Artificial Intelligence applications, once fully operationalised, require significant compute power and data storage. So while public cloud vendors offer enticing pricing models for AI as a service, when it comes time to move beyond experimentation it is often more cost effective to run the systems on premise.
That is according to vendor Pure Storage which last year began offering “AI in a box” in an attempt to address the pricing pain point and begin standardising AI infrastructure.
The flash storage vendor calls its system AIRI, or Artificial Intelligence Ready Infrastructure. It’s a hardware/software package designed specifically for AI use cases, using flash storage arrays and enterprise level GPUs.
And while Pure claims its offering works out cheaper than the same projects running on cloud within two to three years, the upfront costs are considerable – starting at over a million dollars.
Sunil Chavan, Pure Storage’s VP APJ – Flashblade, says AIRI opens AI up to enterprises but more importantly it removes the administrative or low value-add tasks data scientists are dealing with as a result of disparate systems and data storage.
“You can do your research or your use case much faster, we’re offering that infrastructure,” Chavan told Which-50 at Pure’s annual Accelerate conference in Austin.
“It’s really cheap, actually, compared to what they’ll spend if they really have to work it out with different vendors together, they’ll spend a lot more money.”
Chavan said the price tag and hardware means AIRI customers are typically government agencies, researchers, or large private enterprises like banks or telcos.
“The plans they have, they don’t have a choice but to have that AI infrastructure with them [on premise]. They can’t work on a cloud based or a small infrastructure. It’s too expensive.”
Chavan does urge customers to use cloud to test out their AI use cases in the cloud, however, and to invest in AIRI when they are confident in their plans.
What is an AIRI?
The AIRI hardware combines a Pure Storage FlashBlade all flash array with four Nvidia GPUs, with networking handled by either Arista or Ciscon depending on the model. The promise is an easily deployable converged infrastructure designed specifically for AI and machine learning.
Designed by Pure Storage and Nvidia, the advantage over traditional infrastructure, the vendors say, is a removal of complexity and disparate AI infrastructure.
The included software will reduce the time it takes data scientists to establish machine learning projects from weeks to hours, according to Pure.
Chavan says AI workloads and the troves of data they rely on typically sit across different storage and applications that rarely work well together.
“That’s the bridge we built … At one storage infrastructure level I can start aggregating, to processing, to deep learning, to virtualisation. [The] entire process I help manage at one storage level.
Chavan says the product will free up data scientists time – an increasingly prized resource as demand for AI talent increases.
“That’s the biggest issue [data scientists] have: I need an infrastructure where ‘I don’t want to know what’s there. If I need that performance I just need my data to be available at that time, at that speed.’ That’s it, that’s the challenge.”
Use cases and ethical boundaries
Popular use cases so far for AIRI, Chavan, explained, are immigration departments looking to process data at border points. They need to do it quickly, safely, and often face restrictions on the use of cloud technologies or data sovereignty.
AI in a box has also been popular in health-tech, investment banking, wealth management, and anti money laundering applications, where AI and machine learning are needed to analyse huge amounts of transactions.
Australia’s ANZ bank is a Pure Storage customer and Which-50 understands there are also ongoing conversations with fellow big four members.
Chavan says there are customers and use cases which Pure Storage will not sell to. While from a business perspective responsibility ends at the sale, the Pure Storage executive says there is a greater social responsibility in the use of its technology.
“There are certain businesses we refuse to give our AIRI Infrastructure because we just felt that, ethically, it’s not right to work with that [use case]. So we evaluate that side of it as well.”
Disclaimer: The author travelled to Pure Accelerate as a guest of Pure Storage.