The artificial intelligence infrastructure race is expanding the focus beyond processing power as operators evaluate the broader infrastructure requirements needed to build and scale AI data centers effectively. Hewlett Packard Enterprise networking chief Rami Rahim argues that connectivity has become the overlooked layer that determines whether massive AI investments deliver their expected performance. The executive views networking as a strategic multiplier that can unlock greater value from expensive computing resources deployed across large-scale environments. AI builders continue to focus heavily on securing space, managing power demands, and expanding compute capacity, yet network efficiency increasingly defines the operational ceiling. The shift reflects increasing attention toward how infrastructure components such as compute, networking, and power systems must operate together to support large-scale AI deployments. Rahim’s comments highlight a growing industry realization that the future of AI depends on coordination between chips, software, power systems, and high-performance networks.
Connectivity Emerges as the Missing Factor in AI Data Center Expansion
HPE Networking Executive Vice President and General Manager Rami Rahim described connectivity as a critical factor that many AI data center planners have underestimated during expansion cycles. He explained that organizations frequently prioritize GPUs, energy availability, and physical capacity before they evaluate the network foundation needed to maximize those investments. “A cloud provider can spend hundreds-of-millions or billions on GPUs (graphic processing units), ultimately, and it is true that in a given data center the total investment in networking is a relatively small portion, it’s maybe 10 to 15% of the aggregate investment that needs to go into the data center,” Rahim said.
“But what people have come to the realization of is that if your network has congestion, has blocking issues, has reliability problems, what will happen is those GPUs that you spend hundreds-of-millions or billions on could operate at 75% of utilization, 50% utilization, 25% utilization. … Networking is truly become a force-multiplier for massive data center and AI data centers.” Rahim’s comments highlight concerns that network limitations can affect AI infrastructure efficiency as organizations invest heavily in accelerated computing resources. HPE’s combination with Juniper Networks now supports Rahim’s broader vision for addressing evolving AI infrastructure requirements. The expanded portfolio gives HPE additional networking capabilities as the company addresses changing demands from AI workloads and large-scale data center environments.
HPE’s Juniper Integration Targets a New Networking Era
The distinction between scale-across and scale-up networking represents a major architectural change as AI clusters become larger and more interconnected. Therefore, HPE is positioning networking as a foundational element that supports computing environments instead of treating it as a supporting layer. Rahim pointed to scale-across networks as another area where complexity continues to increase as AI deployments expand across multiple systems and locations. Traditional approaches that relied on simpler connections are evolving into sophisticated architectures requiring intelligent routing, optimization, and greater operational visibility. “Sometimes these scale-across networks are fairly simple, they’re point-to-point, but increasingly scale-across has become a new core network with multipathing, and optimizations, and so forth, so that’s a really an important part of our portfolio, and the business momentum there is amazing,” Rahim said. HPE’s strategy emphasizes closer integration between networking capabilities and the workloads that rely on them across modern data center environments.
Automation remains an important area in HPE’s networking strategy as organizations seek to manage increasingly complex infrastructure environments. Rahim highlighted progress toward self-driving network capabilities designed to identify, analyze, and resolve operational challenges with limited human intervention. He explained that current deployments show significant improvement in how network systems respond to incidents and provide actionable intelligence to operators. “It used to be 50%, went to 60, 70. We are now over 80% of these issues either fully self-remediated or the answer is known immediately,” Rahim said. “It’s just a matter of trust of the operator enabling it go on its own.” The advancement reflects a broader industry focus on using automation technologies to improve network operations, incident response, and infrastructure management. Moreover, AI-driven operations could become essential as data centers manage increasingly complex combinations of hardware, applications, and workloads.
Security Strategy Expands Across the Network Layer
HPE is continuing to integrate security capabilities into its networking portfolio as organizations increasingly evaluate approaches that combine connectivity management and security functions. The company has been developing its secure access service edge strategy around closer alignment between networking and security functions. Analysts have questioned whether HPE needs to accelerate its security capabilities compared with competitors building broader networking and cybersecurity platforms. Rahim responded by emphasizing that the company is focused on combining existing security components into a more unified experience. “It comes down first to the ingredients that we have: firewalls, network access control, SSE (security service edge), SD-WAN, SASE solutions. We are putting a lot of pressure on ourselves to integrate … and we want security to be frictionless. We don’t want a standalone product,” Rahim said.
Future AI networking architectures may involve deeper integration between computing systems and optical technologies as the industry evaluates new approaches for improving bandwidth and efficiency. Co-packaged optics, or CPO, remains an important area of research as data centers search for greater bandwidth efficiency and lower power consumption. Rahim explained that Juniper had explored the technology through internal development efforts and customer prototypes before widespread market adoption. “I do believe that CPO will become a thing and it’ll probably become a thing first in scale-out and then in scale-up, and because of the lessons that we have learned with the products and the prototypes that we built, we will absolutely be ready for that,” Rahim said. The company’s approach reflects efforts to evaluate practical deployment requirements before broader adoption of co-packaged optics technologies. CPO adoption will depend on business economics, infrastructure readiness, and customer demand rather than technology availability alone.
AI Data Centers Enter a New Infrastructure Competition
The AI infrastructure market is expanding beyond processor acquisition as organizations evaluate broader factors such as networking, efficiency, and operational management. Companies making significant investments in AI compute are increasingly evaluating whether their network infrastructure can support efficient workload execution. Rahim’s comments underline the growing importance of infrastructure balance as AI workloads become larger and more distributed. Meanwhile, networking vendors are attempting to redefine their role from connectivity providers into strategic partners for AI transformation. Capabilities such as reducing bottlenecks, improving automation, and integrating security are becoming important considerations in future data center architectures. HPE’s networking strategy reflects this changing landscape where performance depends on the entire ecosystem rather than individual components.
Large-scale compute deployments create significant capital commitments, making efficiency improvements increasingly valuable across every layer of the data center. Rahim’s perspective suggests that companies cannot treat connectivity as an afterthought when building next-generation environments. The relationship between networking, compute utilization, and operational reliability is becoming an important consideration for cloud providers and enterprises planning AI infrastructure investments. “Nobody wants to do this just for fun,” Rahim said. “It comes down to the economics of that kind of deep integration versus not, because at the same time, while there’s these sort of advancements in CPO, there are also advancements that are happening in the pluggable optics themselves, and so there’s going to be a little bit of an interesting dynamic that’s going to evolve over the next year or so that we have to stay very much on top of.”
