یک شرکت تکنولوژی نه یه شرکت خرده فروشی

CHAPTER 7

A Technology Company, Not a Retailer

On July 30, 2005, Amazon celebrated its tenth anniversary at a gala at Seattle’s Benaroya Hall. Authors James Patterson and Jim Collins and screenwriter Lawrence Kasdan spoke to employees and their guests, and Bob Dylan and Norah Jones performed and sang a rare duet, Dylan’s “I Shall Be Released.” The comedian Bill Maher acted as master of ceremonies. Marketing vice president Kathy Savitt had persuaded Bezos to splurge on the historic moment, and, characteristically, they organized everything in such a way that it had a benefit for customers: the concert was streamed live on Amazon.com and watched by a million people.

Despite how far Amazon.com had come, it was still often a media afterthought. It was now officially the age of Google, the search-engine star from Silicon Valley. Google cofounders Larry Page and Sergey Brin were rewriting the story of the Internet. Their high-profile ascent, which included an IPO in 2004, was universally watched. Suddenly, clever online business models and experienced CEOs from traditional companies were passé in Silicon Valley, replaced by executives with deep technical competence.

This, it seemed, was to be the era of Stanford computer science PhDs, not Harvard MBAs or hedge-fund whiz kids from Wall Street, and the outside world did not believe Amazon would fare well in this profound shift. In the year leading up to its birthday celebration, Amazon’s stock fell 12 percent as Wall Street focused on its slender margins and the superior business models of other Internet companies. Eighteen of the twenty-three financial analysts who covered the company at the time of the anniversary event expressed their skepticism by putting either a hold or a sell rating on Amazon’s stock. The market capitalization of eBay, still viewed as a perfect venue for commerce, was three times larger than Amazon’s. Google’s valuation was more than four times Amazon’s, and it had been public for less than a year. Fixed-price online retail was simply out of vogue.

Ever since the late 1990s, Bezos had been claiming that Amazon was a technology company pioneering e-commerce, not a retailer. But that sounded like wishful thinking. Amazon still collected a vast majority of its revenues by selling stuff to customers. Despite Bezos’s protestations, Amazon looked, smelled, walked, and quacked like a retailer—and not a very profitable one at that.

A week after the tenth-anniversary show, the New York Times published a lengthy article on the front page of its Sunday business section that suggested Bezos was no longer the right man for the job.1 “It’s time for Mr. Bezos to do as the founders of so many other technology companies have done before him: find a professionally trained chief executive with a deep background in operations to take the reins,” said an analyst quoted prominently in the piece.

The rise of Google did more than shift the mind-set of Wall Street and the media. It posed a new set of challenges to Amazon. Rather than just hopping on Amazon.com and looking for products, Internet users were starting their shopping trips on Google, putting an unwelcome intermediary between Jeff Bezos and his customers. Google had its own e-commerce ambitions and early on opened a comparative shopping engine, dubbed Froogle. Even worse, both Amazon and eBay had to compete with each other to advertise alongside Google results for popular keywords like flat-screen TV and Apple iPod. They were essentially paying a tax to Google on sales that began with a search. To make this new kind of advertising more efficient, Amazon devised one of the Web’s first automated search-ad-buying systems, naming it Urubamba, after a river in Peru, a tributary of the Amazon. But Bezos was wary of helping Google develop tools that it might then extend to Amazon’s rivals. “Treat Google like a mountain. You can climb the mountain, but you can’t move it,” he told Blake Scholl, the young developer in charge of Urubamba. “Use them, but don’t make them smarter.”

Google competed with Amazon for both customers and talented engineers. After its IPO, the search giant opened an office in Kirkland, a twenty-minute drive from downtown Seattle. Google offered its employees lavish perks, like free food, office gyms, and day care for their children, not to mention valuable stock options. For its part, Amazon offered a sickly stock price and a combative internal culture, and employees still had to pay for their own parking and meals. Not surprisingly, Google began to suck engineers out of Amazon en masse.

During this time, Bezos relentlessly advocated for taking risks outside of Amazon’s core business. Between 2003 and 2005, Amazon started its own search engine and devised a way to allow customers to search for phrases inside books on the site. Bezos also helped to pioneer the modern crowd-sourcing movement with a service called Mechanical Turk and laid the groundwork for Amazon Web Services—a seminal initiative that ushered in the age of cloud computing.

Bezos battled a reaction that he dubbed the institutional no, by which he meant any and all signs of internal resistance to these unorthodox moves. Even strong companies, he said, tended to reflexively push back against moves in unusual directions. At quarterly board meetings, he asked each director to share an example of the institutional no from his or her own past. Bezos was preparing his overseers to approve what would be a series of improbable, expensive, and risky bets. He simply refused to accept Amazon’s fate as an unexciting and marginally profitable online retailer. “There’s only one way out of this predicament,” he said repeatedly to employees during this time, “and that is to invent our way out.”

Bezos was certain that Amazon needed to define itself as a technology company instead of a retailer, so he started hiring technologists and giving them obscure job titles. In 2001, he lured Apple veteran and renowned user-interface expert Larry Tesler to Amazon and called him vice president of shopping experience. The next year, he hired a Stanford-educated machine-learning professor named Andreas Weigend and dubbed him chief scientist. Neither did particularly well under Bezos’s demanding tutelage and both quickly grew tired of Seattle. Weigend lasted only sixteen months at Amazon, Tesler a little over three years. Then Bezos found a technologist who thought just as grandly as he did about ways Amazon could branch out in new directions.

Udi Manber was born in Kiryat Haim, a small town in northern Israel, and he earned a PhD in computer science at the University of Washington. In 1989, as a computer science professor at the University of Arizona, he wrote an authoritative book about the problem-solving wonders of complex mathematical formulas called Introduction to Algorithms: A Creative Approach that captured the attention of the Silicon Valley cognoscenti. Manber worked at Yahoo during its glory years but quit in disappointment in 2002 after former Warner Brothers CEO Terry Semel took over as CEO and reoriented the Web portal toward becoming a media company.

Rick Dalzell had heard of Manber’s book and started courting him while Manber was preparing to leave Yahoo. Dalzell introduced Manber to Bezos, and by all accounts, an intoxicating geek bromance was born. One of the first questions Bezos asked Manber was “Why don’t you describe a new algorithm that you invented?” Manber did and then marveled at Bezos’s comprehension. “He not only fully understood it, but did it faster than most people. I did not expect that from a CEO. It would have taken me a month to explain it to most senior Yahoo people,” he says.

