by Brad Stone
Amazon’s original cloud products were conceived by Jeff Bezos in concert with other technical leaders between the years 2004 and 2006. The Simple Storage Service, or S3, and the Elastic Compute Cloud, or EC2, provided most of the functionality of a back-office computer room—but one that could be accessed remotely, and existed inside the massive, air-conditioned data centers that Amazon would construct elsewhere in the country. These would be the dial tones of the twenty-first-century internet explosion. In 2007, Amazon also introduced a primitive database called SimpleDB, to allow customers to store and retrieve organized or “structured” sets of their data.
Entering the business of databases, a seemingly boring bit of commerce that is actually a thriving and competitive $46-billion-a-year industry, would be one of AWS’s most important pathways to success. Amazon itself used Oracle’s relational database to manage Amazon.com, and the company’s ever-growing traffic strained the software and periodically threatened the stability of the site, frustrating Bezos. Across its fulfillment centers and in its online store, Bezos always wanted to minimize Amazon’s dependencies on other companies, because their own primitive database capabilities weren’t up to the task. When SimpleDB also proved to be too clunky and complex to use, AWS engineers started working on a more fast and flexible version, called DynamoDB, to handle the massive volumes of traffic that were endemic to the internet.
SimpleDB was also used avidly by another early AWS customer to store the titles and thumbnail images of its entertainment catalog: Netflix. Reed Hastings’s DVD-by-mail startup wanted to move other parts of its technology operation to the cloud as it transformed itself into a streaming company. To accommodate that, Amazon would need to build the cloud versions of relational databases and a tool called a data warehouse. In 2010, Andy Jassy, the head of the AWS unit, and a vice president named Raju Gulabani started working on the project and then updated the S-team on their progress.
In the meeting, according to a participant, Gulabani projected it would take a decade for Amazon to succeed in relational databases. “I will bet you it will take more than ten years to get this done,” Bezos said, causing momentary consternation among the assembled AWS crew. “So, you better get started now.” Understanding that robust databases would be one of the biggest opportunities in cloud computing, Bezos significantly increased Jassy’s budget request.
Gulabani poached another Indian-born executive, Anurag Gupta, from Oracle, and they opened an office in Silicon Valley. Over the next few years, Gupta assembled a team that would build several AWS databases around free and increasingly popular open-source software tools like MySQL and Postgres. In 2012, AWS introduced Redshift, a so-called data warehouse that allowed companies to analyze the data they stored in the cloud; in 2015, it rolled out Aurora, a relational database. These were typically Amazonian names: geeky, obscure, and endlessly debated inside AWS, since according to an early AWS exec, Bezos had once mused, “You know, the name is about 3 percent of what matters. But sometimes, 3 percent is the difference between winning and losing.”
The name “Redshift” was suggested by Charlie Bell, a former engineer for Boeing on NASA’s space shuttle and the senior vice president in charge of AWS’s operations; it’s the term for the change astronomers see in light, the fastest thing in existence, that’s emitted from a celestial object like a star as it moves away from the observer. Nonetheless, Larry Ellison, then the CEO of Oracle—whose logo happens to be red—saw it as corporate trash talk and began to see red himself. That “never even occurred to us,” Jassy said. “When we were told later that Oracle believed that, we thought it was kind of funny.” A bitter rivalry between Oracle and Amazon, already simmering with Amazon’s entry into databases, intensified.
AWS’s portfolio of cloud-based databases, on top of the classics like S3 and EC2, drew companies big and small toward cloud computing and further into Amazon’s embrace. Once they moved their data onto Amazon’s servers, companies had little reason to endure the inconvenience of transferring it back out. They were also more likely to be attracted to the other profitable applications that AWS introduced. Over the next few years AWS sales and operating margins started to shoot upward. “Of all the services we added, it was the database portfolio that broadened AWS’s appeal,” said Taimur Rashid, a former AWS manager.
