Enterprise AI: Free, premium or a bolt-on?

The latest data from analyst Gartner shows a 10% spike in the datacentre market, driven predominantly by hyperscalers deploying artificial intelligence (AI) servers.

AI in enterprise systems is being marketed as a way to drive productivity and improve efficiency, but there is a cost associated with running AI workloads, and someone has to pay for it.

Given the average cost of an AI server is $32,000, Gartner distinguished analyst John-David Lovelock points out a rack of AI servers will cost over $1m. While the hyperscalers have the funds to build out vast AI infrastructure and the competition among them will keep prices of AI services in check, the same is not true of software-as-a-service (SaaS) providers.

When Computer Weekly raised the question of AI price increases in a recent conversation with Salesforce, its senior vice-president of AI, Jayesh Govindarajan, said: “If the best way to deliver value is to make people 10 times more productive, then we should go off and do that. If you’re truly delivering that value, which is making somebody 10 times more effective, there are people who would be willing to pay.”

During its recent SuiteConnect London event, Brian Chess, senior vice-president of AI at NetSuite, said the company benefited from the partnership Nvidia has with Oracle, which provides access to the graphics processor units (GPUs) needed to run AI workloads in the Oracle Cloud. But he acknowledged the costs associated with operating AI infrastructure, and said Netsuite would need to assess whether it charged more to justify further investment in AI. “We do look at price versus business advantage,” said Chess. “We see a really nice marriage to justify the cost.”

AI is beginning to look like an arms race in enterprise software. Lovelock said companies are starting to give away AI functionality, which means it becomes a fixed cost that has an impact on their revenue. “For those enterprise software companies that are in the cloud, they have no choice but to take on this cost, providing the compute capacity to run AI will mean their operating costs will be going up,” he said.

Lovelock said the challenge for SaaS providers is that this cost is variable. “We still don’t know what the long-term average usage is going to be per user, so at this point, they are in a bit of a precarious situation,” he said. “They either have to charge a reasonable amount for AI functionality, or they’re going to take it out of their profit. Either way, their cost as a percentage of revenue is fixed.”

Gartner has identified a number of categories in the enterprise software market where there is genuine competition, which, according to Lovelock, means SaaS providers will need to offer AI for free or as a bolt-on that IT decision-makers can choose to purchase.

“Their costs could be highly variable,” he said, adding that over the next year, he expects SaaS providers will be busy assessing the average usage per person and the average cost to fulfil that usage. “They’ve got some great models and some great estimates, but we’re in an unknown territory,” said Lovelock.

Looking at a hypothetical example of how two competing SaaS providers could go head-to-head on AI functionality, he said: “The first question is can we change pricing?” If, for instance, Microsoft raised the price of its Outlook email service significantly to incorporate Copilot AI functionality, Lovelock said there is a risk IT buyers would revolt and turn to an alternative such as Gmail if Google’s AI is free or at a lower price.

“The market doesn't give them pricing power,” he said. “The SaaS providers have to decide what their corporate strategy is.”

A SaaS provider that has previously focused on being the low-cost provider in the market may not offer the latest feature functionality, but Lovelock said it needs to offer AI to survive, and that such a company would not have the pricing power to be able to embed AI into its core products and charge everyone more.

Instead, some providers may well offer AI as an add-on. This, he said, would allow such software providers to offer those customers that desperately want the AI functionality they are happy to pay for. “For those that don’t need AI, there’s no harm, no price increase,” said Lovelock. This will lead to a second tier of software that is not AI-enabled.

SAP, for instance, is making AI features only available on its Rise cloud platform, but has now begun offering customers a 50% discount to switch from on-premise enterprise resource planning.