In 2006, Amazon Internet Companies was a fledgling—and a little bit of an oddity. Amazon had taken the cloud-computing applied sciences it had created for its personal operations and turned them right into a enterprise. Any group might use them to construct out a web-based presence with out managing any infrastructure. Amazon watchers struggled to suss out what the e-tailer was as much as: “I’ve but to see how these investments are producing any revenue,” carped one Wall Avenue analyst.
On the very begin—when it was nonetheless an enormous deal if AWS collected $100 in income in a single day—an AWS product supervisor named Matt Garman had lunch with a good friend who labored in one other a part of the corporate. “[The coworker] requested, ‘How is that AWS factor going? I heard about it, and it sounds fairly attention-grabbing,’” Garman recollects. “And I used to be, like, ‘I believe this might be a billion-dollar enterprise for Amazon.’” His lunch mate cautioned him in regards to the daunting ambition of that objective.
Because it turned out, AWS smashed via Garman’s $1 billion objective after which simply saved going, reaching $128.7 billion in income in 2025. Alongside the best way, it got here to ship nearly all of Amazon’s revenue, to the tune of $45.6 billion final yr. As for Garman, his early religion within the firm’s potential led to the final word payoff in June 2024, when he grew to become its CEO, succeeding Adam Selipsky.
Alongside the best way, nothing was assured. “After we began to get a little bit traction, there was this type of meme about how AWS would rapidly grow to be a commodity and every thing would type of normalize out,” he says. “And our crew has proven unbelievable invention to show that that’s not true.”
However AWS’s affect on Amazon, as spectacular because it’s been, fails to convey its affect on enterprise and the world typically. Offloading administration of the myriad applied sciences that energy an internet site to somebody who is aware of what they’re doing simply is sensible. Over time, organizations of all types purchased into that technique, together with huge firms that had been initially cautious of ceding management over such a crucial ingredient of their operations. Eyeing the chance, two different tech giants spun up their very own AWS-like items, Microsoft Azure and Google Cloud. Cloud computing grew to become one of many tech business’s fiercest aggressive battlegrounds.
On March 13, AWS formally marked its twentieth anniversary, which dates to the introduction of its Easy Storage Service, higher referred to as S3 and nonetheless one among its flagships. (It’s important to be a tech historical past obsessive to do not forget that an earlier model of the idea, initially referred to as Amazon.com Internet Companies, launched in 2002.) The corporate continues to be the dominant pressure within the class it created, however after years of pursuit, Microsoft and Google have narrowed its lead. Again within the first quarter of 2020, AWS held 32% of the market in contrast with Azure’s 18% and Google Cloud’s 8%, in keeping with Synergy Analysis Group. Within the first quarter of 2026, AWS’s share was 28%, Azure’s was 21%, and Google Cloud popped to 14%.
Because of artificial intelligence, the three cloud suppliers are hardly squabbling over their respective slices of a pie of mounted measurement. It’s a testomony to the revolution AWS spawned that there’s been no debate about whether or not most firms will get their AI as a cloud service. In fact they may. Given the overwhelming computational assets essential to make giant language fashions (LLMs) function at scale, it’s the one sensible solution to make the know-how pervasive.
AI, says Garman, “is an enormous know-how leap that modifications every thing about how know-how is consumed. It modifications every thing about how all of our clients are going to function their companies, how industries are going to work.” As a supplier of AI on demand, AWS is charged with driving that change. Nevertheless it’s additionally the largest change the corporate and its class have seen of their first 20 years—and an opportunity for its rivals to make additional inroads.
Garman calls AI “an unlimited tailwind to our enterprise already” however acknowledges that the problem of getting it proper is simply starting. “All know-how disruptions needs to be seen as each a risk and alternative,” he cautions. On a number of fronts, AWS is evolving to satisfy this second.
