The great cloud migration, which began about a decade ago, brought about a significant revolution in the field of IT. Initially, small startups and businesses without the means to build and manage physical infrastructure were the primary users of cloud services. Additionally, companies saw the benefits of moving collaboration services to a managed infrastructure, leveraging the scalability and cost-effectiveness of public cloud services. This environment enabled cloud-native startups like Uber and Airbnb to thrive and grow rapidly.
In the subsequent years, a vast number of enterprises embraced cloud technology, driven by its ability to reduce costs and accelerate innovation. Many companies adopted “cloud-first” strategies, leading to a wholesale migration of their infrastructures to cloud service providers. This shift represented a paradigm change in IT operations.
However, as the cloud-first strategies matured, certain limitations and challenges have emerged. The efficacy of these strategies is now being questioned, and returns on investment (ROIs) are diminishing, resulting in a significant backlash against cloud adoption. This backlash is primarily driven by three key factors: escalating costs, increasing complexity, and vendor lock-in.
The widespread adoption of the cloud has led to a phenomenon known as “cloud sprawl,” where the sheer volume of workloads in the cloud has caused expenses to skyrocket. Data-intensive processes such as shop floor machine data collection should never have been considered for the cloud. Manufacturers are finding that datasets of hundreds of gigabytes should never have left the premises. Enterprises are now running critical computing workloads, storing massive volumes of data, and executing resource-intensive programs such as machine learning (ML), artificial intelligence (AI), and deep learning on cloud platforms. These activities come with substantial costs, especially considering the need for high-performance resources like GPUs and large storage capacities.
In some cases, companies spend up to twice as much on cloud services as their previous on-premises systems. This significant cost increase has sparked a realization that the cloud is not always the most cost-effective solution. As a result, a growing number of sophisticated enterprises are exploring hybrid strategies, which involve repatriating workloads from the cloud back to on-premises systems.
By developing true hybrid strategies, organizations aim to leverage the benefits of both cloud and on-premises systems. This approach allows them to optimize their IT infrastructure based on the specific requirements of different workloads and data science initiatives. Moreover, hybrid strategies offer greater control over costs, reduced complexity, and increased flexibility to avoid vendor lock-in.
In fact, leading technology companies like Nvidia have estimated that moving large and specialized AI and ML workloads back on premises can result in significant savings, potentially reducing expenses by around 30%.
In conclusion, while the great cloud migration brought undeniable advantages in terms of scalability and innovation, the limitations and challenges associated with cloud-first strategies have triggered a backlash. To address these issues, enterprises are embracing hybrid strategies, repatriating critical workloads to on-premises systems and leveraging the benefits of cloud and traditional infrastructure. This evolution represents the next generational leap in IT, enabling organizations to support their increasingly business-critical data science initiatives while regaining control over costs and complexity. If your organization has data being collected and stored in the cloud, you may want to start to plan to migrate that ever-growing data back to on-premise and mitigate the costs. If your organization is thinking of a cloud solution, think again.
Thomas Robinson is COO of Domino Data Lab,