Nearly as remarkable as AWS’s evolution into a profitable business over the first half of the 2010s was its emergence as its own distinctive organization, an ablation calved from the glacier of Amazon itself. In 2011 the division broke off from the company’s main campus in South Lake Union and moved a half mile away to 1918 Eighth Avenue, a five-hundred-foot-tall glass skyscraper that Amazon dubbed Blackfoot. Ever the Bezos disciple, Jassy hung on the walls not the admiring articles but the critical ones, including a 2006 Businessweek story whose sub-headline read: “Amazon’s CEO wants to run your business with the technology behind his Web site. But Wall Street wants him to mind the store.”
AWS’s culture was a microcosm of Amazon’s: tough, unrelenting, and focused on meeting impossibly high standards. Jassy and his fellow managers asked searing questions of their underlings and hammered anyone without suitable answers or who didn’t embrace accountability for a problem within their purview. Daily operations were driven by data-filled six-page narratives and the obsessive contemplation of the needs of customers. When employees returned strong results, attention always turned to the ways in which they could have done better. One former executive described the mentality this way: “We were really good at going up to the gold medal podium and complaining that our medals weren’t shiny enough.”
Engineers were given pagers and were assigned to on-call rotations, when they were expected to be available at all hours to address system outages. If a serious technical problem erupted while the pagers were muted during meetings at AWS, Amazon’s pager program would automatically circumvent “silent mode” and the meeting would erupt in a rolling orchestra of electronic pings.
In many respects, Jassy’s business philosophies were a distillation of Bezos’s. A few years after he joined Amazon from Harvard Business School in 1997, Jassy had narrowly avoided getting fired in an early purge of Amazon’s marketing department. Bezos saved him, dubbing him as “one of our most high potential people,” according to former S-team member Diego Piacentini. For eighteen months, he was Bezos’s first full-time shadow, or technical advisor. This brand-new role entailed the almost slavish following of the CEO, and colleagues gently teased Jassy for it.
Jassy totally embodied Amazon values like frugality and humility. He usually wore inexpensive sport coats and loudly trumpeted his enthusiasms for diversions like New York sports teams, buffalo wings, and the Dave Matthews Band. Even as his net worth skyrocketed along with AWS’s value—he received a $35 million stock grant in 2016 alone—he shunned the ostentatious trappings of success, like traveling via private aircraft. He held an annual Super Bowl party in the replica sports bar that he’d fashioned inside the basement of his own Seattle home. Bezos attended every year until 2019, when in another glimpse of dramatic changes ahead, he showed up at the actual game, sitting in the commissioner’s box.
Bezos liked to say that “good intentions don’t work, but mechanisms do.” Inside AWS, Jassy applied that adage ferociously. The rhythms of a week at AWS revolved around several formal “mechanisms” or well-honed processes or rituals. Ideas for new services, their names, pricing changes, and marketing plans were meticulously written as six-page documents and presented to Jassy in his twentieth-floor meeting room, dubbed “the Chop” (the name Jassy and his roommate had given their Harvard dorm room, from a novel they were assigned in European literature, Stendhal’s The Charterhouse of Parma). Executives asked hard technical questions and Jassy usually spoke last. Colleagues said he exhibited almost inhuman levels of discipline, sitting in meetings for ten hours a day and digesting dense and complex documents without flagging.
The highlights of the AWS week were two Wednesday mo
rning meetings. Jassy ran the ninety-minute midday business review, where the top two hundred managers discussed the minute details of customers, competitive developments, and the financial health of each product unit. But the real centerpiece of the week was the forum that preceded that meeting: the two-hour operations review to assess the technical performance of each web service. Held in the large conference room on the third floor, it was run by the intimidating and direct Charlie Bell, the former space shuttle engineer.
AWS execs and engineers typically describe this remarkable session with a combination of awe and post-traumatic stress disorder. Around the big table at the center sat more than forty vice presidents and directors, while hundreds of others (almost all of whom are men) stood in the wings or listened over the phone from around the world. On one side of the room was a multicolored roulette wheel, with different web services, like EC2, Redshift, and Aurora, listed around the perimeter. Each week the wheel was spun (until 2014 when there were too many services and software mimicked the function of the wheel), with the goal being, Jassy said, to make sure managers were “on top of the key metrics of their service all week long, because they know there’s a chance they may have to speak to it in detail.”