Manber had serious reservations about moving to Seattle. His wife was a professor at Stanford and they had two young daughters in school. But Bezos agreed to let him split his time between Seattle and Silicon Valley. Manber joined Amazon that fall, and Bezos gave him a typically obscure job title: chief algorithms officer. A few months later, he joined the S Team. “Udi and Jeff had instant chemistry,” says Dalzell.

Manber’s mission was a broad one: use technology to improve Amazon’s operations and invent new features. He would see Bezos once a week—an exception to the CEO’s aversion to one-on-one meetings—to review ongoing projects and brainstorm new ideas. Manber always had Bezos’s full attention, even on a day when they met just a few hours before Amazon’s quarterly earnings announcement.

One of Manber’s first projects at Amazon captured the interest of both the media and the New York publishing establishment for the sheer scope of its ambition. Before Manber joined the company, Amazon had introduced a tool called Look Inside the Book, an effort to match the experience of a physical bookstore by allowing customers to browse through the first few pages of any title. Manber took that idea much further. He proposed a service called Search Inside the Book that would let customers look for specific words or phrases from any book they had purchased. Bezos loved the idea and raised the stakes: he wanted customers to be able to search any book on the site, and he gave Manber a goal of getting one hundred thousand books into the new digital catalog.2

“We had a very simple argument” for book publishers, Manber says. “Think of two bookstores, one where all the books are shrink-wrapped and one where you can sit as long as you want and read any book you want. Which one do you think will sell more books?”

Publishers were concerned that Search Inside the Book might open up the floodgates of online piracy. Most, however, agreed to try it out and gave Amazon physical copies of their titles, which were shipped to a contractor in the Philippines to be scanned. Then Manber’s team ran optical character-recognition software over the book files to convert the scanned images into text that Amazon’s search algorithms could navigate and index. To reduce the chance that customers would read the books for free, Amazon served up only snippets of content—one or two pages before and after the search term, for example, and only to customers who had credit cards on file. It also dropped a small piece of code, called a cookie, in each customer’s computer to ensure he didn’t keep coming back to read additional pages without paying.

It was a computationally intensive process, and Amazon did not provide Manber and his team with much in the way of computer resources. Manber almost had to resort to running his software on employee computers at night and on weekends, but one of his employees found a batch of idle PCs that had been set aside for emergencies. He was allowed to commandeer those machines, although with the understanding that they could be taken back at any time.

Amazon introduced Search Inside the Book on October 2003—and for the first time in three and a half years, there was a feature story on the company in Wired magazine, celebrating its significant innovation. The article revived Bezos’s vision of the Alexandria Project, that 1990s-era fever dream of a bookstore that stocked every book ever written. Perhaps such a universal library could be digital and thus infinitely more practical? Bezos cautiously told Wired that Search Inside the Book could indeed be such a beginning. “You have to start somewhere,” he said. “You climb the top of the first tiny hill and from there you see the next hill.”3

As Amazon was adding product categories throughout the 1990s, its executives came to an inevitable conclusion: the company had to become good at product search. Early in its history, Amazon had licensed a now-defunct search engine called Alta Vista, a spinoff of computer maker Digital Equipment Corp., but it had quickly proved insufficient. In the late 1990s, Amazon engineers Dwayne Bowman and Ruben Ortega led the development of an internal product-search tool called Botega (a mash-up of their surnames) that capitalized on Amazon’s vast trove of customer data, information the website had been collecting from the moment it officially opened for business. The system identified the top products customers clicked on for a given search term and then positioned those products higher in ensuing searches. That worked, for a time. But as Amazon’s catalog grew ever more complicated and Google got exceedingly good at indexing and organizing the Web, Amazon had to confront the awkward truth that one of its chief rivals could search Amazon’s site better than its own search engine could.

At that point, several factors led Amazon directly into the broader Web search arena—and into its first head-to-head confrontation with Google. Amazon was having a difficult time luring technical talent to Seattle, and its divisions often found themselves competing for the same engineers. So in late 2003, Jeff Holden, Udi Manber, and several colleagues travelled to Palo Alto to interview potential hires. The trip was so fruitful and the Seattle labor market had grown so challenging that the company decided to open its first North American office outside Seattle.

Bezos and Dalzell came to call these satellite locations remote development centers. The idea was to place the offices in regions with rich pools of technical talent and set teams to work on specific, isolated projects, harnessing the energy and agility of a startup while minimizing the need for communication with the mother ship in Seattle. Amazon’s lawyers, concerned that this might require the company to collect state sales tax, signed off on the strategy but only if the offices were set up as independent subsidiaries and stayed away from transactions with customers.

After one year in Seattle, Manber was already tired of his commute, and he was asked to head up the new Palo Alto office. In October 2003, Amazon’s first development center was opened on Waverly and Hamilton Streets in downtown Palo Alto. Staying true to his affinity for mathematical abbreviations, Bezos called it A9—shorthand for algorithms. Despite his move, Manber kept up his weekly meetings with Bezos via conference calls and regular trips to headquarters.

They were still thinking big. A9 not only worked on revamping product search on Amazon.com but also, in a direct attack on Google’s turf, developed a general Web search engine. The company licensed the Google search index but built on top of it—simultaneously partnering with and challenging Google. “Search is not a solved problem,” Manber said in April of 2004 when Amazon unveiled a Web search engine at A9.com. “There are lots more things that can be done. This is just the beginning.”

A9 would give Bezos and Manber a forum to try out some of their more ambitious ideas, most of which had nothing to do with Amazon’s core business. In one brainstorming session, they decided the Web presented a natural opportunity to reinvent the Yellow Pages and ginned up a project called Block View that matched street-level photographs of stores and restaurants with their listings in A9’s search results. This was two years before Google announced a similar (more successful and ultimately controversial) initiative called Street View.

Google would blanket the country with a fleet of company-owned trucks outfitted with expensive, specialized cameras to get its street views, but Amazon approached the problem with its usual emphasis on frugality. Manber’s budget for the project was less than a hundred thousand dollars. A9 flew photographers and portable equipment to twenty major cities and rented vehicles.

By late 2005, with Google gaining in both popularity and market capitalization, the general Web search at A9.com started to look like a noble but failed experiment. Web search, it became apparent, was not something that could be done cheaply or by piggybacking on a rival’s search index. Manber had a dozen engineers working on Web search, while Google had several hundred. Still, the A9 development center was showing promise. It made modest improvements to product search on Amazon.com and started work on an advertising service called Clickriver, which would allow advertisers (a television installer, for example) to purchase links within search results on Amazon.com (a search for HDTVs, for instance). Clickriver contained the seeds of a new advertising business, seeds that would later sprout into a healthy source of revenue for the company. Manber’s time at Amazon was productive in other ways too: after three years, he had more than twenty patent applications, several of which carried Bezos’s name too.

