The robots are mesmerizing. Inside a warehouse in Erith, on the outskirts of East London, more than a thousand of them glide across a vast steel and aluminum grid. Each is the size and shape of an office copy machine, topped with stubby antennae and a shining neon-green LED. Following individual routes, they whiz off, accelerating at rates rivaling those of a Ferrari. They stop on a dime, reverse, shoot left or right, or momentarily pause to allow fellow robots to pass—a meticulously choreographed electric ballet.

The robots’ grid is actually the top of a giant three-dimensional lattice—a modular cage packed with groceries. Each time a robot stops, it drops a clawlike attachment into the bowels of the lattice (“the hive,” as human workers call it), descending as many as three stories. The claw grabs the sides of a white plastic crate containing fruit, vegetables, cereal—any of 55,000 different items—and retracts it up into the robot’s belly. The robot then carries the crate to another grid square and lowers it into the “pick tunnel,” which sits beneath the hive on the warehouse’s ground floor. There, workers pick items out of the crates to fill customers’ orders, placing the groceries into red plastic bins, which are then loaded onto trucks for delivery.

This warehouse, or customer fulfillment center (CFC), as logistics pros call it, is one of the most sophisticated and automated on the planet, one that can handle many tens of thousands of orders a week. It belongs to Ocado, a pioneering British online grocer that is positioning itself as a white knight for the beleaguered grocery sector—and possibly other industries too—offering to help supermarket chains compete in an automated age.

Ocado’s robot-powered warehouses thrum with activity on ordinary days; since the coronavirus crisis erupted, they’ve been in roaring overdrive. The pandemic has given the company a chance to prove it can keep an online grocery business humming, even when its human workforce faces unprecedented strain. Yet at the same time, the crisis’s upending of daily life has threatened to knock Ocado off its growth trajectory, just when it seemed tantalizingly close to becoming a global force.

Like most grocers, Ocado has faced skyrocketing demand fueled by social distancing measures and panic buying. Its U.K. grocery sales in March leaped more than 20% year over year. At one point, visits to its website were 100 times the normal rate—a level so high, it triggered the company’s cybersecurity systems to believe the website was under attack. “This is the peakiest peak we’ve ever had in the history of the business,” says David Shriver, Ocado’s group director of communications.

Ocado is hardly alone in this. Consultants McKinsey & Co., in a note to grocery clients on March 19, reported that online grocers worldwide were struggling to meet demand spikes as high as 700%. There’s a good chance the pandemic will have a lasting impact on consumer behavior, converting many more customers to online grocery shopping long after the crisis recedes. Supermarkets will need ways to meet this new demand as efficiently as possible—which ought to boost interest in Ocado’s automated warehouses, robots, and software.

Still, like its peers, Ocado has sometimes stumbled as it races to keep up. To try to slow the overwhelming order volume, Ocado shut down its mobile app and implemented a queuing system on its website. But demand was simply too great, and with all its delivery slots booked for a week out, Ocado was forced to temporarily shut down its website in mid-March. When it came back online, almost a week later, the company restricted orders to existing customers; even so, delivery slots remain difficult to find. And, like many grocers, Ocado temporarily limited customers to two or three of any individual item to deter hoarding.

Ocado is now preparing for other worst-case scenarios; it has held rehearsals for what will happen if its employees begin to fall ill and entire teams need to self-isolate. (One possible solution: drafting furloughed workers from other industries to staff warehouses and drive trucks. The company has already reached out to recruit idled Uber drivers.)

Even before the coronavirus hit, grocers were under tremendous pressure. “Both growth and profitability have been on a downward trajectory due to higher costs, falling productivity, and race-to-the-bottom pricing,” McKinsey wrote in a recent report. Sales growth for North American and Western European grocers has been just 2% over the past decade. Net profit margins for U.S. grocery chains, meanwhile, average just 1% to 2%, according to consulting firm Mercator Advisory Group. On top of these dismal trends, executives in the sector fear the twin Death Stars of retail—Amazon and Walmart—both of which have made clear their intentions to dominate the grocery business. Walmart is already by far the largest seller of groceries in the U.S., with a 21.3% market share, according to UBS, more than twice the share of its nearest competitor, supermarket chain Kroger.

