Amazon’s new cloud boss inherits a business under pressure to keep pace in AI

“It’s very early days in generative AI,” said Jassy, who succeeded Jeff Bezos as CEO in 2021. “It’s very high potential, and we’re investing quite a bit in it and expect to be a leader.”

For Selipsky, who took over AWS when Jassy was promoted, the days got late quickly.

In the most significant shake-up of Jassy’s tenure at the helm, Amazon announced last week that Selipsky, 57, was exiting AWS and will be succeeded by Matt Garman, 48, a veteran AWS executive who most recently led sales and marketing.

The problem for Selipsky and the challenge for Garman is that Amazon has yet to emerge as a leader in generative AI despite throwing billions of dollars behind OpenAI competitor Anthropic and rolling out its own large language models, or LLMs. In the developer universe and among startups, the company is battling the perception that it’s falling behind cloud rivals Microsoft and Google, in addition to lagging OpenAI in developing AI tools.

After years of rapid expansion, growth at AWS decelerated to 13% in 2023, down from 37% in 2021 and 29% in 2022, reflecting more conservative spending by businesses on IT and cloud services. Amazon has downsized across the board, including at least two rounds of layoffs at AWS since last year.

AWS remains the leader in cloud infrastructure, but Microsoft is quickly closing the gap. AWS’ market share slipped to 31% in the first quarter of this year from 32% three years earlier, while Microsoft Azure jumped to 25% of the market from 19% in 2021, according to Canalys. Google is also picking up share, accounting for 10% of the market, up from 7% in early 2021.

In the past few quarters, Microsoft has cited surging demand for AI tools as a catalyst for its momentum.

Gil Luria, an analyst at D.A. Davidson, told CNBC that Amazon was “caught flat-footed” by the generative AI boom.

“It allowed Microsoft Azure to run laps around them, which should not have happened, and ultimately there was a price to pay for that,” Luria said, referring to Selipsky’s departure.

Garman’s selection for the top job “indicates Mr. Jassy and perhaps Mr. Bezos believe he’s the person most likely to help Amazon close the lead and maybe establish a lead of their own,” said Luria, who recommends buying Amazon shares.

A source close to Amazon, who asked not to be named because they weren’t authorized to speak on the matter, described Garman to CNBC as a “wartime” leader and said change was needed to get more aggressive in AI. 

Jassy said in a staff memo announcing the move that he and Selipsky agreed years ago, when they were discussing the role, that Selipsky would “likely do it for a few years, and that one of the things he’d focus on during that time was helping prepare the next generation of leadership.” 

Casey McGee, an AWS spokesperson, told CNBC in a statement that Selipsky is leaving the cloud division in a “strong position.”

“The growth, innovation, and profitability of AWS over the past three years speaks for itself, with AWS producing more absolute dollar growth quarter-over-quarter by our numbers so far this year than any other cloud provider,” McGee said. He said AWS is leading in security and reliability as well as the “overall breadth and depth of our services.”

Amazon’s annual shareholder meeting, which took place virtually Wednesday, arrived at a precarious time. It was held just days after Selipsky’s departure and was overshadowed by AI-focused events at top tech companies.

Last week, OpenAI introduced GPT-4o, a faster model with improved capabilities in text, video and audio. Google followed a day later at its developer conference, rolling out the company’s lightest and most efficient AI models. And this week, Microsoft announced new computers with advanced chips designed to run AI features in Windows.

During a Q&A session on Wednesday, Jassy was asked twice about the status of Amazon’s generative AI efforts. He said the company is “seeing a lot of momentum” in generative AI within AWS to where it’s now a multibillion-dollar business based on annualized revenue.

He reminded shareholders that Amazon owns Alexa, which was a popular consumer offering long before the latest chatbots hit the market.

“If you don’t believe there’s going to be a really broad personal assistant, you have your head in the sand,” Jassy said, adding that the company is building a “much more expansive” AI model to power Alexa. Amazon has previously said it intends to use generative AI to make Alexa more conversational. CNBC reported on Wednesday that Amazon plans to charge a subscription fee for the more powerful version.

Garman joined Amazon in 2005 as an intern, and was hired full time the following year as an early product manager in AWS, working on the core computing service called EC2. He worked his way up to senior vice president in 2020, overseeing sales, marketing and global services.