Getting selected could be a career-defining moment at AWS. Managers could boost their prospects with a comprehensive and confident presentation. But if they employed ambiguous language, erred with their data, or conveyed even the whiff of bullshit, then Charlie Bell swooped in, sometimes with awesomely patronizing flair. For managers, a failure to deeply understand and communicate the operational posture of their service could amount to career death.
Nevertheless, as AWS approached its ten-year birthday, and as its revenues increased and profits mounted, it became the most desirable division for Amazon’s tech elite, a kind of Ivy League among all of its business units. Staying and thriving amid the geniuses and their diabolical rituals amounted to earning the medal of honor.
* * *
In the early years, Bezos waded into the details of AWS himself, editing web pages for the first products and reviewing revenue reports from EC2 and occasionally replying with smiley faces. Over time, as he fixated on newer things, like Alexa and the Amazon Go stores, he allowed Jassy to run AWS autonomously. He receded from regular view, save for reviewing significant investment decisions and overseeing the annual OP1 and OP2 sessions, where he usually pressed for ways to connect AWS with other parts of Amazon’s business. “Jeff was very involved almost as an investor in AWS,” said Joe DePalo, a former AWS exec. “He would ask questions and poke and review. But day to day, Andy operated it independently.”
Bezos also served as a kind of strategic guru for Jassy and his leadership team. As Google and Microsoft awoke to the potential of cloud computing and began investing heavily in their competing initiatives, he urged Jassy to think about ways to protect Amazon’s advantages. “You’ve built this lovely castle, and now all the barbarians are going to come riding on horses to attack the castle,” Bezos said, according to a former AWS exec who reports hearing the comment. “You need a moat; what is the moat around the castle?” (Amazon denied that Bezos said this.)
In January 2015, one of Jassy’s answers was Amazon’s $400 million acquisition of an Israel-based chipmaker, Annapurna Labs, to build low-cost, high-performance microprocessors for Amazon servers, and seek a cost advantage in Amazon’s data centers that competitors couldn’t match.
Bezos had one other impact at AWS: both he and Jassy lobbied to conceal the division’s financial details from public view, even amid the widespread skepticism that throttled the company and its stock price in 2014. But in 2015, Amazon’s finance department argued that the division’s revenue was approaching 10 percent of Amazon’s overall sales and would eventually trigger reporting requirements under federal law. “I was not excited about breaking our financials out because they contained useful competitive information,” Jassy admitted.
Nevertheless, that January, Amazon signaled that it would report AWS’s financial results in its quarterly report for the first time, and investors girded in anticipation. Many analysts predicted that AWS would be revealed as just another Amazon “science project”—a lousy, low-margin business that was sapping energy from the company’s more advanced efforts in retail.
In reality, the opposite was true. That year, AWS had a 70 percent growth rate and 19.2 percent operating margin, compared to the North American retail group’s 25 percent growth rate and 2.2 percent operating margin. AWS was gushing cash, even as it rapidly consumed most of it to build even more computing capacity and keep up with the fast-growing internet companies like Snapchat that were piling onto its servers.
This reporting was a huge surprise for the analysts and investors who monitored and scrutinized Amazon, and likely even a bigger one for Microsoft, Google, and the rest of the enterprise computing world. Analyst Ben Thompson facetiously called that April 2015 earnings report “one of the technology industry’s biggest and most important IPOs.” After the reveal, Amazon’s valuation jumped almost 15 percent in a day, crossing the $200 billion mark for the first time and laying to rest the myth of Amazon as a perpetually money-losing machine.
* * *
A few months before that earnings report, the S-team had been analyzing their deteriorating competitive position in China and the success of Alibaba’s annual holiday shopping extravaganza, Singles Day. For the last five years, Jack Ma’s e-commerce juggernaut had turned the date of 11/11 into a hybrid of Black Friday and Valentine’s Day, offering a frenzy of deals which in 2014 generated more than $9 billion sales and a dependable tsunami of free press.
