AWS turns 20, and AI makes its bet pay

AWS turns – From a skeptical start with Wall Street questioning early profits to a staggering 2025 revenue surge, Amazon Web Services has marked its 20th anniversary—now driven by AI services such as Bedrock, SageMaker, and agentic tools. With AWS holding a still-leading

When Matt Garman had lunch with a coworker back in the early days, it felt like talking about something both promising and impossible. The message from across the company was simple: “How is that AWS thing going? I heard about it, and it sounds pretty interesting.”

Garman’s reply carried the kind of conviction that gets tested later—“I think this could be a billion-dollar business for Amazon.” His lunch mate cautioned him about the daunting ambition of that goal.

In 2006, AWS was still a fledgling.. Amazon had taken cloud-computing technologies it had created for its own operations and turned them into a business. offering organizations the ability to build an online presence without managing infrastructure.. Even then, some analysts struggled to see how it would pay off.. One Wall Street analyst said, “I have yet to see how these investments are producing any profit.”

Two decades later, those doubts read differently.. AWS officially marked its 20th anniversary on March 13.. The date ties to the introduction of its Simple Storage Service. better known as S3. still one of its flagship products.. Earlier concepts had appeared as far back as 2002, when Amazon.com Web Services launched.

By 2025, AWS had reached $128.7 billion in revenue, after smashing through Garman’s $1 billion goal. Last year, it delivered Amazon’s majority of profit, totaling $45.6 billion. Garman, whose early faith had helped set his own path, became CEO in June 2024, succeeding Adam Selipsky.

S3 may be the anniversary anchor, but AI is the force rewriting what AWS means to businesses now.. Garman frames it as a shift that stretches beyond products.. “is a massive technology leap that changes everything about how technology is consumed.. It changes everything about how all of our customers are going to operate their businesses. how industries are going to work.” As an AI provider on demand. AWS is tasked with driving that change—and also absorbing the disruption it creates for its own category.

He calls AI “an enormous tailwind to our business already,” but adds that the hard part is only starting. “All technology disruptions should be viewed as both a threat and opportunity,” he cautions.

That balance—between momentum and uncertainty—runs through the rest of AWS’s 20-year story. because nothing has ever been guaranteed.. Garman points to a meme he says circulated as AWS gained traction: the idea that it would quickly become a commodity and normalize.. His team, he says, proved that narrative wrong through “incredible invention.”

Cloud’s competition is tighter than it once was, too.. AWS remains the dominant force in the category it created, but Microsoft and Google have narrowed the gap over time.. Back in the first quarter of 2020. AWS held 32% of the market compared with Azure’s 18% and Google Cloud’s 8%. according to Synergy Research Group.. By the first quarter of 2026, AWS’s share was 28%, Azure’s was 21%, and Google Cloud’s was 14%.

Still, AI changes how companies view the pie.. Garman argues that there’s been no real debate about whether most businesses will get their AI as a cloud service.. The reason is straightforward: the computational resources needed to make large language models (LLMs) run at scale make cloud delivery the practical route.

That AI-driven reality helps explain why AWS is doubling down on the infrastructure and the software layers that sit under modern applications.. And it also explains why the competitive fight feels different now—less about whether cloud is relevant. more about who can deliver AI capability quickly enough. reliably enough. and in ways customers can actually use.

AWS built its AI strategy long before the generative AI boom became an industry obsession. “We obviously didn’t project a lot of the generative AI explosion that’s happened in the world today,” Garman says. “But we’ve long known that [AI] was going to be critically important.”

One early step came in 2017. when Swami Sivasubramanian—who joined Amazon as a research intern a dozen years earlier—became AWS’s VP of AI.. Later that year. at its annual re:Invent conference in Las Vegas. AWS introduced SageMaker. a platform for creating. training. and otherwise wrangling machine-learning models.. SageMaker remains one of AWS’s core AI offerings, upgraded and expanded many times since.

At the time, Google’s TensorFlow dominated AI development.. AWS believed customers would come to value choice.. Sivasubramanian explains that even internally. building applications made it clear that “you need multiple models even for a single application to make it happen.” That realization fed into Bedrock. introduced in 2023.

Bedrock lets customers run dozens of AI models from major companies, including Amazon itself, Anthropic, Nvidia, DeepSeek, Qwen, Mistral, TwelveLabs, and—via a new partnership—OpenAI.

AWS also pursued more control over its computing backbone by building custom AI processors instead of relying entirely on Nvidia.. Amazon’s 2015 acquisition of Israeli startup Annapurna Labs led to multiple generations of chips for inference and training. most recently Trainium3. announced last December at re:Invent.