But then a series of conflicts erupted that rocked the S Team, broke up the Bezos-Manber partnership, and sent Bezos all the way back to the drawing board in his ongoing attempts to prove to the world that Amazon was something more than just a boring retailer, or a technology company that had chosen the least inspired business model of a new age.

At ten years old, Amazon could be a deeply unhappy place to work. The stock price was flat, there were strict limits on annual raises, and the pace was unrelenting. Employees felt underpaid and overworked. When the new development centers opened in Palo Alto and elsewhere, the joke inside Amazon was that it was a necessary move because everyone in Seattle was aware of how abjectly miserable employees at the company were.

In the engineering department, employees were constantly trying to fix a technical infrastructure that was now an aging, sprawling mess. The company had outgrown the original framework devised by Shel Kaphan in the 1990s, the monolithic code base dubbed Obidos that for years was held together by what Amazon executive Werner Vogels later called “duct tape and WD40 engineering.”4 And when Amazon cloned its clunky code base to run the websites of Target and Borders, those deals were lucrative but they magnified the company’s infrastructure problems. Instead of fighting flames emanating from a single building, engineers often had to deal with a neighborhoodwide inferno.

Like a lot of other technology companies at the time, Amazon got an education in the wisdom of moving to a simpler and more flexible technology infrastructure, called service-oriented architecture. In this kind of framework, every feature and service is treated as an independent piece and each can easily be updated or replaced without breaking the whole.

Led by Amazon’s chief technology officer at the time, an avid pilot named Al Vermeulen, whom colleagues fondly called Al V., the company rebuilt its technology infrastructure as a series of these independent but interconnected parts. The awkward and extended transition to this new code base, one element of which Amazon called Gurupa (after a section of the Amazon river where the tributaries diverged), took over three years and caused all kinds of excruciating pain among its network engineers, who were forced to carry pagers so they could respond promptly to the numerous problems.

As a result, dozens of these talented technicians left, many of them defecting to Google. Steve Yegge was one such engineer who made the move around this time. He would publish his opinion of his former employer years later by writing a screed on the Google+ social network and accidentally making it public for the entire Internet to read. “My challenge with Amazon is finding a way to describe it without making me puke,” Yegge wrote. “But I’ll figure something out, eventually. In many ways they’re a world-class operation—primarily in ways that matter to their customers; employees, not so much. But I guess in the end it’s the customers that matter.”

In late 2004, another window opened on the mood and inner workings of Amazon. Toys “R” Us sued Amazon in federal court, contending that Amazon had violated the agreement to allow the chain store to be the exclusive seller of the most popular toys on the Amazon website. The issues in the case were numerous and complex and hinged on some of the arcane legal language in the original contract. But they boiled down to a clash of goals and worldviews. Toys “R” Us thought it was paying Amazon hefty annual fees and a percentage of sales for exclusivity as the seller of the most popular toys on Amazon. But Amazon and its CEO could not abide anything that impeded their drive to give customers the ultimate selection, and Amazon constantly angered its partner by conceiving of new ways to allow other sellers to list competing toys on the site.

The trial was held in September of 2005 in a stuffy courtroom in Paterson, New Jersey. Bezos testified over the course of two days, and from court records, it does not appear that he enjoyed himself. Judge Margaret Mary McVeigh questioned Bezos’s inability to recall key decisions and ultimately ruled in favor of Toys “R” Us, allowing the toy seller to break its contract with Amazon and revive its own website. In her ruling, the judge described Amazon employees as contemptuous toward their offline counterpart and worshipful and apprehensive of their own CEO and his demands. “It was certainly my perception that nothing major happened at Amazon without Jeff Bezos’s approval,” she wrote in her judgment, quoting the testimony of a Toys “R” Us executive.

Amazon appealed the settlement but lost and was required to pay $51 million to its former partner. The dispute with Toys “R” Us would become exhibit A in the argument that Amazon was so fixated on catering to its customers and on the mechanics of its own business that the corporation was often hostile to the large companies it partnered with. (At the same time as the Toys “R” Us suit, another partnership, with travel site Expedia, also dissolved in litigation. That matter was settled out of court.)

With the toy business now in transition after the dissolution of the agreement with Toys “R” Us, the hard-lines retail division was cast into further disarray. Part of the problem was that categories like electronics and jewelry were not yet profitable but were growing faster than the older media businesses, which dragged down the company’s finances. Bezos felt he needed to give the issue specific attention, and so in late 2004, he hired Kal Raman, the former Drugstore.com executive who had played a supporting role in the employee poaching that had led to the Walmart lawsuit in 1998. Overnight, Bezos cleaved in two the domain of Diego Piacentini, then the senior vice president for worldwide retail, and he handed hard lines over to Raman. Bezos announced the move on a Tuesday in an internal e-mail to the company. Almost everyone from that time says the message was a shock not only to them but also to Piacentini (though Piacentini insists that he knew about the change before the e-mail went out).

Raman was a native of a small village in southern India. His father had died when he was fifteen, plunging his family into poverty. He bootstrapped his way to a degree in electrical engineering, then to a job at Tata Consulting Engineers in Mumbai, and then to a consultant gig at Walmart in Texas, where he climbed the ranks of its IT department and met Rick Dalzell.5 Raman was whip-smart, a tireless worker, and he had a reputation as an exceedingly demanding manager. He also had some memorable habits, including chewing an Indian betel leaf called pan during meetings and spitting into the garbage pails. Diane Lye, who ran Amazon’s data warehouse at the time and reported to Raman, sums it up this way: “Kal was a screamer.”

Applying his experience from Walmart, Raman pushed to build systems that finally realized Bezos’s vision of Amazon as a company with data at its heart. His groups created automated tools that allowed buyers to order merchandise based on dozens of variables such as seasonal trends, past purchasing behaviors, and how many customers were searching for a particular product at certain times. Raman’s teams also improved the software for pricing bots, which were automated programs that crawled the Web, spied on competitors’ prices, and then adjusted Amazon’s prices accordingly, ensuring that Bezos’s adamant demand that the company always match the lowest price anywhere, offline or online, would be met.