In 2021, after Amazon announced that Jassy would be taking over the CEO role from Bezos, many people speculated that Garman would be named CEO of AWS. Instead, Amazon chose Selipsky, who had previously spent 11 years at Amazon but was running Salesforce’s Tableau Software at the time.

Shortly after the transition, the economy turned against AWS. Inflation started to rapidly pick up, leading to a steady rise in interest rates and forcing businesses into capital preservation mode. By mid-2022, Amazon was telling investors that it was “prepared to help customers optimize their costs” due to the economic challenges they were facing. AWS acknowledged it was taking a short-term revenue hit to preserve customer relationships over the long term.

Then came ChatGPT. OpenAI, backed by Microsoft, released the chatbot in November 2022 and watched it go viral. Months later, Microsoft invested billions more into OpenAI, and became its exclusive cloud partner, giving Amazon’s chief cloud rival a new competitive edge. 

Over the course of the past year, Jassy has gushed about Amazon’s opportunities in generative AI, both for offering automated services to advertisers and sellers and in providing technology within AWS for running sophisticated models and workloads.

The company has also boasted about the success of AWS’ Trainium and Inferentia chips, with Anthropic using them to build and train its models, a process often performed on Nvidia graphics processing units.

“I don’t know if any of us has seen a possibility like this in technology in a really long time, for sure since the cloud, perhaps since the internet,” Jassy said on the company’s first-quarter earnings call in April, speaking about generative AI.

But realizing that opportunity is proving to be a major hurdle.

AWS took months to come out with an AI model that could go up against ChatGPT. The company is now offering its own LLMs as well as those from third parties, including one from Anthropic, which Amazon backed.

Last year, Amazon released Q, a chatbot for businesses. An AWS employee, who used Q, told CNBC that it felt underwhelming because the chatbot would respond to queries with information that wasn’t particularly relevant or valuable. The employee asked to remain anonymous because he wasn’t authorized to speak on the matter.

AWS said its Q chatbot has been gaining traction among a range of customers including Accenture, Toyota, GoDaddy and GitLab. Bedrock, which lets users access AI models from Amazon and others, now has tens of thousands of customers and partners, the company said.

The week before his departure, Selipsky made some changes to the Q team. He tapped Dilip Kumar, a longtime Amazon executive who helped develop and launch its cashierless checkout technology, to oversee the “Amazon Q Business set of services,” according to a memo sent to employees that was viewed by CNBC. Kumar will report to Swami Sivasubramanian, vice president of AI and data at AWS.

A former AWS employee, who asked not to be named in order to speak about private matters, said the company has limited the ability of some staffers to use AI services like its software tool SageMaker and data visualization tool QuickSight for work internally, citing security reasons. The practice, known as dogfooding, is commonly used at software companies so employees can test products and services for bugs and to help make improvements.

AWS said all applications made available to employees are subject to a security review, but it denied that it’s restricted staffers’ use of Amazon’s AI tools.

Despite all of its AI challenges, Wall Street continues to rally around Amazon, which last month reported better-than-expected first-quarter results and a more than 200% increase in operating income. Sales at AWS rose 17%, a modest acceleration from the past few quarters.

Amazon shares are up 21% this year, topping the 12% gain in the Nasdaq, after jumping 81% in 2023. The stock hit a record earlier this month.

Jamie Meyers, a senior investment analyst at Laffer Tengler Investments, which owns shares of Amazon, said he views the leadership transition at AWS as a “natural progression,” adding that Garman has “always been looked at as a successor.” 

“AWS has always been about investing in growth,” Meyers said, a strategy he said is unlikely to change under Garman.

Garman is viewed internally as someone who is highly technical, and is well respected among engineers. At the time Jassy picked Garman to lead the AWS sales organization in 2020, he was looking for a technical leader and a person who “knew everything inside out,” another former AWS employee said. Garman’s appointment to the role was widely seen internally as a step toward priming him to lead AWS, the person added.

In his memo to staffers last week, Jassy referenced Garman’s background “on both the product and demand generation sides” of AWS, noting he has “an unusually strong set of skills and experiences for his new role.”

“I’m excited to see Matt and his outstanding AWS leadership team continue to invent our future,” Jassy said. “It’s still such early days in AWS.”

— CNBC’s Jordan Novet and Kate Rooney contributed to this report.

Correction: CNBC reported Wednesday that Amazon plans to charge a subscription fee for a more powerful version of Alexa. An earlier version misstated the day.