AI Growth Fuels Alternative Clouds 

Imagine you followed the gold rush into the cryptocurrency space and were left high and dry over the multitude of regulatory challenges? Do you cry over spilt milk or just pivot over into a provider of GPU infrastructure to the next gold rush? That’s precisely what CoreWeave has done and is today sitting pretty with $1.1 billion in new funding and a $19Bn valuation. 

So what’s changed? Quietly simply the appetite for alternative clouds has just gone through the roof – thanks to the growing commitment to generative alternative intelligence (GenAI) as the magic wand that can help enterprises reinvent their processes, put the resources freed up on more constructive uses and save costs. 

Alternative cloud based GPUs are in

CoreWeave, backed by NVidia, almost tripled its valuation within just five months of deciding to pivot from crypto to infra to support the growing computing demand that is required for AI to flourish. When the company managed to raise $5 billion in debt and equity, it also raised several eyebrows as CoreWeave is less than a decade old and such valuations appear extravagant. 

Just so that you don’t think CoreWeave hit the jackpot leaving nothing for others, we would like to share that they’re not alone. Lambda Labs offers cloud-hosted GPU instances and early last month got a $500 million worth special purpose vehicle close on the heels of a Series-C round of $320 million. And there are many more. 

GenAI needs costly computing hardware

The reasons are quite obvious as the GenAI boom is resulting in fast-growing demand for hardware that can run and train generative AI models. What’s more, as companies are pushing the agenda, more operational areas are seeking use cases for GenAI, which means data capture, and training needs to be at scale. 

Which is where GPUs come into play as the right choice for training models, fine-tuning them and running them. The presence of thousands of cores working in parallel to perform linear algebraic equations make these GPUs the logical choice for everyone dabbling with AI and GenAI out there. Small wonder that alternative clouds are back in vogue. Not that they ever went anywhere! 

In the past GPUs were in vogue for companies handling massive loads of data and analytics in select industries such as insurance, fintech, healthcare etc. Now, with more industries experimenting with GenAI, the need for a cheaper alternative installing one’s own GPU became an automatic requirement. And the answer, as always, lay in shared usage or cloud.

Alternative clouds cost much less

We have already seen robust growth among providers of cloud computing space – made up essentially of three big names viz., Amazon Web Services, Microsoft Azure and Google Cloud. These three corner nearly two-thirds of the market and are eyeing more through a mix of software and hardware alignment for even smaller customers. 

However, now it is becoming evident that for running and training some GenAI models, alternative clouds could be a cheaper alternative with quality delivery guaranteed. Here’s an example: CoreWeave rents an Nvidia A100 4GB at $2.39 an hour that works out to just $1200 a month, which on Azure costs $2482 and on Google Cloud $2682. 

And since genAI workloads are performed on GPU clusters, the costs could quickly mushroom out of control. Which is why industry analysts are confident that this new age GPU-as-a-service makes overall sense, as we had stated in our earlier post published in March 2023 about how Nvidia was offering super computing as a service. 

What began a year ago is now showing results. And one doesn’t need to look beyond how Microsoft actually signed a multi-billion dollar deal with CoreWeave to ensure adequate computing power for its partner OpenAI. Nvidia saw it coming and reportedly offered some alternative cloud providers preferential access to its GPUs last year. 

But wait! Hyperscalers in the race too

Of course, for the current spell of growth to sustain, these alternative cloud providers would have to really hustle on growing the GPU volumes and offering them at lower prices than the hyperscalers. The price factor could be a challenge as the Big-3 are ramping up their own investments into custom hardware that can run and train AI models. 

Google already offers its TPUs while Microsoft came up with two custom chips – Azure Maia and Azure Cobalt in order to catch up with AWS and their Trainium, Inferentia and Graviton.  That hyperscalers will leverage custom silicon to nudge Nvidia out is a given, what remains to be seen is whether the likes of CoreWeave can innovate faster and pivot once again. 

Of course, there is also that real threat of the GenAI bubble actually bursting at some point in the near future. Given that most experiments are still awaiting results, such an occurrence cannot be ruled out and if it does happen, we might just have an oversupply that could cause prices to automatically crash.