In his presentation on China, international boss Diego Piacentini proposed that Amazon might fashion its own such shopping holiday. Jeff Bezos thought it was a good idea, but at the time he was consumed with linking everything to Amazon’s seductive Prime service. He suggested that the company roll out the holiday globally and use it to try to add new members to Prime.
The assignment was passed to Greg Greeley, a vice president in charge of Prime, who in turn handed it to one of his deputies, longtime Amazon executive Chris Rupp. Rupp knew that Amazon customers were accustomed to spending freely on Black Friday and Cyber Monday and that another Christmas shopping event would simply shift their discretionary spending by a few weeks. She also knew that Amazon was not particularly adept at tapping the summer shopping season known as “back to school,” a period marked by such retail rituals as Nordstrom’s Anniversary Sale.
Rupp’s subsequent proposal to hold the event midsummer sparked contentious debate inside Amazon. She argued that customers had money to spend over the summer, while Amazon could exploit the excess warehouse space built for the peak season. Amazon’s supply chain executives, who spent their comparatively tranquil summers preparing for the holidays, had little interest in handling a midyear surge. “I got every kind of pushback on that, but I had darn good reasons for doing it,” Rupp said.
Greeley and Rupp presented the paper to the S-team in January 2015 and got a sign-off from Bezos. “Don’t make this convoluted. Prime Day needs to mean one thing, and we have to do it really well,” he told them. In the appendix, he highlighted a section that targeted ten thousand deals for Prime Day, a larger selection than Black Friday. To accomplish that, they would have to persuade Amazon’s merchandising teams to coalesce behind the goal and harangue their vendors to supply the discounts.
In early March, the event was assigned to a member of Rupp’s team, a thirty-year-old product manager named Meghan Wulff. Wulff would be Prime Day’s “single-threaded leader,” with her sole focus on the event (as well as trying to blot out the characteristically Amazonian paranoia that, at any moment, she might screw things up and get fired). Because it was a global event and meant to be chock-full of surprises, Wulff and a colleague code-named it “Project Piñata,” requiring an awkward keyboard maneuver every time she wrote the name in a document or email. “I will never put an enye in a project name ev
er again,” Wulff joked.
Wulff now had to mint an entirely new shopping holiday on an impossible deadline. Amazon wanted to hold it on July 15, to mark the twentieth anniversary of the first sale on Amazon.com. In May, she embarked on a whirlwind trip, traveling to Tokyo, London, Paris, and Munich, to try to yoke together a confederation of Amazon’s merchandisers, marketers, and supply chain executives to get behind an initiative that almost everyone was dubious about. Prime Day did not exist yet, so there was little reason for Amazon’s retail and advertising teams to drop everything and convince suppliers to support it. “I felt like I was running a Ponzi scheme,” Wulff said. Only the imprimatur of Bezos motivated other Amazon execs to abandon their apathy and fall into line.
As the date approached, Wulff and Rupp started to realize there might be more riding on Prime Day than they’d suspected. Bezos wanted to wade into the details and review the promotional materials. Good Morning America was interested in previewing the event. “At first we thought, ‘this is fantastic,’ ” Rupp said, “and then ‘uh-oh, wait a minute, this might be bigger than we thought.’ ”
Amazon in 2015 had moved into the zeitgeist in a way that perhaps even its own executives hadn’t yet recognized. The event kicked off in Japan, where the local website promptly crashed due to overwhelming interest, then cascaded into Europe, and finally into the U.S., where social media reaction was swift—and brutally negative. Armchair shoppers ignored Amazon’s proposed hashtag, #HappyPrimeDay and took to Twitter to criticize the company for sold-out items, discounts on trivial products like dishwashing detergent, and an abundance of underwhelming deals. “I keep going back to the Amazon #PrimeDay sale like a girlfriend who is convinced it’s going to get better,” said one typical Twitter post. “The best deal I’ve seen so far is 15% off a box of pop tarts,” read another.
