By David Linthicum
The rapid evolution of generative AI is poised to influence the significant adoption and expansion of multicloud architecture. What is most interesting is that multicloud is occurring mainly behind the scenes, without much fanfare, taking a backseat to the hype around generative AI. I believe it’s just as important, and enterprises need to pay attention.
Everyone saw this coming
Generative AI models, especially large-scale neural networks, require immense computational power and scalable infrastructure. Multicloud architecture is essentially a complex distributed architecture that spreads workloads across multiple cloud service providers, on-premises systems, edge, and anything that can store or process stuff. Multicloud offers the requisite scalability and flexibility whether you’re hosting generative AI systems or not.
By leveraging different cloud environments, enterprises can dynamically allocate resources, ensuring that AI workloads are efficiently managed without bottlenecks. This flexibility is particularly crucial for generative AI models, which often necessitate bursts of high-performance computing and vast amounts of storage.
One of the primary benefits of a multicloud strategy is cost optimization. Generative AI workloads can be expensive to run continuously. Using a multicloud approach, organizations can optimize costs by selecting the most cost-effective cloud provider for specific tasks.
This has been a big bugaboo for me, considering that enterprises are attempting to go with an all-AWS approach for their generative AI deployments, or all-Microsoft or all-Google. They focus on homogeneous architecture for convenience and simplicity but miss a great deal of value by not considering other cloud platforms. They’re also likely spending two to three times as much on systems that are already expensive to build and run.
For example, an enterprise might use one provider for data storage at lower costs and another for high-performance computing due to superior processing capabilities. This strategic allocation helps minimize expenses while maximizing the efficiency of generative AI processes. While this sounds obvious, it’s not. I see a discouraging number of deployments that are wholly underoptimized. When companies limit their platforms to just the “preferred vendors,” these two words often mean “money wasted.”
Best-of-breed is best
The multicloud approach enables organizations to leverage the best-of-breed services from various cloud providers. This is the value of multicloud—not redundancy or the ability to beat cloud lock-in (both are multicloud myths I debunk about three times a week). Lock-in still exists with multicloud; we still must write our systems down the native APIs that are, for all purposes, proprietary.
Multicloud is advantageous for generative AI development, which may require specialized tools and environments. Enterprises can integrate advanced services such as AI-specific GPUs, specialized machine learning platforms, and unique data analytics tools from different providers. This fosters innovation and cutting-edge AI applications. Innovation drives value, which is the game we’re playing.
By utilizing multiple providers, enterprises can create a multilayered security approach, ensuring that data and applications are protected across different environments. This distribution also mitigates the impact of security breaches. The compromise of one cloud provider does not necessarily expose the entire system.
However, it would be less than truthful to say that multicloud drives better security. The complexity of these systems makes them more complex and more costly to secure. Breaches are often caused by the lack of attention and resources, not the heterogeneity, which is an argument I often hear.
Lack of understanding
The biggest issue I see right now is enterprises don’t understand that multicloud is an option. Or, they have adopted multicloud by accident and don’t understand what it really requires for security, operations, deployment, cost, etc.
The popularity of generative AI means that this will occur even faster, and enterprises that have no understanding of multiclouds will suddenly own one. My advice is that you do multiclouds on purpose and that you do the necessary planning and design to design an architecture that returns the most value to the business. That’s what it is all about.
Do yourself a favor and get ahead of this now, no matter if generative AI is in your short-term plans or not.