Buyers were held strictly accountable for keeping their products in stock and their prices competitive. If they somehow failed to deliver—if their shelves were suddenly empty or if Amazon’s prices were higher than a rival’s—then “Kal was going to personally hunt you down and kill you,” says Diane Lye, who worked for Raman for eighteen months. “There was so much fighting and yelling at each other. The technology was broken all the time and because the technology was broken, the data was often wrong. We would bring it to Jeff Bezos and it was all contradictory and he would be yelling and screaming at us. Oh, it was horrible.”

Raman spoke fast and had a thick accent, and his malapropisms, dubbed Kalisms, were legendary. “You all must be smoking cracks!” he yelled. Or “Can I have some of what you’re drinking so I can feel good about your business too?” He lasted at Amazon less than two years, but people at the company still talk about him.

“Kal was brutal,” says Jason Goldberger, a retail manager who worked for Raman. “He’s like out of a movie. The year after Katrina, I took over the home-improvement business, and he could not understand why generator business had fallen [compared to the increase that had accompanied the storm]. He’s such a driven personality.”

The turbulence caused by all these changes added to the overall dysfunction gripping Amazon at the time. The S Team was beset by a variety of internecine rivalries, perhaps typical for a large company. Raman and Piacentini, uncomfortably splitting ownership of the retail business, did not get along. Raman also battled with Jeff Wilke. At one point, Wilke heard that Raman had spoken negatively about the fulfillment team and he confronted him in a large meeting. “I heard there’s something you want to say to me,” Wilke said. “Do you want to say it in front of all these people?” Onlookers thought they might come to blows. In addition, Kathy Savitt, the vice president of communications, didn’t get along with Piacentini, and Jason Kilar, who had fully imbibed Bezos’s principles and mannerisms, had committed to run the video site Hulu but stuck around while he gave Amazon months to find his replacement.

Bezos handled it all poorly; it was as if the personal dramas were happening on a different dimensional plane that he couldn’t or didn’t want to access. As a result, the S Team, according to several of its members, became a highly combustible forum, a group in which everyone felt the need to be outspoken and curry favor with the boss and where political disputes were allowed to fester.

One of the biggest of those disputes was between Udi Manber and another technical leader of the company, Jeff Holden—Bezos’s former colleague at D. E. Shaw, the onetime teenage hacker who had dubbed himself the Nova.

Holden had been at Amazon longer than anyone else on the management team and had the closest personal relationship with Bezos. If members of the S Team were planets revolving around the sun, then Holden was Mercury, occupying a privileged orbit and drawing a fair amount of criticism, partly based on jealousy. Now in his midthirties, Holden remained a fast talker and a prodigious diet soda and Frappuccino drinker who always paced intensely during product meetings. Like Bezos, he was an aggressive manager who wanted to see results fast.

As senior vice president of worldwide discovery, Holden oversaw more than five hundred employees in Personalization, Automated Merchandising, Associates, E-Mail Marketing—and the department in charge of the search engine. It had been partly his idea to have Manber return to Palo Alto and run A9. But after a while, Holden began to feel that Manber’s group was too absorbed with the abstract challenges of general search and wasn’t focused enough on the practicalities of running the search for the Amazon website and solving nagging problems, such as latency, or the amount of time it took for searches on Amazon.com to generate results. The problem was that Holden retained ownership of the search experience on the website while Manber held responsibility for the search technology; they were basically dependent on each other.

Eventually, after growing increasingly frustrated, Holden concluded the situation was unworkable and, with an engineer named Darren Vengroff, started his own secretive effort in Seattle to rebuild Amazon’s search engine using the open-source tools Lucene and Solr. After a few months, Holden demonstrated the prototype to Bezos, who agreed to let them test it. Holden told Bezos he wanted to develop the Solr-based engine further and, if things went well, move search back to Seattle. Bezos said he’d think about it and later ran the proposal by Manber, who felt it was a sneak attack on his turf.

Now everyone was in a difficult spot. Bezos came back and told Holden and Manber to form a joint team to evaluate the new approach. There are various versions of what happened next, but the bottom line is that Manber and Holden didn’t like each other and couldn’t work well together. After the evaluation period ended, Bezos decided that search should remain the purview of A9. Holden was crestfallen. He argued that his organization had spearheaded the project and was doing the hard work to fix the persistent search problems on the site. Bezos pointed out that those were emotional concerns, not logical ones.

Feeling that Bezos had chosen Manber over him, Holden planned his departure from Amazon. With Vengroff, he would start a mobile search company called Pelago (which Groupon later acquired). Though this was a difficult time in their relationship, Holden and Bezos remained friends, and Bezos invested in Pelago. But when Holden left, Bezos lost one of his oldest friends at the company and one of Amazon’s most versatile innovators. Fortunately, he still had Udi Manber.

And then Manber decided to leave.

Manber said he didn’t like running a remote office and felt isolated from the decision-making in Seattle. Privately, he was annoyed that Bezos had allowed Holden’s rival search effort to take root in Seattle. He told Bezos and Rick Dalzell that he was considering going back to academia to do research in the science of memory. Bezos pleaded with Manber to stay on as what he called an Amazon Fellow. Manber said he would consider it.

Meanwhile, Urs Hölzle, one of Google’s first employees and its vice president of engineering, wanted to relinquish his oversight of search to focus on Google’s infrastructure. Hölzle invited Udi Manber to have dinner with him and surprised the Israeli scientist by asking if he was interested in replacing him as Google’s head of search engineering. Manber demurred at first, saying he was planning on getting out of the field. Then a few weeks later he changed his mind and decided that he might as well hear Google’s offer. Hölzle arranged a dinner that January with Larry Page in a private room of Il Fornaio, a restaurant in downtown Palo Alto. Page and Manber made sure to enter the restaurant separately. In the middle of dinner, Sergey Brin joined them. Google CEO Eric Schmidt showed up for dessert. It was an impressive full-court press.

By February, Manber had received an extraordinarily lucrative offer to run the search team at Google, and he decided to take it. Money aside, for any search engineer at the time, going to Google meant stepping onto the biggest playing field in the world and joining a championship-caliber team. For its part, Google had snagged one of the brightest minds in search and simultaneously decapitated the efforts of a competitor with one swift stroke.

Now Manber had to inform Bezos, right in the midst of so many other defections to Google. He delivered the news over the phone. Amazon employees would describe what happened next as one of Bezos’s all-time biggest nutters. Manber anticipated that Bezos would be disappointed and perhaps try to persuade him again to stay. “That’s what I had expected Jeff to do, but that’s not what he did,” Manber says. “He was clearly angry and he was dumping on me. I don’t recall now his exact words, but it was something like ‘No! No! No! You can’t do that!’ He was blaming me almost like I was a kid who did something very wrong.”

