Kelly Herrell is CEO of Hazelcast. Powering mission-critical applications that move the economy.
This year has been a watershed moment for AI. Billions of dollars have poured into AI startups. Regulations are swirling, trying to find a balance between protecting consumers and encouraging innovation. Some companies are beginning to see results from their AI investments, but many others are not yet there.
No doubt, AI will someday impact every industry and what seems like limitless work processes. The stars are slowly aligning to make it happen, and it will happen, in some companies and industries than in others.
But one thing is certain: To be ready, executives need to build the infrastructure to enable that future.
I think of technology infrastructure as akin to the infrastructure of a building. When someone builds an office complex, they build it for the long haul. They don’t put in pipes that degrade in five years. They install pipes meant to last for decades, and cables that still work even as cable networking improves. Need proof of a technology equivalent? Just look at mainframes and their continued revenue growth for companies like IBM.
AI workloads, including GenAI and advanced analytics, will snowball in the next few years. Infrastructure will need to be nimble enough to handle all of them. It cannot be so niche that it works great for a point solution but nothing else and needs to be swapped out year after year.
Infrastructure Decision Making
Here are the top seven considerations for executives when making decisions related to the technology infrastructure that will buoy AI workloads:
How will you continually shape your data?
Data will determine the success of enterprise and generative AI use cases. More importantly, how will you integrate your enterprise data with pre-trained models from the likes of OpenAI and others? How will you combine structured and unstructured data? And how will you organize and optimize the data? It is critical to recognize that these are among the more important challenges to solve, especially if you don’t have the right data infrastructure.
An IDC study of 2000 executives found that “dealing with data” was their biggest hurdle as they approached AI and that “much of their AI development time is spent on data preparation alone.” No surprise. Data needs to be clean, organized and easy to access, especially unstructured data, such as video, text and images, which must be labeled before beginning to train an AI model.
How fresh is the data?
The fresher the data, the more valuable it is. Generic large language models (LLMs) that don’t easily ingest fresh data will not be of much use to enterprises solving a custom problem. Ensure your infrastructure can readily ingest and integrate fresh data to inform AI solutions and achieve the desired ROI for your use case. Fresh data combined with the context of historical data will be key to informing AI predictions when a real-time and accurate response will be more valuable than a delayed one.
How well and fast do you process that data?
What’s your infrastructure like today for processing data? Is it robust enough to handle the significant demands that AI workloads will create? Forrester Research finds that “inadequate infrastructure to support the consumption, storage and sharing of massive volumes of data” is a major pain point for companies, VentureBeat reports, along with “difficulties in integrating with existing infrastructure.” Data’s value is its highest the moment it is born. Being able to process it instantly—and perhaps even combining it with historical data—will lead to more impactful AI outcomes.
Computation
Intensive is the keyword for AI and computation, involving more than silicon. The software computation layer is at the heart of an AI architecture. You’ll want high-performance capability, whether on-premises or in the cloud. And you want compute capabilities that can scale—without changing everything else in your infrastructure to accommodate it. You cannot do AI without compute, which is where you combine logic with data to deliver a result. It could be traditional logic, machine learning or AI and you need compute infrastructure to do each one for the foreseeable future. All three types of computation will coexist inside a company’s workflows.
Storage
With so much data, high performance storage is required to handle the volume, velocity and variety of data. Consider the scale and speed at which your data storage needs to ingest and serve data. Similarly, storing, merging and accessing structured AND unstructured data must be addressed, especially if there is a response time requirement for the use case.
Complexity
Given that AI workloads require a lot more data, a lot more compute processing and even more storage, it could be tempting to create an increasingly complex infrastructure with many different point solutions, which would be a costly mistake. Complexity requires more integrations of all kinds of components, which takes time and attention away from application development and deployment. Always look for ways to simplify and unify your infrastructure, such as unified platforms that can provide the core architecture for your applications. These platforms will accelerate your time-to-market, reduce total cost of ownership (TCO), and improve your ROI.
Legacy vs. new
There is always a time to recycle the old, and then there’s a time to invest in something new. Every business will face a different equation here, depending on its AI workloads, technical debt from legacy infrastructure and how real-time it needs its AI solutions. An AI-effective infrastructure should be attainable for every application, whether it serves a thousand people or a hundred thousand people. Every company can support an AI initiative with the right infrastructure and the right data.
Be Excited, Yet Prepared
No doubt, it’s easy to get excited about the potential of AI in today’s economy and society, and companies are rushing headfirst into experimentation. However, it doesn’t make sense to get excited about what could happen until you have the right foundational infrastructure—especially the data and data processing infrastructure—to make it happen. It’s the early days of a long construction process, and picking the right plumbing matters intensely.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?
