- Demand for AI continues to grow, but more AI doesn’t mean better returns
- Businesses are underestimating costs as they go from pilot to production
- Seat- and consumption-based pricing both have their places, it seems
Businesses are at risk of confusing generic AI adoption within their workforces with actual meaningful value, Oracle’s Head of AI Strategy Hammad Hussain warns, with many of its customers now struggling to manage increasingly unpredictable AI costs.
We already know that three in five businesses agree unpredictable billing hikes cause them more strain than a higher but flatter baseline cost, thanks to a recent Splunk study, but it seems cloud vendors already know this too.
Speaking with a group of journalists at Oracle’s recent AI World event, Hussain explained that while demand for AI continue to grow, businesses need to focus more on measurable outcomes than just adopting AI for the sake of it.
Latest Videos FromTechRadar
AI adoption does not equal value
“One of the fatal mistakes that clients can do… is they might perceive usage as value,” Hussain said. “And that’s not the case.”
Companies should consider whether they’re creating more experiences, more efficiencies, more products, and so one – rather than just whether they’re using more AI, at the risk of doing so blindly.
This all comes as they struggle to maintain a sort of balancing act between investing enough in AI to make a meaningful impact on business outcomes while still maintaining cost visibility and control. It’s that second element that many are now struggling with.
In fact, even the largest enterprises and early AI adopters who have experienced considerable success are looking at ways to scale back costs and only spend on the AI that’s necessary.
Unpredictable costs are now becoming a major focus
Multiple studies have already uncovered the impact that unpredictable costs are having on AI consumers, and this follows a period of transition from flat, seat-based models to variable, consumption-based pricing as AI vendors themselves look to make the economics work.
Hussain warned that many customers risk underestimating the financial implications of rolling out tools on a bigger scale. “The production economics is quite different from pilot economics,” he added.
Looking ahead, Hussain implied that a hybrid approach is emerging as a ‘best of both worlds’ scenario. Companies can still use the latest and most advanced frontier models to handle their biggest workloads, but some of the lighter work might just need conventional CPUs without the expense of GPUs and advanced AI altogether.
Somewhere in the middle seems to be a new solution that we’re only really starting to see at scale this year – leaning on open-weight models.
“One way of [balancing costs and impact] is also using dedicated AI clusters, where they pay fixed amounts for a cluster, which might have open-weight models, so they’re not paying per tokens.”
Hussain explained that having access to a pre-defined set of compute resources and a slightly less advanced model gives companies a space to throw their less demanding work for a cheaper – but importantly, more predictable – price.
Unlike accessing a third-party model through an API, running an open-weight model on their own dedicated infrastructure allows organizations to process certain workloads without incurring extra costs for tokens.
Of course, they’ll still need to pay for the underlying compute infrastructure, but it offers a far more predictable and budgeting-friendly cost.
Oracle’s AI Strategy Head doesn’t expect consumption-based pricing to disappear altogether. The pendulum has already swung from seats to tokens, but he now sees it swinging back to settle somewhere in the middle, where organizations will mostly be given the opportunity to choose which pricing strategy fits their needs best.
“We started off with fixed pricing,” Hussain shared. “But then we realized that some are heavy users and some are low users.”
More AI doesn’t mean higher or quicker ROI
But regardless of how customers choose to pay for their AI, Hussain argues that the real measure of success should be the returns those investments actually make, and more isn’t always better.
Defining AI ROI has been an unresolved topic for years, now, but Hussain offered a glimpse into some of the calculations companies can make to both determine whether they’re been a success, but also to help guide their strategies initially.
On the upper side of a fraction-like equation Hussain detailed were benefit and adoption – AI investments need to create some sort of benefit, whether that’s unlocking more productivity, driving more revenue or something else – but workers also need to adopt the new AI. “Benefit multiplied by zero is zero,” he warned.
But then you’ve got the balancing act on the bottom half of that fraction – cost. Direct costs that stem from buying AI and compute, but also indirect costs relating to upgrading infrastructure and training employees.
Our discussion also revealed that only around 20% of organizations capture 75% of the value, and Hussain thinks he knows why.
“The thing which separates that 20% from the rest of the 80% is the fact that they can rewire their business, or they’re truly reimagining their business using AI.”
The future might not be so clear, but the underlying message is – personalization is key, and emerging hybrid approaches probably hold at least some of the keys to that.
That’s hybrid in terms of how much workload actually needs to be handled by AI in the first place, but also hybrid in the way that consumers actually pay for their compute.
Ultimately, with AI costs becoming harder to predict, the most successful adopters will become the ones who find the most effective ways to turn investments into ROI – not just the ones that have the biggest budgets and the most extensive AI portfolios.
Follow TechRadar on Google News and add us as a preferred source to get our expert news, reviews, and opinion in your feeds.
https://cdn.mos.cms.futurecdn.net/DanvsyBYwaEUGw7m7DUXzn-2560-80.jpg
Source link
Craig Hale