An ever-expanding toolkit
Like all of the tech giants at the moment jockeying to steer the current AI revolution, Amazon Internet Companies was quietly, persistently critical in regards to the know-how effectively earlier than it grew to become the business’s number-one obsession. “We clearly didn’t venture quite a lot of the generative AI explosion that’s occurred on the earth as we speak,” Garman says. “However we’ve lengthy identified that [AI] was going to be critically necessary.”
In 2017, Swami Sivasubramanian, who’d joined Amazon as a analysis intern a dozen years earlier, grew to become AWS’s VP of AI. Later that yr, at its mammoth annual re:Invent convention in Las Vegas, the corporate launched SageMaker, a platform for creating, coaching, and in any other case wrangling machine-learning fashions. Upgraded and expanded many occasions since, it stays one among AWS’s core AI choices.
On the time AWS was formulating its plans for SageMaker, Google’s TensorFlow software program library dominated AI improvement. However AWS believed that clients would come to prize alternative. “Even internally, after we constructed purposes, we seen you want a number of fashions even for a single software to make it occur,” Sivasubramanian explains. That realization knowledgeable 2023’s Bedrock, which lets clients use AWS to run dozens of AI fashions from main firms, together with Amazon itself, Anthropic, Nvidia, DeepSeek, Qwen, Mistral, TwelveLabs, and—through a new partnership—OpenAI.
Together with constructing out AI’s software program layer, AWS has spent years creating its personal customized AI processors, affording it extra management over its infrastructure than if it had been fully depending on Nvidia for computing muscle. Amazon’s 2015 acquisition of Israeli startup Annapurna Labs has led to a number of generations of chips for inference and coaching, most lately the Trainium3, introduced final December at re:Invent.
Lately, agentic AI—types of the know-how that may carry out complicated, typically time-consuming duties with some measure of autonomy—has come to dominate the dialog about the place AI goes. Reflecting on this improvement led Sivasubramanian to “a realization that AI brokers will essentially change how all of us work and dwell.”
Wanting to assist AWS seize this chance, he explains, “I spun myself out.” In March 2025, he gave up his outdated job as VP of AI to grow to be VP of AWS Agentic AI, overseeing a bunch targeted on creating merchandise which can be, in a technique or one other, agent-centric.
By July, this funding started to repay in new AWS companies. Kiro is a coding surroundings that lets software program engineers flip over a few of their heavy lifting to an LLM-powered agent. Bedrock AgentCore helps them construct brokers of their very own. DevOps Agent, introduced at December’s re:Invent 2025, screens different AWS companies to detect and resolve issues earlier than they require human intervention.
At AWS, like elsewhere, lots of agentic AI’s earliest massive wins are coming from its means to hurry software program improvement by writing code. Sivasubramanian factors to clients resembling Thomson Reuters, which used a AWS agentic AI service referred to as Remodel to assist modernize purposes constructed way back utilizing creaky applied sciences resembling Microsoft’s .NET. Work that will as soon as have consumed three to 4 years now takes six to 12 months, Sivasubramanian marvels.
The advantages are hardly restricted to massive firms slogging via mundane however necessary technical initiatives. “Even my 10-year-old daughter, who doesn’t absolutely know but find out how to construct in Python, was in a position to spin up and construct an internet site to handle calendars for the whole family,” he says. “And he or she constructed it on AWS.”
Past infrastructure
When Colleen Aubrey joined AWS as a senior VP in 2024, she was a brand new recruit—but in addition an outdated hand at Amazon, the place she’d labored for almost 20 years. Till then, most of her expertise was in its promoting arm. Her lengthy immersion within the firm’s distinctive tradition smoothed the transition from adverts to infrastructure, although the shift in jargon was a little bit of a shock: “The acronyms had been completely completely different,” she says.
Aubrey wasn’t at AWS to do infrastructure in its traditional kind. As a substitute, she was charged with spearheading its growth into an space the place it had far much less expertise: full-blown enterprise productivity purposes.
“At Amazon, we’ve constructed lots of our personal purposes, and we realized quite a bit from that,” she says. “And my speculation was that we might convey to life a few of that studying for AWS clients within the type of enterprise purposes. And that the time was an excellent time, as a result of we might concurrently take into consideration what we’d construct as we speak given the capabilities of AI and the place we would see that going.”