Custom chips such as Trainium2 trace back to that same 2015 acquisition. AWS’s chip-and-platform approach matters because AI workloads are not just a software challenge; they’re a hardware one too.

More recently, agentic AI has taken center stage. These are forms of technology that can perform complex tasks with some measure of autonomy. Sivasubramanian says agentic AI forced a personal realization: “AI agents will fundamentally change how we all work and live.”

To pursue that shift directly, he “spun” himself out. In March 2025, he gave up his old job as VP of AI to become VP of AWS Agentic AI, overseeing a group focused on creating products that are agent-centric.

By July, that investment had begun to appear in new AWS services.. Kiro is a coding environment that lets software engineers hand off heavy lifting to an LLM-powered agent.. Bedrock AgentCore helps them build agents of their own.. DevOps Agent. announced at December’s re:Invent 2025. monitors other AWS services to detect and resolve problems before they require human intervention.

Sivasubramanian says early big wins often come in software development, where agents accelerate the process of writing code.. He points to Thomson Reuters. which used an AWS agentic AI service called Transform to modernize older applications built using Microsoft’s .NET.. Work that once took three to four years, he says, now takes six to 12 months.

The value is not only for large companies. Sivasubramanian adds that “Even my 10-year-old daughter, who doesn’t fully know yet how to build in Python, was able to spin up and build a website to manage calendars for the entire household,” and that she built it on AWS.

AI’s expansion at AWS isn’t limited to infrastructure.. Colleen Aubrey joined AWS as a senior VP in 2024 after nearly two decades at Amazon, where she worked in advertising.. She moved from ads to infrastructure, but found the cultural transition easier than the jargon.. “The acronyms were totally different,” she says.

Aubrey wasn’t hired to replicate old infrastructure instincts.. Her job was to lead AWS’s expansion into business productivity applications.. She explains. “At Amazon. we’ve built many of our own applications. and we learned a lot from that. ” and her “hypothesis was that we could bring to life some of that learning for AWS customers in the form of business applications.” She adds that the timing matched what AI makes possible now. while also signaling where it might go.

In April. at an event in San Francisco. AWS introduced a line of cloud-based. AI-powered products aimed at automating common business processes.. Amazon Connect Decisions focuses on supply-chain management.. Amazon Connect Talent conducts job interviews.. Amazon Connect Health helps doctors’ offices with tasks such as scheduling and medical history review.. Amazon Connect Customer is the latest version of a customer service contact center platform AWS originally launched in 2017.

Amazon Connect Decisions uses AI to bring a conversational interface to supply-chain management.. Generative AI enables the Connect products to offer chat-like interfaces and voice input.. Aubrey says the goal is “to offer software ‘that works in a way where. as a person. I don’t have to learn how to use a new tool.. I interact with it in ways that are familiar.’”

That emphasis on usability comes after years in which AWS excelled at administrative dashboards for technologists to configure. manage. and monitor services.. But as anyone who has used similar tools from Azure and Google Cloud can attest. those interfaces aren’t exactly built for consumers.. To improve the experience, AWS hired Hector Ouilhet as AWS Solutions VP of Design in January 2025.

Ouilhet spent more than 14 years at Google. where he helped shape Material Design. the design language behind a cohesive set of tools for intuitive. recognizably Google-y interfaces.. He describes the AWS challenge as broader than visuals: “We build the whole thing ourselves in terms of the experience. Not only how it looks. but how it feels. how it sounds. how it behaves. how it interrupts. how it listens.” He says design is now “way broader. ” especially because AI both enables and demands new interaction approaches.

Ouilhet calls AWS’s approach to AI agent interfaces “humorphism.” He says he detailed its principles—such as “Route work to whoever can do it best” and “Synthesize and tailor information for the moment”—at a website he created and would be “delighted if other companies followed the lead.”

Approachability also sits behind the latest updates to Amazon Quick, an AI assistant introduced last year.. At the April event. AWS said Quick taps into business tools including Google Workspace. Microsoft’s Teams and Outlook. and Slack to support research and task automation.. AWS also announced new Quick apps for MacOS and Windows that aim to make Quick more directly competitive with Anthropic’s Claude Cowork.. It also started letting users sign up for Quick’s freemium service with a standard-issue Amazon account. so they can get started in minutes without confronting the broader AWS setup.

The rollout still has limits.. “The limit at the moment is about 300 employees,” says Jigar Thakkar, AWS’s VP of agentic AI for business.. A Microsoft veteran and a Teams co-creator, he joined AWS in January.. Thakkar adds that if users are much larger than that. they would want the enterprise account where “we do a lot more governance and security.”

Even as AWS grows into more customer-facing experiences, its core remains what it has always been good at: providing reliable ingredients that other innovations depend on. When those ingredients fail, outages can suddenly make AWS visible—because so many websites depend on it.

