Artificial intelligence (AI) technology is currently “more expensive than it should be” according to Amazon CEO Andy Jassy.
Jassy spoke to CNBC this week to discuss how the company is responding to the release of DeepSeek and what must happen to bring the cost of AI down.
How to bring down the cost of AI
When Chinese-made AI model DeepSeek burst onto the scene in January, people were excited about its perceived lower cost benefits.
“We would love nothing better than to make the cost of AI meaningfully less expensive for customers,” Jassy said. “The lower we can make the cost of AI, the more customers are going to use it.”
Customers love when you take the cost per unit of something down, Jassy added. “It allows them to save money in what they’re building – but they don’t actually spend less. It unleashes them to do a lot more innovation – and spend more.”
Jassy was asked what it will take to bring the cost of AI down – and how long it will take.
“If you sat in the meetings with the AWS team right now, [you’d see that] they feel it’s their responsibility and their mission to make the cost of AI meaningfully less for customers than it is today,” he said.
There are two big drivers here, Jassy added. “What’s interesting is that most of the spend in AI has been on training [models]. When you get to the scale of production applications for AI, it flips. The overwhelming majority of the spend is on the inference, which are the predictions for the model.”
The two things that need to happen to bring down the cost of AI is for the price performance of the chips to be meaningfully less (partly why Amazon has built its own custom chips) and for the cost of inference to come down, according to Jassy. “That involves a whole bunch of things including techniques like distillation, prompt caching and how you handle different model architectures.”
Amazon is working hard in these areas to drive meaningful reductions in the cost of AI, Jassy concluded.
Join us at All Access: AI in PEX 2025 to learn how to successfully integrate AI into process improvement initiatives
Time to assess AI economics
“I’m relieved to hear a prominent AI leader finally address the elephant in the room: the high cost of AI inference,” commented Mel Morris, co-founder and CEO of Corpora.ai.
“Right now, many AI vendors absorb these expenses by incurring multi-billion-dollar losses, propped up by investors hoping for future payoffs,” Morris said. “While Jassy is right that costs must fall dramatically, 30 to 40 percent gains in hardware performance alone won’t suffice. Vendors may need to reduce service delivery costs by a factor of five – or more – for AI to achieve near-term profitability.”
It’s time for both AI vendors and investors to take a hard look at the economics, rather than focusing solely on ever more sophisticated models. “Sam Altman’s talk of $25,000-per-month AI PhD ‘researchers’ or expecting hundreds of millions of users to pay $200 a month is fanciful at best – underscoring just how far we still have to go,” said Morris.
AI is the biggest OPEX and business transformation investment area
AI is the biggest operational excellence (OPEX) and business transformation investment area right now, according to the PEX Report 2024/25.
The report compiles and examines the results of a recent survey of almost 200 professionals, painting a comprehensive picture of today’s process excellence environment. Almost half of businesses (47 percent) said they are looking to invest in AI solutions, with 58 percent having discussed potential AI projects.
The areas where AI is being leveraged the most are in operations (35 percent), customer service and call centers (29 percent) and data processing and management (24 percent).
Modern organizations are heavily investing in conversational AI assistants, which are revolutionizing both front-office and back-office operations, said Ricardo Henriques, business automation professor at the Católica Lisbon School of Business and Economics and PEX Network Advisory Board member. “These solutions, such as natural voice interactive voice response systems (IVRs), promote self-service operations by removing the need for human intervention, thus enhancing customer experience.”
[inlinead-1]