In that moment, Manber felt like he had lost a friend. He pleaded with Bezos that an engineer with his background and interests could not possibly decline the opportunity to run search at Google. Bezos viewed it as a personal betrayal. This time, he couldn’t brush away an employee’s departure easily. “He was not mincing words, and I felt horrible. He was always very good to me, the closest to a mentor that I ever had, and I was letting him down,” Manber says. “I don’t know if he ever forgave me, probably not, but I didn’t really have a rational choice. I [had] already decided to leave Amazon, so it was between moving to the top of my field or starting from scratch in a new field.”

A few days later Bezos calmed down and tried to get Manber to change his mind, but it was too late. Bezos had now lost his two closest colleagues and technical leaders, and just at the time that Amazon’s attempt to break out of retail and embrace an identity as a technology company was faltering. The general search engine at A9.com was a failure and was shut down a year after Manber left. Block View would be overtaken by Google’s Street View. Search Inside the Book was interesting but hardly a game changer, and the world’s best engineers were fleeing a poisonous Amazon culture and flocking to Google and other hot Internet companies in Silicon Valley. If Bezos was going to prove to the world that Amazon was indeed the technology company that he so desperately claimed it to be, he needed a dramatic breakthrough.

In early 2002, Web evangelist and computer book publisher Tim O’Reilly flew to Seattle to bend Jeff Bezos’s ear. O’Reilly, who would go on to help create a popular series of Web 2.0 technology conferences and the traveling festival for hardware hobbyists called the Maker Faire, thought that Amazon was acting too much like an isolated Web destination. He wanted the company to make available its sales data that could, for example, allow him and other book publishers to track various trends and help them decide what to publish next.6 Bezos hadn’t considered providing a broad range of such services to the outside world and initially replied that he didn’t see how that would benefit Amazon.

Over the years, O’Reilly and Bezos would have a friendly but sometimes adversarial relationship. In February of 2000, O’Reilly had organized an online protest against Amazon when it refused to allow other Internet retailers to use its patented 1-Click system. (Bezos cleverly blunted the campaign by joining in O’Reilly’s criticism of the patent system and supporting his idea for an independent company called BountyQuest, which, until it folded, allowed companies to post rewards for documents that undermined patents.7) O’Reilly also wrote a blog post urging fans of local bookstores to make their purchases there even if prices were cheaper online, arguing that those merchants would otherwise go away. That missive was taped to the cash registers of more than a few independent bookstores around the country.

But on this particular visit to Bezos in 2002, O’Reilly had a cogent case to make, and Bezos listened. The publisher showed Bezos Amarank, a sophisticated tool his company had created that visited the Amazon website every few hours and copied the rankings of O’Reilly Media books and the books of its competitors. It was a clunky process that relied on a primitive technique called screen scraping, and O’Reilly suggested that Amazon should develop a series of online tools called application programming interfaces, or APIs, that allowed third parties to easily harvest data about its prices, products, and sales rankings. O’Reilly spoke ambitiously about parceling out entire sectors of the Amazon store and allowing other websites to build on top of them. “Companies need to think not just what they can get for themselves from new technologies but how they can enable others,” he said.8

After O’Reilly’s visit, Bezos convened a meeting with Rick Dalzell, Neil Roseman, and Colin Bryar, the head of Associates at the time, to talk about the issue. Dalzell pointed out that there was already something like this under way inside the company and told Bezos about a young engineer named Rob Frederick whose mobile commerce startup, Convergence, Amazon had acquired in 1999. Frederick’s group was working on APIs that would allow non-PC mobile devices like phones and PalmPilots to access the Amazon store. After that meeting, Bezos invited O’Reilly to speak to a group of engineers, and later at an Amazon all-hands meeting, about lessons from computer history and the importance of becoming a platform.

Bezos added Frederick’s team to the Associates group under Colin Bryar and tasked them with creating a new set of APIs to let developers plug into the Amazon website. Soon other websites would be able to publish selections from the Amazon catalog, including prices and detailed product descriptions, and use its payment system and shopping cart. Bezos himself bought into the Web’s new orthodoxy of openness, preaching inside Amazon over the next few months that they should make these new tools available to developers and “let them surprise us.” The company held its first developer conference that spring and invited all the outsiders who were trying to hack Amazon’s systems. Now developers became another constituency at Amazon, joining customers and third-party sellers. And the new group, run by Colin Bryar and Rob Frederick, was given a formal name: Amazon Web Services.

It was the trailhead of an extremely serendipitous path.

Amazon Web Services, or AWS, is today in the business of selling basic computer infrastructure like storage, databases, and raw computing power. The service is woven into the fabric of daily life in Silicon Valley and the broader technology community. Startups like Pinterest and Instagram rent space and cycles on Amazon’s computers and run their operations over the Internet as if the high-powered servers were sitting in the backs of their own offices. Even large companies rely on AWS—Netflix, for example, uses it to stream movies to its customers. AWS helped introduce the ethereal concept known as the cloud, and it is viewed as so vital to the future fortunes of technology startups that venture capitalists often give gift certificates for it to their new entrepreneurs. Various divisions of the U.S. government, such as NASA and the Central Intelligence Agency, are high-profile AWS customers as well. Though Amazon keeps AWS’s financial performance and profitability a secret, analysts at Morgan Stanley estimate that in 2012, it brought in $2.2 billion in revenue.

The rise of Amazon Web Services brings up a few obvious questions. How did an online retailer spawn such a completely unrelated business? How did the creature that was originally called Amazon Web Services—the group working on the commerce APIs—evolve into something so radically different, a seller of high-tech infrastructure? Early observers suggested that Amazon’s retail business was so seasonal—booming during the holiday months—that Bezos had decided to rent his spare computer capacity during the quieter periods. But that explanation is widely debunked by Amazon insiders, in part because it would require Amazon to kick developers off its servers every fall.

The shift to offering these infrastructure services actually began with the transition to Gurupa and a more reliable technology infrastructure, a process that gathered momentum in 2003. While Amazon’s internal systems had been broken down into more durable individual components, Amazon’s technical staff was still organized conventionally as a single team, headquartered in a separate office building downtown near Seattle’s Union Station. This group strictly controlled who could access Amazon’s servers, and various teams inside the company had to plead for resources to try out their new projects and features. The process slowed down and frustrated many Amazon project managers. “You had a set of folks running these machines who were the priesthood of hardware, and the rest of us were railing against it,” says Chris Brown, a software-development manager at the time. “We wanted a playground where we could go to freely try things out.”

