Cisco 2024: AI adoption will drive network modernization, data-center evolution

AI-related technology will drive many of the trends Cisco executives expect to impact the enterprise in 2024, while key networking and data-center developments will change the future of infrastructure.

On the AI side, there are developments coming that Liz Centoni, Cisco executive vice president, chief strategy officer and general manager, applications, says will lead to “a once-in-a-generation shift…opening vast new opportunities and transforming industries, modes of operation, and career paths.”

“GenAI will fast expand into the business world with GenAI-powered [natural language interfaces], customized LLMs, tailored B2B applications and business context,” Centoni wrote in a blog outlining her predictions. Natural language interfaces (NLI) powered by GenAI are expected for new products, and more than half will have this by default by the end of 2024, Centoni stated.

“GenAI will also be leveraged in B2B interactions with users demanding more contextualized, personalized, and integrated solutions,” Centoni stated. “GenAI will offer APIs, interfaces, and services to access, analyze, and visualize data and insights, becoming pervasive across areas such as project management, software quality and testing, compliance assessments, and recruitment efforts. As a result, observability for AI will grow.”

In 2024, businesses will look for innovative ways to use AI without the complexity and cost of building their own platforms, and APIs will play a pivotal role, Centoni predicted. “APIs will increasingly act as an ‘abstraction layer’ – seamless bridges that integrate a multitude of pre-built AI tools, services, and systems with little development or infrastructure setup. With access to a vast array of AI capabilities through APIs, teams will automate repetitive tasks, gain deeper insights from data, and enhance decision making,” Centoni stated.

Along with the potential benefits AI may provide, security will remain a key concern, and fending off threats will be an important part of AI’s development. “In 2024, AI-enabled disinformation, scams, and fraud will continue to grow as a threat to businesses, people, and even candidates and elections. In response, we’ll see more investments in detection and risk mitigation,” Centoni wrote. “Inclusive new AI solutions will guard against cloned voices, deepfakes, social media bots, and influence campaigns. AI models will be trained on large datasets for better accuracy and effectiveness. New mechanisms for authentication and provenance will promote transparency and accountability.”

AI impact in networking

Specific to networking, AI will reinforce the need for more versatile switching infrastructure, according to Jonathan Davidson, executive vice president and general manger of Cisco Networking.

“Enterprise Al network fabrics will continue to evolve to accommodate both traditional and Al/ML workloads, highlighting the importance of versatile switching fabrics. New approaches to solving the I/O challenges will take the industry through cycles of proprietary innovation and standards development, resulting in a transformed landscape of fabrics inside and between servers,” Davidson said.

“For instance, moving data faster and more efficiently between models and data stores will bring GPUs closer to primary storage and blur the lines between internal and external fabrics in server design. Ultimately, Cisco believes there will be a wide range of use cases and customer needs when it comes to AI networking,” Davidson said.

“The industry needs to find solutions to power Al networks and configure them using multiple architectures, including telemetry-based, fully scheduled fabrics, and Ethernet/non-oversubscribed choices, if we are to unlock the promise of AI,” Davidson said.

Ethernet in AI networking

Ethernet, in particular, will undergo significant evolution in the coming year.

In a move that demonstrates the potential role of Ethernet in AI networking, 27 new members in November joined the Ultra Ethernet Consortium.

AMD, Arista, Broadcom, Cisco, Eviden, HPE, Intel, Meta and Microsoft established the UEC, which is hosted by the Linux Foundation, in July to develop physical, link, transport and software layer Ethernet advances in particular to handle the performance, scale and bandwidth to keep up with AI demands. The 27 new members include Alibaba Cloud, Dell, DriveNets, Fujitsu Limited, Huawei, Juniper Networks, Nokia and others.

“The formation of the UEC and the rapid membership growth is a testament to the strong interest in using Ethernet as the basis for networking solutions targeting AI/ML workloads. It’s also a recognition that broad industry standardization and cooperation is necessary to solve large, industry-wide challenges,” Davidson said. “In 2024, we anticipate the UEC will generate and publish initial specifications. We believe this will only be the start of the UEC’s impact and influence on the networking industry’s adoption of Ethernet-based solutions for the AI/ML space. With its broad industry scale and customers’ existing investment in the technology, Ethernet provides a wealth of flexible solution options.”

There will be continued efforts to reinvent Ethernet technology to deliver better performance and resiliency as it becomes the de facto networking fabric for Al networking, Davidson said. 

“As most customers look to build out their networks to handle new types and greater volumes of workloads, most want a single architecture to reduce operational complexity and costs – one network fabric to run them all,” Davidson said. “Ethernet has emerged as that preferred technology for Al training and inference, offering a standardized, flexible, and cost-effective alternative to proprietary technologies like InfiniBand.”

To understand why, it’s important to acknowledge Ethernet’s evolution and role in enabling other major technology shifts, Davidson said. “Traditional workloads such as voice, video, storage, high-performance computing (HPC), and high-performance databases have created requirements toward making Ethernet ‘lossless’ that significantly improves reliability, lowers latency, and increases performance. Ethernet also provides economical scaling to large core clusters and smaller edge clusters,” Davidson said.

Network modernizations on tap

Modernizing network and data center infrastructure will continue to be an enterprise investment priority to keep pace with the explosion of workloads, create performance capacity and drive operational efficiency, Davidson said.   

“Today 100G network fabric technology is common, yet it’s not a stretch to think the industry will soon reach a tipping point where network architectures require higher performance capacities to support new applications, data, and workflows including AI/ML,” Davidson said. “There is an increasing demand for higher network speed and bandwidth advancements that accompany innovations in GPUs and NICs, reaching up to 400G and 800G speeds.”

Such demands mean fundamental changes to how data centers are built and operated, Davidson said. “The development of 400G/800G leaf-spine architectures, with 100G SerDes, addresses the growing demand for higher bandwidth on server and client sides. We will also see a growing emphasis on security and multi-tenancy to segregate data for Al workloads. And we will begin to see customers incorporate liquid cooling where applicable for optimal Al infrastructure energy consumption,” Davidson said.

Environmental, social, and governance initiatives are top priorities for companies today, and this has put added pressure on IT organizations, including network operations teams, to support sustainable IT and data center initiatives, Davidson said. “Modernizing network and data center infrastructure can be a key step, as many of the newest products were designed at the architecture level to operate more efficiently,” Davidson said.

In addition to the networking changes, 2024 will shine a spotlight on the importance of network visibility and assurance, Davidson added.

“We believe the dynamic nature of changing Internet performance will require always-on end-to-end assurance to plan and mitigate inevitable disruptions,” Davidson said. “In today’s digital economy, the connective tissue between customers, applications, and brands is the public Internet. As organizations continue to embrace cloud, SaaS, and SD-WAN, we know that they are grappling with the complexity of now being dependent on external networks that sit outside of their control. Outages and other disruptions are a matter of when, not if,” Davidson said.

The ability to connect, secure, and assure the delivery of digital experiences, no matter the underlying infrastructure, is critically important to mitigate any disruption to business operations, Davidson said.