In April, at an occasion in San Francisco, the corporate launched a line of cloud-based, AI-powered merchandise for automating common business processes. Amazon Join Choices focuses on supply-chain administration. Amazon Join Expertise conducts job interviews. Amazon Join Well being helps docs’ workplaces with duties resembling scheduling and medical historical past overview. And Amazon Join Buyer is the newest model of a customer support contact middle platform that AWS initially launched in 2017.
Generative AI permits the Join merchandise to supply chat-like interfaces and voice enter. Based on Aubrey, the objective is to supply software program “that works in a approach the place, as an individual, a human within the enterprise, I don’t should learn to use a brand new software. I work together with it in methods which can be acquainted.”
AWS’s first 20 years didn’t essentially set it as much as create such experiences. The corporate has loads of experience at creating administrative dashboards that allow technologists configure, handle, and monitor its companies. However anybody who’s used them—or their counterparts at Azure and Google Cloud—is aware of they’re not precisely grasp lessons in polished, consumer-grade usability. To up its sport, the corporate employed Hector Ouilhet as AWS Options VP of Design in January 2025.
Ouilhet spent greater than 14 years at Google, the place he was one of many folks answerable for Material Design, the design language that gave the corporate its first cohesive set of instruments for creating interfaces that had been each intuitive and recognizably Google-y. He compares the problem at AWS to that one, with the added twist of AI each enabling and demanding new approaches to how folks work together with computer systems.
“We construct the entire factor ourselves in phrases of the expertise,” he says. “Not solely how it appears to be like, however the way it feels, the way it sounds, the way it behaves, the way it interrupts, the way it listens. So now, the observe of design is approach broader.” Ouilhet calls AWS’s method to AI agent interfaces “humorphism.” Its rules—resembling “Route work to whoever can do it greatest” and “Synthesize and tailor info for the second”—are detailed at a website he created; he says he’d be delighted if different firms adopted the lead.
Approachability additionally drove the newest updates to Amazon Quick, an AI assistant, launched final yr, that faucets into enterprise instruments resembling Google Workspace, Microsoft’s Groups and Outlook, and Slack for functions resembling analysis and activity automation. On the April occasion, AWS introduced new Fast apps for MacOS and Home windows that make it extra straight aggressive with the likes of Anthropic’s Claude Cowork. It additionally began letting customers join the freemium service with a standard-issue Amazon account, permitting them to stand up and operating in minutes with out confronting the doubtless intimidating full-on AWS expertise.
“The restrict at the second is about 300 workers,” says Jigar Thakkar, AWS’s VP of agentic AI for enterprise, a Microsoft veteran (and Groups co-creator) who joined AWS in January. “When you’re a lot bigger than that, you need to get the enterprise account, the place we do much more governance and safety.”
Secret components
New enterprise apps apart, AWS’s core enterprise stays offering dependable components for different technologists’ improvements. Its position is that of a silent accomplice, and solely the occasional outage reveals its involvement by making clear what number of websites rely on it.
When the corporate was younger, its clients tended to be smaller outfits that had been open to contemporary concepts and knew they wanted assist scaling. Each Garman and Sivasubramanian point out SmugMug, the photo-sharing service whose early embrace stays a totemic success story. SmugMug’s CEO, Don MacAskill, negotiated AWS’s preliminary asking value of 40¢ a gigabyte for cloud storage down to fifteen¢, then took the plunge. He couldn’t make sure that Amazon would keep dedicated to its new enterprise: “Lots of people instructed me I used to be loopy on the time, simply tons and tons and tons,” he instructed me in 2017.
Immediately, AWS has the boldness of among the world’s best-known firms, who name on it for components that go far past on-line storage. AI is barely accelerating their consumption of its companies.