Bezos was getting annoyed as well. The company had improved on its pick-to-light system in the FCs, and its infrastructure had been successfully recast into component services, but the provisioning of computer resources remained a bottleneck. It got so dysfunctional that project leaders would present the S Team with their six-page narratives and then in the discussion afterward admit they had been unable to actually test their projects. Rick Dalzell recalls a particularly significant meeting when Matt Round, the head of Personalization at the time, complained that he didn’t have resources for experimentation. “Jeff finally just exploded at me,” Dalzell says. “I always handled Jeff’s outbursts pretty well, but to be honest about it, he had a right to be angry. We were stifling the flow of creativity. Even though we were probably faster than ninety-nine percent of companies in the world, we were still too slow.”

At the same time, Bezos became enamored with a book called Creation, by Steve Grand, the developer of a 1990s video game called Creatures that allowed players to guide and nurture a seemingly intelligent organism on their computer screens. Grand wrote that his approach to creating intelligent life was to focus on designing simple computational building blocks, called primitives, and then sit back and watch surprising behaviors emerge. Just as electronics are built from basic components like resistors and capacitors, and as living beings spring from genetic building blocks, Grand wrote that sophisticated AI can emerge from cybernetic primitives, and then it’s up to the “ratchet of evolution to change the design.”9

The book, though dense and challenging, was widely discussed in the book clubs of Amazon executives at the time and it helped to crystallize the debate over the problems with the company’s own infrastructure. If Amazon wanted to stimulate creativity among its developers, it shouldn’t try to guess what kind of services they might want; such guesses would be based on patterns of the past. Instead, it should be creating primitives—the building blocks of computing—and then getting out of the way. In other words, it needed to break its infrastructure down into the smallest, simplest atomic components and allow developers to freely access them with as much flexibility as possible. As Bezos proclaimed at the time, according to numerous employees: “Developers are alchemists and our job is to do everything we can to get them to do their alchemy.”

Bezos directed groups of engineers in brainstorming possible primitives. Storage, bandwidth, messaging, payments, and processing all made the list. In an informal way—as if the company didn’t quite know the insight around primitives was an extraordinary one—Amazon then started building teams to develop the services described on that list.

In late 2004, Chris Pinkham, head of the company’s IT infrastructure, told Dalzell that he had decided to return with his family to their native South Africa. At this point, A9 had taken root in Palo Alto, and Dalzell was busy establishing remote developer centers in Scotland and India, among other places. Dalzell suggested to Pinkham that instead of leaving Amazon, he open an office in Cape Town. They brainstormed possible projects and finally settled on trying to build a service that would allow a developer to run any application, regardless of its type, on Amazon’s servers. Pinkham and a few colleagues studied the problem and came up with a plan to use a new open-source tool called Xen, a layer of software that made it easier to run numerous applications on a single physical server in a data center.

Pinkham took colleague Chris Brown along with him to South Africa and they set up shop in a nondescript office complex in Constantia, a winemaking region northeast of Cape Town, near a school and a small homeless encampment. Their efforts would become the Elastic Compute Cloud, or EC2—the service that is at the heart of AWS and that became the engine of the Web 2.0 boom.

EC2 was born in isolation, with Pinkham talking to his colleagues in Seattle only sporadically, at least for the first year. The Constantia office had to make do with two residential-grade DSL lines, and during the hot summer of 2005, one of the country’s two nuclear reactors went offline, so engineers worked amid rolling brownouts. Pinkham later said that the solitude was beneficial, as it afforded a comfortable distance from Amazon’s intrusive CEO. “I spent most of my time trying to hide from Bezos,” Pinkham says. “He was a fun guy to talk to but you did not want to be his pet project. He would love it to distraction.”

The dozen engineers concurrently developing what would become Amazon’s Simple Storage Service, or S3, did not have that luxury, despite their best attempts to keep to themselves. They worked in an office on the eighth floor of Pac Med, ate lunch together every day for nearly two years, and often played cards together after work. Neither they nor their manager, Alan Atlas, a veteran of crosstown digital-media startup Real Networks, were able to hide half a world away.

Bezos was deeply interested in the evolution of Web services and often dived into the minutiae of S3, asking for details about how the services would keep up with demand and repeatedly sending engineers back to the drawing board to simplify the S3 architecture. “It would always start out fun and happy, with Jeff’s laugh rebounding against the walls,” Atlas says. “Then something would happen and the meeting would go south and you would fear for your life. I literally thought I’d get fired after every one of those meetings.”

Atlas said that while working on the S3 project, he frequently had difficulty grasping just how big Bezos was thinking. “He had this vision of literally tens of thousands of cheap, two-hundred-dollar machines piling up in racks, exploding. And it had to be able to expand forever,” Atlas says. Bezos told him, “This has to scale to infinity with no planned downtime. Infinity!”

During one meeting, Atlas blundered by suggesting they could figure out how to keep up with any unexpected growth after the service launched. That triggered a Bezos nutter. “He leaned toward me and said, ‘Why are you wasting my life?,’ and went on a tirade about Keystone Cops,” Atlas says. “That was real anger. I wasn’t keeping up with him. There were a number of times like that. He was so far ahead of us.”

For the launch of Simple Storage Service, Atlas had commemorative T-shirts made up for his colleagues; he used the design of Superman’s costume but with an S3 rather than an S on the chest. Naturally, he had to pay for the shirts himself.

As their ambitions for Web services expanded between South Africa and Seattle, Bezos and Dalzell began to consider who would lead the effort. Bezos suggested Al Vermeulen, Amazon’s chief technology officer, but Al V. commuted to Seattle by plane every day from Corvallis, Oregon, and said he didn’t want an administrative job. He demoted himself to an engineer working on S3 with Alan Atlas. So Dalzell recommended Andy Jassy, who had inauspiciously started his career at Amazon so many years ago by hitting Jeff Bezos with a kayak paddle during the company’s first game of broomball.

If the new era in high-tech was indeed to be the age of computer science PhDs, then Jassy would prove to be a conspicuous anomaly. A graduate of Harvard Business School with a passion for buffalo wings and New York sports teams, Jassy seemed unlikely to fit in at a geeky technology startup. Perhaps as a result of this, his path at Amazon had been meandering and sometimes difficult. It was Jassy who presented the original business plan in 1998 for Amazon to enter the music business, but then he watched in disappointment as another executive was selected to lead the charge. A few years later, in a companywide reorganization, Jassy was chosen to oversee the personalization group, but then the engineers in that department objected to being led by someone they viewed at the time as nontechnical.

