking portfolio and it recently had a big AI networking customer win at Meta. It’s projecting AI networking revenue of $750 million in the current year.
That brings us to the proposed HPE-Juniper tie-up, which was recently blocked by U.S. authorities. With the AI boom developing, HPE saw it as a golden opportunity to contend for a top spot in the networking market when it bid $14 billion for Juniper early last year. The goal was to create a number-two networking player to Cisco’s leadership position in the market (Cisco is estimated by most research firms to control 60%-70% of the networking market). However, the U.S. Department of Justics has ironically found this situation “anticompetitive” and sued to block the merger, leading to doubts that the deal could come to fruition.
Juniper checks both the boxes in AI, with AI for networking via Mist and Apstra AI orchestration smarts; as well as networking for AI, where it has recently gained some impressive wins and shown growth. Several sources believe Juniper is in the mix to supply the 100,000-GPU xAI datacenter in Memphis, which may split networking allocations among Juniper and NVIDIA.
Meanwhile, additional public companies are upping their AI networking game. Nokia is resurging in the datacenter with solid customers such as Apple and CoreWeave, a player in the fast-growing neocloud market. And Extreme Networks has also been enjoying strong growth as AI drives vitality in enterprise and campus networking.
Public Companies and Startups May Converge
If anything, all this activity on the networking has provided new dynamics to the industry, with all of the major networking companies looking to take advantage of market shifts and growth. This may also include consolidations, as the larger companies buy startups to fuel innovation.
To sum it up: Look for a big scrum in this market as major players position themselves for AI. NVIDIA isn’t going away, and neither is Cisco or Arista Networks. All of this raises the stakes for HPE and Juniper to fight the DOJ to make the deal happen or else look at other strategies.
Next week, in part two, we’ll take a look at some of the dozens of private networking companies providing innovative technology to fuel AI and security for the growing distributed cloud.
Source link