As AWS built its early customer base. those clients tended to be smaller companies eager for scale and willing to try something new.. Garman and Sivasubramanian both point to SmugMug, the photo-sharing service whose early embrace became a defining example.. SmugMug’s CEO. Don MacAskill. negotiated AWS’s initial asking price of 40¢ a gigabyte for cloud storage down to 15¢. then took the plunge.. In 2017. MacAskill recalled that “A lot of people told me I was crazy at the time. just tons and tons and tons.” He couldn’t be sure Amazon would stay committed.

Today, AWS counts some of the world’s best-known companies as customers, and AI is accelerating their consumption of its services.

At United Airlines, AWS is “part of everything we do,” according to its CIO Jason Birnbaum.. The airline began working with AWS in 2018. the same year it launched a customer-service program called “Every Flight Has a Story.” Instead of leaving travelers wondering what caused a takeoff delay. the initiative provided explanations—initially handcrafted by a human “storyteller.” Birnbaum says that gesture was “amazingly well-received—it just was tough to scale.”

Now. United uses AI to write more than half of those messages. which has helped it cover more scenarios and be more transparent with customers.. Birnbaum says passengers on more than half a million flights have received messages generated by AWS AI. calling it “a home run for us. and it’s been a home run for our customers.”

At Mondelēz International. the maker of Oreos. Clif bars. and Cadbury eggs. the company wasn’t an AWS shop when Chris Hesse joined as CTO in 2021.. Now, Hesse says the majority of its cloud runs on AWS services.. He also said the company recently rolled out the Quick assistant to 50. 000 office workers. a mass deployment he admits was early compared with Quick’s state when he decided to move forward.. “I saw things that were maybe not as polished, and I was afraid people would talk about that,” he says.. “But instead, everyone went, ‘Look at this thing that I built, look at this thing that it does.. This helps me so much.’ That kind of thing.”

Capital One, meanwhile, has been building on AWS for over a decade.. Its senior VP of infrastructure. Will Meyer. describes the bank as “a tech company that has this amazing risk management capability of a really savvy bank.” Recently. Capital One’s AI projects include an agentic car-buying experience for its auto loan business. AI assistance for 20. 000 human customer service agents. and AI-enhanced fraud case resolution.

Meyer says even a bank that thinks like a tech company couldn’t ramp up AI-infused products in parallel without AWS support.. “There’s this whole category of stuff that AWS calls ‘undifferentiated heavy lifting’ that we wanted to get our teams out of. ” he says.. “But for us, it’s also always been about tapping into the innovation that the cloud can deliver.. It’s not just renting hard disks and CPUs from someone.”

He adds that AWS has “been really good at just helping real customers solve real problems. And that’s a strategy I think is aging pretty well.”

That customer feedback loop also shapes what AWS builds next.. Garman says “Some of the very best information that we get on what to build next comes from really leaning into folks like Capital One and saying. ‘What are the [blockers] that would prevent you from putting everything on top of AWS?’” He adds AWS asks. “How do we help you have better security?. How do we help your development teams innovate faster?”

Sivasubramanian says listening matters, but not only in a literal sense. “Nine out of 10 times, we do exactly what customers want,” he says. “And one out of 10 times, we read between the lines and [conclude] they’re asking for a faster horse instead of a car. Then we build a car.”

AWS’s ability to keep up with customers’ expanding needs appears to be paying off even as Microsoft and Google intensify competition. In the first quarter of 2026, AWS’s revenue was $37.6 billion, up 28%. Operating income was $14.2 billion, up 23%.

AI is playing a central role in that acceleration. Bedrock’s model platform now has 125,000 customers, including 80% of the Fortune 500. During the quarter, Bedrock processed more tokens than in its entire prior history, driving 170% quarter-over-quarter revenue growth.

Garman says it’s rare to see a business opportunity grow as fast as AWS while still having more runway ahead.. “You don’t often find a business opportunity that’s grown as fast as AWS where there’s much more opportunity in front of it than behind it. ” he says.. He contrasts it with the usual pattern: “A lot of the time. by the time you get to something this big. you’re eking out single-digit percent growth as you try to optimize around the edges.”

For all the unpredictability of AI’s future. it’s hard to see how AWS wouldn’t define the next two decades.. After 20 years, the question no longer seems to be whether cloud will matter.. The question now is which business systems can learn to live inside the speed of AI—and how quickly AWS can keep earning the right to be the layer beneath it.

AWS Amazon Web Services AI generative AI Bedrock SageMaker agentic AI S3 cloud computing Microsoft Azure Google Cloud Trainium3 re:Invent Matt Garman Colleen Aubrey Swami Sivasubramanian