So Jassy was given a unique opportunity. Bezos asked him to become his first official shadow—a new role that would entail Jassy’s following around the CEO and sitting with him in every meeting. Other technology companies like Intel and Sun had similar positions, and Bezos had tried this before with executives who were new to the company—including a D. E. Shaw engineer named John Overdeck and Accept.com founder Danny Shader—but it had never been a full-time job, and many of those previous shadows had subsequently left the company. Jassy was conflicted about the proposal. “I was flattered by the offer to work closely with Jeff but wasn’t initially excited about it because I’d seen the way it had gone down before,” he says. “I asked Jeff what success would look like. He said success would be if I got to know him and he got to know me and we built trust in each other.” Jassy agreed and spent much of the next eighteen months by Bezos’s side, traveling with him, discussing the events of each day, and observing the CEO’s style and thought process. Jassy would define the shadow role as a quasi chief of staff, and today the position of Bezos’s shadow, now formally known as technical adviser, is highly coveted and has broad visibility within the company. For Bezos, having an accomplished assistant on hand to discuss important matters with and ensure that people follow up on certain tasks is another way to extend his reach.

As Jassy’s tenure as shadow ended, he became a natural candidate to step in as the new head of AWS. One of his first jobs was to write a vision statement; he had to tinker with the margins to get it under six pages. The paper laid out the expanded AWS mission: “to enable developers and companies to use Web services to build sophisticated and scalable applications.” The paper listed the primitives that Amazon would subsequently turn into Web services, from storage and computing to database, payments, and messaging. “We tried to imagine a student in a dorm room who would have at his or her disposal the same infrastructure as the largest companies in the world,” Jassy says. “We thought it was a great playing-field leveler for startups and smaller companies to have the same cost structure as big companies.”

Jassy, Bezos, and Dalzell presented the plan for the new AWS to the Amazon board, and the institutional no came close to rearing its ugly head. John Doerr, expressing what he would later call a “healthy skepticism,” asked the obvious question: At a time when Amazon was having difficulty hiring engineers and needed to accelerate its international expansion, “Why would we go into this business?”

“Because we need it as well,” Bezos replied, suggesting that Amazon’s demand for such a service reflected the broader market need. Jassy remembers Doerr telling him after the meeting that he was lucky to work at a company that would invest in something so daring.

Around this time, Bezos was pursuing another project whose origins were shaped by resistance from Amazon’s board of directors. When the founders of the comparison-shopping site Junglee had left Amazon in the late 1990s, they had done so on good terms and with the agreement that they would stay in touch with Amazon executives and even coordinate their efforts. Anand Rajaraman and Venky Harinarayan, two of the cofounders, then started their own Internet incubator called Cambrian Ventures, and Bezos wanted Amazon to invest in it. But in a rare act of resistance to Bezos’s will, the Amazon board vetoed the move. So Bezos ended up investing personally. The circuitous result of that decision was yet another unlikely Web service from Amazon and another sign that the company was trying to evolve beyond online retailing.

At the time that Cambrian Ventures was starting up in Silicon Valley, the peer-to-peer file-sharing service Napster was dominating headlines and panicking the music business. The Cambrian engineers thought about Napster and the power of networks that linked people who were scattered around the world. Could you do something of value with those distributed networks, they wondered, something better than stealing music? That question became the seed of an initiative they called Project Agreya—the Sanskrit word for “first.”

The idea was to build software and harness the Internet to coordinate groups of people around the world to work on problems that computers weren’t very good at solving. For example, a computer system might have difficulty examining a collection of photos of domestic pets and reliably selecting the ones that depicted cats or dogs. But humans could do that easily. The Cambrian Ventures executives hypothesized that they could build an online service to coordinate low-wage workers around the world and then sell access to this workforce to financial firms and other large companies. In 2001, they filed for a patent for the idea and called it a “Hybrid machine/human computing arrangement.”10

The world would later come to know this idea, and embrace it, as crowd-sourcing. But Project Agreya was ahead of its time, and financial firms didn’t know quite what to make of it. The Agreya team was in New York trying to pitch the concept the week of 9/11. Venture capital dried up after the terrorist attacks, so they shut down Project Agreya and moved on.

In 2003 Rajaraman and Harinarayan decided to close Cambrian Ventures to work on a new company called Kosmix, which would develop technology to organize information on the Web by specific topics. As part of winding down Cambrian, they had to deal with Bezos and his investment in the firm. Not surprisingly, Bezos proved to be a tenacious negotiator and a stubborn defender of his own financial interests, even when the stakes were a relatively inconsequential fraction of his net worth. Rajaraman and Harinarayan recall a brutal two-month-long negotiation during which Bezos leveraged the dissolution of Cambrian Ventures into a stake in Kosmix. In the midst of this, they happened to tell Bezos about the Agreya patent, and he was immediately interested and asked for it to be included in the overall deal. Seeing an opportunity to conclude the arrangement, they quickly agreed to sell it to him.

To their surprise, Bezos then actually developed a version of Project Agreya inside Amazon. He renamed it Mechanical Turk, after an eighteenth-century chess-playing automaton that concealed a diminutive man—a chess master—who hid inside and guided the machine’s moves. About two dozen Amazon employees worked on the service from January 2004 to November 2005. It was considered a Jeff project, which meant that the product manager met with Bezos every few weeks and received a constant stream of e-mail from the CEO, usually containing extraordinarily detailed recommendations and frequently arriving late at night.

Amazon started using Mechanical Turk internally in 2005 to have humans do things like review Search Inside the Book scans and check product images uploaded to Amazon by customers to ensure they were not pornographic. The company also used Mechanical Turk to match the images with the corresponding commercial establishments in A9’s fledgling Block View tool. Bezos himself became consumed with this task and used it as a way to demonstrate the service.

As the company prepared to introduce Mechanical Turk to the public, Amazon’s PR team and a few employees complained they were uncomfortable with the system’s reference to the Turkish people. Bezos liked the name for its historical association but agreed to let the communications staff and Mechanical Turk team brainstorm alternatives. They seriously considered Cadabra, an allusion to magic and the original corporate name of Amazon. But in the end, Bezos shrugged off the concerns and said that he personally would bear the responsibility for any backlash.

Mechanical Turk quietly launched in November 2005. Now any Internet user could perform what Amazon called human-intelligence tasks, typically earning a few cents per job. Other companies could list jobs on the Mechanical Turk website, with Amazon taking a 10 percent cut of the payments.11 One of the first applications, from a company called Casting Words, paid workers a few cents per minute to listen to and transcribe podcasts.

