Digital infrastructure saw some pretty revolutionary changes in 2023. The U.S. government allocated billions of dollars to bridge the digital divide through the CHIPS and Science Act. AI became the top topic of conversation across all industries. And after years of flailing, edge computing is getting some real use cases.
So what’s on the horizon for 2024? Will IoT finally become meaningful to the average person? (Probably, I can now run my dishwasher from an iPhone app.) Will cybercriminals finally gain the upper hand? (Maybe, DDoS attacks grew in triple-digit percentages across many measurements.) Will AI supplant human cognition? (Doubtful, Ray Kurzweil still looks on track with his 2029 prediction for AI to match human intelligence.)
What is more likely to be observed in 2024? AI’s need for greater bandwidth and processing will continue expanding. The talent shortfall - including a dearth of hybrid experts and a lack of new talent in the infrastructure space - will remain a concern. And shifting views on the utility of public cloud resources will drive change.
#1: Building network capacity to power AI’s rise
AI’s appetite for bandwidth and processing will keep growing with blinding speed for the foreseeable future. AI system's demand for bandwidth and processing power will likely evolve at a much faster pace than the physical infrastructure required to support it. Power for AI processing and network capacity to move data are all potential bottlenecks for AI advancement. Averting shortfalls requires rapid physical infrastructure development, starting yesterday.
To illustrate this trend, Nvidia sold half a million of its H100 chips in 2023 and expects to sell four times that in 2024; Facebook alone is requesting 350,000. The amount of power, specifically in data centers, required to power the chips produced in 2024, about 3200 MW, is double the amount of power used in top data center markets in the world in 2022. A new data center building takes 2 years to build, which means construction anticipating the additional data center power required by AI would have had to start in 2022 - well before chatGPT attained rock status in 2023. Power, especially environmentally friendly power, takes considerable time to bring online. Where the average data center takes two years to build, the average dam required for hydroelectric power takes over eight years. Construction for the power infrastructure required for the new AI workloads would have needed to start in 2016.
Data center buildings must not only exist where there is power to train your new Llama 2 AI model to reliably recognize pictures (for example, of cats), but also where there is network connectivity to 1) import the training data (lots of cat pictures) to the equipment running the Llama model, and 2) provide an answer back to users (is the picture in question indeed a cat).
Data center buildings are ultimately manifestations of the most difficult to develop resources: power and fiber optic networks. The two don’t seem related at first, but they have a long and involved relationship. Data center buildings must be built where power is ubiquitous and cheap, such as hydroelectric. However, these locations are often not around major cities or located anywhere that existing fiber optic cables exist. New fiber routes must be created to these locations, which can take anywhere from six months to three years depending on the location.
The combination of long time frames for infrastructure development combined with unknown demand for network connectivity creates a high risk of infrastructure not keeping pace with the meteoric requirements of AI. The infrastructure groundwork created today is required to help us realize AI’s full potential tomorrow.
#2: Strategic repatriation from the cloud
Organizations are rethinking their cloud strategies, recognizing that not all applications, data, workloads, or bank account balances benefit from public cloud hosting. Concerns about security, privacy, and cost will drive companies to move partially to company-owned and operated infrastructure. Companies opting to remain wholly (or mostly) cloud-reliant instead of identifying which data, applications, and workloads should be relocated back to an on-premises infrastructure risk losing ground to more agile rivals and missing out on attractive cost optimization. Basecamp illustrated just how financially savvy cloud repatriation can be by the company's complete move out of the cloud in 2023, which will save $7 million in costs over the next 5 years.
Does this mean we’ll see a mass exodus from cloud services? Not at all. Instead, expect a better optimization of cloud computing functionality. Companies adopting a more balanced approach to leveraging public and private cloud and on-premise solutions will find greater affordability, efficiency, and versatility. This realignment - driven by the need for tighter controls, data safety, and application-specific tuning - may seed an on-site networking renaissance.
An evolving cloud approach means enterprise IT teams will need to reassess network strategy because the two go hand-in-hand. As companies repatriate data and applications, their networking decisions must keep pace, enabling secure, high-performance connectivity between public cloud, private cloud, multi-tenant data centers, and on-premise computing.
#3: The need for hybrid experts
The communications infrastructure industry not only has to grapple with rapidly moving trends, it also desperately needs experts who can bridge the gap between legacy systems and cutting-edge technology. Organizations wrestle with integrating antique assets into contemporary solutions as they shift operations to (and from) the cloud to achieve scale, savings, and productivity. With this new wave of digital infrastructure gathering momentum, the need for skilled professionals able to operate with mastery of both spaces will increase.
The rapid uptake of Infrastructure as Code has removed old divides. The divided roles of yesteryear - network engineers, software developers, dev ops, and cloud architects - no longer have sharp edges. Network devices have APIs and network operators can transform their workflows and save heaps of company time and money by being proficient in Python. DevOps can use Terraform to turn down Kubernetes clusters and turn up network connectivity. Software developers are working in more complex systems that require cloud and network knowledge to debug difficult problems. Cloud architects need to be able to design systems that incorporate private cloud resources, applications that run on-premise, and complex security concerns. Multi-disciplinary experts will see increasing demands even in the employer-favored economy in 2024.
While 2023 opened the door for major transformative change, 2024 will be the year we see those changes truly come into action. Companies that continue to prioritize and invest in their digital infrastructure and hybrid experts will be well-positioned to capitalize on the continued growth of AI and other emerging trends and keep themselves ready for what’s next.
