- AWS sees plenty of runway for public cloud growth
- But rising demand for private cloud offerings could throw a wrench in the works
- Private cloud has emerged as the answer for companies looking to run AI while also protecting their data
If you ask Amazon Web Services CEO Matt Garman, the public cloud has plenty of runway. Speaking at a Goldman Sachs conference this week, Garman said only 10-20% of workloads have moved to the cloud, meaning the “vast majority” are still there for the taking. But despite Garman’s optimism that the public cloud will be the primary beneficiary, an emerging trend toward private cloud offerings could throw a wrench in hyperscalers’ plans.
The private cloud appears to have popped up in answer to demand from customers who want to take advantage of artificial intelligence (AI) tools but don’t want to surrender sensitive data.
AvidThink founder Roy Chua told Fierce that there is already a fairly broad spectrum of private cloud options. These include virtual private clouds within the public cloud that are used to train or fine-tune foundation models “so that the data doesn't get sent to some multi-tenant AI service somewhere” as well as private clouds built into multi-tenant data centers or colocation facilities from the likes of Equinix and others. There’s also on-premises private cloud infrastructure that a company owns or otherwise directly controls.
Chua said these are just a few of the offerings out there today, though there are likely “many different combinations I'm sure that will emerge.”
Kevin Sullivan, Global and US Oracle Alliance Practice Leader at consulting firm PwC, told Fierce it particularly sees private cloud as an option for on-prem and edge AI deployments.
"Within our work we do see an angle for private cloud in products and apps where there are extremely large volumes of proprietary data on prem and edge device, which require a highly tuned LLM and tight control over infrastructure performance," Sullivan said. "Think embedding in products as opposed to traditional enterprise processes."
Indeed, Apple has already come up with its own Private Cloud Compute solution to go along with the AI it’s rolling out for its suite of devices.
As Apple explained, the idea is to bring its on-device security to the cloud to ensure customers’ personal data can be used for AI but never shared with anyone.
“The root of trust for Private Cloud Compute is our compute node: custom-built server hardware that brings the power and security of Apple silicon to the data center, with the same hardware security technologies used in iPhone,” Apple said in a blog. “We paired this hardware with a new operating system: a hardened subset of the foundations of iOS and macOS tailored to support Large Language Model (LLM) inference workloads while presenting an extremely narrow attack surface.”
New Street Research's Antoine Chkaiban told Fierce that the fact that Apple is building its own data centers makes sense given it's scale. "We expect that to be the case for another dozen of Enterprises beyond Apple: Tesla, successful AI start-ups building their own (x.AI), big-pharma, etc."
Private cloud on the rise
Kendall Clark, CEO of enterprise knowledge graph company Stardog, told Fierce there are three primary things driving private cloud adoption: a desire for data privacy, a need for data integration into AI and a need to close the distance between where the data resides and where the AI runs. There’s also the realization that the best economics come from owning your own compute infrastructure, especially when it comes to AI.
Stardog itself came to this realization, he said, and built its own GPU private cloud infrastructure to run workloads for its Stardog Voicebox AI offering.
All of those things, he added, are putting pressure on the idea that the public cloud is the end goal for enterprises.
“I think private GenAI will flatten that adoption curve unless the hyperscalers can do something else,” Clark said, referencing Garman’s comments about the public cloud’s growth opportunity.
While hyperscalers’ current strategy is to “buy all the GPUs,” Clark speculated that Nvidia is going to “realize at some point that having that customer concentration is bad for their stock and bad for their business and start thinking about how they can sell more GPUs to [companies] other than those four big whales.”
“Without GenAI the cloud migration story, the cloud adoption story would look much, much better for the shareholders of the hyperscalers than it’s going to look for the next few years,” he concluded. “Public cloud growth and private cloud growth, I would suspect, will look more like 50-50 than we would have said three years ago. Nobody would have said that three years ago.”
According to Chkaiban, New Street has predicted that in 2027 enterprise private AI infrastructure will account for 25-30% of the total market, with hyperscalers' internal AI infrastructure accounting for another 25-30%. Public cloud will still make up around 45% of the market, he said.
Update 9/11/2024 10:17 am ET: This story has been updated to include comments from New Street Research and PwC.