Here's how AI is driving the rise of GPU cloud specialists

Both Microsoft and Google Cloud hailed artificial intelligence (AI) as a key growth driver in their recent Q1 2023 earnings calls. But beyond giving tech giants a boost, the technology is also fueling the rise of a new class of cloud provider: the GPU specialist. 

Silverlinings profiled one of these players – CoreWeave – last month. However, the company is just one of a growing cohort flourishing in the cloud giants’ shadow. Others in this space include the likes of Lambda Labs, RunPod and Arc Compute. 

All are trying to solve the same problem: the fact that AI compute workloads – especially AI training workloads, like those needed to create something like ChatGPT – are inherently pricey because they require GPU resources. The idea is that by specializing in only GPU compute, these companies can provide lower prices than the big three generalized cloud competitors.  

For instance, on-demand access to NVIDIA’s A100 series of GPUs costs $3.37 per hour with Google Cloud but starts at $2.06 with CoreWeave, $2.04 per hour at RunPod, $1.69 per hour (at the weekly rate) at Arc Compute and $1.10 per hour at Lambda Labs. 

The GPU players 

Lambda Labs appears to be one of the older GPU specialists. Founded in 2012 to help develop facial recognition technology, Lambda launched its internal GPU cloud in 2015 and a public GPU Cloud product in 2018 dedicated to serving deep learning use cases. In 2021 it secured a total of $24.5 million in financing to scale its GPU Cloud and on-prem AI infrastructure products. And this past March, it bagged another $44 million in funding to deploy next-generation H100 GPU capacity and develop new features. 

“Lambda solves difficult problems at the intersection of AI, data center-scale computing and GPU virtualization,” Lambda’s co-founder and CEO Stephen Balaban said in a video announcing the news in March. “Over the past couple of years, we’ve seen extreme growth in our cloud product.” Though Lambda doesn’t disclose earnings since it is a private company, Balaban displayed a chart indicating the company’s revenue is expected to more than triple year on year in 2023. 

Arc Compute was founded in 2020 to tackle the same market. It initially provided NVIDIA GPU instances, but subsequently became a provider of GPU optimization products with the development of the GVM Server. 

Then there’s RunPod, which according to its LinkedIn page was founded in 2022. It, too, is catering to the GPU niche with container-based instances, bare-metal and virtual machine products. A serverless GPU platform is currently in closed beta and it is working toward a full virtual machine (VM) solution slated for release in the back half of this year. 

GPU expertise needed 

Hariprasad Pichai, principal at telecom and cloud advisory firm Arthur D. Little, told Silverlinings the reason a company might choose a niche vendor over a cloud giant likely either relates to the vendor’s geographic location or their ability to offer access to scare compute resources. There’s also the price consideration, but that’s more likely second fiddle to access, he said.  

“Typically, what they have as a value proposition is a better location or a range of locations not reachable by the public cloud. Number two, more tactically they have capacity and bandwidth that’s usable in chunks and attractively priced. And the last one is the actual compute, storage and networking equipment that are tailored to specialized workloads” Pichai explained. “With all the supply chain challenges, ready availability of compute becomes a near-term differentiating factor. If you have access to a fleet of GPUs or other specialized compute and you’re able to provide those services at reasonable prices, you are in the game because those are hard to come by.” 

Right now, specialized GPU cloud providers are catering to AI and other data intensive workloads. But Pichai said there could be the need for other cloud niches in the future, for instance around streaming and video processing, gaming, music, virtual reality or certain edge use cases. 

There could also be niche clouds for certain verticals – think the oil and gas or mining industries – or geographies, with the latter designed to comply with data sovereignty laws.  

Options in the latter category are starting to emerge. In late 2021, Deutsche Telekom’s T-Systems division partnered with Google Cloud to build a sovereign cloud service for German enterprises.