Mechanical Turk gave Bezos another opportunity to demonstrate Amazon’s ability to innovate outside of its core retail business and show off his own curious attempts to crystallize abstract concepts. He called Mechanical Turk “artificial, artificial intelligence” and gave interviews about the service to the New York Times and the Economist. The ethnic reference in the name was never criticized, but labor activists targeted the service as a “virtual sweatshop” and “the dark side of globalization.”12

By 2007 there were a hundred thousand workers on Mechanical Turk in more than one hundred countries.13 But it didn’t take off in the way Bezos clearly hoped it would, or at least it hasn’t yet. One obvious reason is that the exceedingly low wages on Mechanical Turk have the greatest appeal in less developed countries, yet most impoverished workers in the third world do not own Internet-connected PCs. When Amazon’s other Web services unexpectedly took off in the following years, Bezos devoted considerably more attention and resources to them. Just as in Amazon’s early days, when automated personalization replaced editorial, machines, not people hiding inside them, would drive Amazon’s long-awaited big breakthrough.

In March 2006, Amazon introduced the Simple Storage Service, which allowed other websites and developers to store computer files like photos, documents, or video-game player profiles on Amazon’s servers. S3 remained alone and somewhat overlooked, like a section of a fence that had not yet been finished. A month after the launch, Alan Atlas recalled, it crashed for nine hours, and hardly anyone in the outside world noticed. Then a few months later, the Elastic Compute Cloud went to public beta, allowing developers to actually run their own programs on Amazon’s computers. According to Chris Brown, who returned from South Africa for the launch, Amazon opened the first servers to customers on the East Coast of the United States, and developers rushed in so quickly that the initial batch of computers was taken up before Amazon had a chance to let in folks on the West Coast.

Part of AWS’s immediate attraction to startups was its business model. Bezos viewed Web services as similar to an electric utility that allowed customers to pay for only what they used and to increase or decrease their consumption at any time. “The best analogy that I know is the electric grid,” Bezos said. “You go back in time a hundred years, if you wanted to have electricity, you had to build your own little electric power plant, and a lot of factories did this. As soon as the electric power grid came online, they dumped their electric power generator, and they started buying power off the grid. It just makes more sense. And that’s what is starting to happen with infrastructure computing.”14

Bezos wanted AWS to be a utility with discount rates, even if that meant losing money in the short term. Willem van Biljon, who worked with Chris Pinkham on EC2 and stayed for a few months after Pinkham quit in 2006, proposed pricing EC2 instances at fifteen cents an hour, a rate that he believed would allow the company to break even on the service. In an S Team meeting before EC2 launched, Bezos unilaterally revised that to ten cents. “You realize you could lose money on that for a long time,” van Biljon told him. “Great,” Bezos said.

Bezos believed his company had a natural advantage in its cost structure and ability to survive in the thin atmosphere of low-margin businesses. Companies like IBM, Microsoft, and Google, he suspected, would hesitate to get into such markets because it would depress their overall profit margins. Bill Miller, the chief investment officer at Legg Mason Capital Management and a major Amazon shareholder, asked Bezos at the time about the profitability prospects for AWS. Bezos predicted they would be good over the long term but said that he didn’t want to repeat “Steve Jobs’s mistake” of pricing the iPhone in a way that was so fantastically profitable that the smartphone market became a magnet for competition.

The comment reflected his distinctive business philosophy. Bezos believed that high margins justified rivals’ investments in research and development and attracted more competition, while low margins attracted customers and were more defensible. (He was partly right about the iPhone; its sizable profits did indeed attract a deluge of competition, starting with smartphones running Google’s Android operating system. But the pioneering smartphone is also a fantastically lucrative product for Apple and its shareholders in a way that AWS has not been, at least so far.)

Bezos’s belief was borne out, and AWS’s deliberately low rates had their intended effect; Google chairman Eric Schmidt said it was at least two years before he noticed that the founders of seemingly every startup he visited told him they were building their systems atop Amazon’s servers. “All of the sudden, it was all Amazon,” Schmidt says. “It’s a significant benefit when every interesting fast-growing company starts on your platform.” Microsoft announced a similar cloud initiative called Azure in 2010. In 2012, Google announced its own Compute Engine. “Let’s give them credit,” Schmidt says. “The book guys got computer science, they figured out the analytics, and they built something significant.”

Just like Creation author Steve Grand had predicted, the creatures were evolving in ways that Bezos could not have imagined. It was the combination of EC2 and S3—storage and compute, two primitives linked together—that transformed both AWS and the technology world. Startups no longer needed to spend their venture capital on buying servers and hiring specialized engineers to run them. Infrastructure costs were variable instead of fixed, and they could grow in direct proportion to revenues. It freed companies to experiment, to change their business models with a minimum of pain, and to keep up with the rapidly growing audiences of erupting social networks like Facebook and Twitter.

All of this took years and required significant effort, and its developers encountered many challenges and setbacks along the way. Andy Jassy, along with technical lieutenants Charlie Bell and Werner Vogels, outpaced rivals by layering additional services like Flexible Payment Services and Amazon CloudSearch alongside of EC2 and S3. Groups within Amazon were told to use AWS while the services were still immature, a demand that led to another round of consternation among its engineers. As startups and even some big companies began to rely heavily on AWS, outages had widespread repercussions, and chronically secretive Amazon found it had to get better at explaining itself and speaking to the public.

But the emergence of Amazon Web Services was transformational in a number of ways. Amazon’s inexpensive and easily accessible Web services facilitated the creation of thousands of Internet startups, some of which would not have been possible without it, and it provided larger companies with the ability to rent a supercomputer in the cloud, ushering in a new era of innovation in areas like finance, oil and gas, health, and science. It is not hyperbole to say that AWS, particularly the original services like S3 and EC2, helped lift the entire technology industry out of a prolonged post-dot-com malaise. Amazon also completely outflanked the great hardware makers of the time, like Sun Microsystems and Hewlett-Packard, and defined the next wave of corporate computing.

Perhaps the greatest makeover was of Amazon’s own image. AWS enlarged the scope of what it meant to be the everything store and stocked Amazon’s shelves with incongruous products like spot instances and storage terabytes. It made Amazon a confusing target for Walmart and other rival retailers and gave the company fresh appeal to the legions of engineers looking to solve the world’s most interesting problems. Finally, after years of setbacks and internal rancor, Amazon was unquestionably a technology company, what Bezos had always imagined it to be.