At United Airways, AWS is “a part of every thing we do,” says its CIO Jason Birnbaum. The airline started working with the corporate in 2018, the identical yr it launched a customer-service program referred to as “Every Flight Has a Story.” Moderately than leaving vacationers questioning in regards to the concern that had triggered a takeoff delay, the initiative offered them with a proof of what had gone amiss—one handcrafted, at first, by a human “storyteller.”
That gesture, Birnbaum says, “was amazingly well-received—it simply was robust to scale. We use AI now to jot down greater than half of these messages, which has enabled us to cowl far more situations and be far more clear with our clients.” Passengers on greater than half one million flights have acquired messages generated by AWS AI. “It’s been a house run for us, and it’s been a house run for our clients,” he provides.
When Mondelēz Worldwide CTO Chris Hesse joined the snack-food behemoth in 2021, it wasn’t an AWS store. Now nearly all of its cloud runs on AWS companies. The maker of Oreos, Clif bars, and Cadbury eggs lately rolled out the Fast assistant to 50,000 workplace employees, a mass deployment that Hesse admits was on the early aspect, given Fast’s state when he determined to maneuver ahead. “I noticed issues that had been perhaps not as polished, and I used to be afraid folks would speak about that,” he says. “However as a substitute, everybody went, ‘Have a look at this factor that I constructed, take a look at this factor that it does. This helps me a lot.’ That type of factor.”
Capital One—whose senior VP of infrastructure, Will Meyer, says likes to think about itself as “a tech firm that has this wonderful danger administration functionality of a very savvy financial institution”—has been constructing on AWS for over a decade. Lately, a lot of that constructing has had an AI angle. Its initiatives have included an agentic car-buying expertise for its auto mortgage enterprise, AI help for 20,000 (human) customer support brokers, and AI-enhanced fraud case decision.
Even a financial institution that tries to assume like a tech firm wouldn’t have been in a position to ramp up all these AI-infused merchandise in parallel with out assist. “There’s this entire class of stuff that AWS calls ‘undifferentiated heavy lifting’ that we needed to get our groups out of,” Meyer says. “However for us, it’s additionally at all times been about tapping into the innovation that the cloud can ship. It’s not simply renting exhausting disks and CPUs from somebody.”
AWS, he provides, has “been actually good at simply serving to actual clients clear up actual issues. And that’s a technique I believe is getting older fairly effectively.”
These sorts of main clients’ worth to AWS transcend the checks they write. “A number of the best info that we get on what to construct subsequent comes from actually leaning into of us like Capital One and saying, ‘What are the [blockers] that will forestall you from placing every thing on high of AWS?’” Garman says. “‘How will we assist you’ve got higher safety? How will we assist your improvement groups innovate sooner?’”
That listening is crucial: By definition, AWS’s clients’ technological priorities grow to be its personal. Sivasubramanian notes, nevertheless, that it’s not nearly giving folks what they ask for. “9 out of 10 occasions, we do precisely what clients need,” he says. “And one out of 10 occasions, we learn between the strains and [conclude] they’re asking for a sooner horse as a substitute of a automobile. Then we construct a automobile.”
In each varieties, maintaining with clients’ ever-expanding wants appears to be paying off for AWS, whilst Microsoft and Google present extra strong competitors. In the first quarter of 2026, AWS’s $37.6 billion income represented progress of 28%. Its working revenue, $14.2 billion, was up 23%. Stats present AI making an outsized contribution: The Bedrock mannequin platform, for example, now has 125,000 clients, together with 80% of the Fortune 500. Throughout the quarter, Bedrock processed extra tokens than in its whole prior historical past, leading to 170% quarter-over-quarter income progress.
“You don’t typically discover a enterprise alternative that’s grown as quick as AWS the place there’s rather more alternative in entrance of it than behind it,” Garman says. “Numerous the time, by the point you get to one thing this massive, you’re eking out single-digit p.c progress as you attempt to optimize across the edges.”
Unpredictable although AI’s future is, it’s robust to check it dropping momentum anytime quickly—or failing to outline the following 20 years for AWS.
