The alternative cloud market is experiencing significant growth, with companies like CoreWeave, Lambda Labs, Voltage Park, and Together AI attracting substantial investments. This surge in interest and funding can be attributed to the increasing demand for hardware to support generative artificial intelligence (AI) models. GPUs are the preferred choice for training and running these models due to their parallel processing capabilities. However, the installation of GPUs is expensive, prompting developers and organizations to turn to the cloud instead.
While major cloud computing providers like Amazon Web Services (AWS), Google Cloud, and Microsoft Azure offer GPU instances optimized for generative AI workloads, alternative cloud providers can offer cost advantages and better availability for certain models and projects. For example, renting an Nvidia A100 40GB GPU on CoreWeave costs $2.39 per hour, while on Azure, it costs $3.40 per hour, and on Google Cloud, it’s $3.67 per hour. As generative AI workloads typically require clusters of GPUs, the cost differences become more significant.
These alternative cloud providers, referred to as specialty “GPU as a service” cloud providers, offer an alternative route to accessing GPUs. They have gained traction even among big tech firms facing compute capacity challenges. For instance, Microsoft signed a multi-billion-dollar deal with CoreWeave to ensure sufficient compute power for OpenAI’s generative AI models.
According to Sid Nag from Gartner, these alternative cloud providers have found success because they can offer access to GPUs without the infrastructure burdens faced by incumbent providers. The dominance of hyperscalers in the public cloud market requires substantial investments in infrastructure and services that may not generate significant revenue. In contrast, challengers like CoreWeave can focus on premium AI services without such burdens.
However, there are concerns about the sustainability of this growth. Lee Sustar from Forrester believes that alternative cloud providers’ expansion will depend on their ability to bring GPUs online in high volume and offer competitive prices. Incumbents like Google, Microsoft, and AWS are investing in custom hardware, such as Google’s TPUs and Microsoft’s Azure Maia and Azure Cobalt chips, to run and train models. This could pose a challenge for alternative providers competing solely on pricing.
Additionally, not all generative AI workloads require GPUs, particularly those that are not time-sensitive. CPUs can perform the necessary calculations, albeit at a slower pace. Moreover, there is a possibility that the generative AI bubble may burst, leading to a surplus of GPUs and a lack of demand. However, in the short term, industry analysts like Sustar and Nag expect a steady stream of upstart cloud providers in the GPU-oriented AI space.
In conclusion, the alternative cloud market is thriving due to the growing demand for GPUs to support generative AI models. While major cloud providers offer GPU instances, alternative providers can offer cost advantages and better availability. However, their sustainability depends on factors such as competitive pricing, the development of custom hardware by incumbents, and the longevity of the generative AI trend. Nonetheless, there is optimism for continued growth in the short term, especially among customers who are comfortable with multi-cloud environments and value credible leadership, financial backing, and timely GPU availability.