The cutting edge of the Web just bled a little.
On Friday night, lightning in Virginia took out part of Amazon’s cloud computing service, called Amazon Web Services, which hundreds of companies use for data storage and computation. Well-known sites like Netflix , Pinterest and Instagram were not accessible for hours. There was little information for customers about what had happened, or even whether user data was safe.
The interruption underlined how businesses and consumers are increasingly exposed to unforeseen risks and wrenching disruptions as they increasingly embrace life in the cloud. It was also a big blow to what is probably the fastest-growing part of the media business, start-ups on the social Web that attract millions of users seemingly overnight.
Besides Internet-connected computers, tablets and smartphones, which people rely on daily, consumers and businesses are connected to the Web through everything from cars and appliances to utility monitors and surveillance cameras. How well the systems behind these are built, and how they handle unforeseen disruptions, will increasingly affect the larger economy.
Amazon has built a thriving business in cloud computing, with a range of customers including Intercontinental Hotels, Fox Entertainment, Unilever, Spotify, as well as 187 government agencies and hundreds of small start-ups looking for the cheapest possible computing.
None of these big customers reported any service disruptions. While it was not clear if they were using the Amazon center that was crippled, analysts said the disruption would cause renewed scrutiny of their dependence on cloud computing.
“The way companies view it is in terms of reliability generally,” said Michael Chui, a senior fellow at McKinsey & Company. Big customers of Amazon, he said, “have the opportunity to shape the marketplace and make demands that make products better. They will push for improvements.”
They will also have another option. On Thursday, Google said it would offer computing over the Internet at half the price of Amazon.
The weekend’s disruption happened after a lightning storm caused the power to fail at the Amazon Web Services center in Northern Virginia containing thousands of computer servers. For reasons Amazon was still unsure of on Sunday, the data center’s backup generator also failed.
By midday Saturday, Amazon said in a statement that it had restored service “to most of our impacted customers, and continue to work to restore service to our remaining impacted customers,” adding, “we will share more details on this event in the coming days.” The company had no further comment.
It was at least the second major failure for Amazon in that area. In April 2011, a problem in Amazon’s networking at a nearby data center took down a number of applications and popular Web sites, including Reddit and Quora, for more than a day.
To be sure, Amazon Web Services held up better than many utilities affected by the storm, which left more than four million homes and businesses without power on the East Coast. By Sunday, some Instagram customers in the Washington area could, through their cellphones, share pictures on the Web before they had lights or safe drinking water.
Netflix, along with Pinterest and Instagram, did not respond to requests for comment Sunday. Netflix was able to restore service in a couple of hours. Many more Amazon customers, including the small start-ups that make up the bulk of its customers, did not report any problems, but may have scrambled to make sure their systems were secure.
“We were late at work, playing an online video game that went down, then got a notification from our own system that it was starting to fail,” said Benjamin Coe, the founder of Attachments.me, a start-up in San Francisco that connects e-mail attachments to online data storage services. “We had built for redundancy, but our system failed in two areas, making it hard for the third to work.”
Mr. Coe said he stayed up until 4 a.m. fixing the problem. It was, he said, the price of working cheaply with a service that enabled his six-person company to potentially serve millions of customers, who he said would be understanding about brief shutdowns.
“For a consumer product that is free, there is an understanding all around that things can fail occasionally,” he said. “If you were flying a plane or something, this would be totally unacceptable.”
The ability to deal with failures has long been a feature of any computing system, but like much else in the cloud, there are no common standards to guide how much protection against disaster is enough. Many start-ups appear not to take advantage of more expensive redundancy features in Amazon, like swapping data between the East Coast and West Coast Amazon facilities.
Bigger companies are moving to the cloud as well, but may now look at Amazon Web Services as a stopgap as much as a primary provider.
“Maybe 15 percent of our customers now keep their data with us in the cloud, but in three years that could be 75 percent,” said Carl Bass, chief executive of Autodesk, a leading maker of design software. Most of that data is stored on Amazon servers for now, he said, but the company is weighing managing its own cloud, as customers use cheap cloud computing to generate huge numbers of design simulations.
“It’ll cost $10 million to get started, and we’ll probably still rely on A.W.S. to handle big workloads,” he said. “We’re going to need tools for infinite computing.”