Some industry insiders argue that the only way “legacy” grocers can compete with the titans is by matching their state-of-the-art technology and logistics infrastructure. That automation could also help them trim labor costs: another McKinsey report, from May 2019, estimated that by implementing existing technologies, a grocer could potentially run a store with 55% to 65% fewer labor-hours.

Ocado’s pitch to grocers stresses those benefits and adds a compelling twist: Ocado can build the automation infrastructure for them, sparing them the pains and costs of developing their own.

For many years, Ocado’s talk of becoming a tech platform seemed to be just that: talk. Equity analysts were skeptical, and the stock became a perennial favorite among short-sellers. But Ocado’s logistics prowess has gradually won converts. Beginning in 2017, the company announced a series of licensing deals with grocery chains on four continents—including a huge partnership with Kroger. Since then, Ocado’s market capitalization has quadrupled to north of $10 billion, while revenue has been growing at about 10% annually, reaching $2.2 billion in 2019. That kind of growth is nothing for a tech company—but it’s exceptional for a grocery.

Investors seem confident that Ocado can capitalize on the current moment; its shares have risen 28% since Feb. 28, even as global markets plummeted. The future, however, looks murkier. Ocado’s licensing deals require it to spend heavily upfront to build dozens of CFCs. Those obligations have left some analysts wondering if the company, which currently carries more than $750 million in debt, has taken on too much risk. And, even once the pandemic passes, an existential question looms: whether any grocer, even with Ocado’s robotic helping hand, can withstand the onslaught of the Everything Store and the Behemoth of Bentonville.

Ocado’s roots stretch back to the dotcom boom, when three twentysomething Brits working as traders at Goldman Sachs—Tim Steiner, Jonathan Faiman, and Jason Gissing—got bitten by the startup bug. The trio founded Ocado in April 2000. (The name is an invented word, chosen because it could work across languages and because the founders liked how it looked as a logo.) Steiner, Ocado’s CEO, is the only founder still involved with the company. Compact and trim, with close-cropped gray hair and pale blue eyes, he exudes a pugilistic intensity as he walks through Ocado’s history at a rapid-fire clip.

A strategic partnership helped get the new company going. Realizing it was too small to get the best prices from wholesalers, Ocado signed a deal with the upmarket U.K. supermarket chain Waitrose, allowing Ocado to piggyback on its deals with suppliers. In exchange, Ocado agreed to carry Waitrose’s white label products for ten years (a deal that was later extended for an additional decade). Waitrose’s parent company, the retailer John Lewis, invested $45 million for a 40% stake in the company.

When Ocado made its debut, established British grocery chains such as Tesco, Sainsbury’s, and Walmart-owned Asda already had e-commerce operations. Those chains were using their stores to fulfill online orders, with store clerks gathering the goods and loading them onto delivery trucks. This process, known as “store pick,” is the way most retailers have grafted e-commerce onto their existing operations. Store pick requires little additional capital or labor investment, but it has disadvantages. Many stores’ stockrooms are too cramped to accommodate a sizable picking operation, which means employees may have to fill online orders from the supermarket floor, putting them in competition with in-store customers. With smaller inventories, there’s also a greater chance that items won’t be available—a leading driver of customer dissatisfaction.

Ocado, which had no stores, took a different approach. It built an automated central distribution center in Hatfield, on London’s northern edge, and delivered all its orders from there—a strategy that helped it minimize inventory shortfalls. The business quickly became popular, consistently topping consumer surveys. Its rate of substituted or erroneous items in orders—which it drove down below 0.5%—was much better than those of its rivals.

The problem: The technology at Hatfield left a lot to be desired. The equipment Ocado was using—giant conveyor belts and sorting machines—was designed for the manufacturing sector, where factories churn out mass volumes of identical items. It was ill-suited for the grocery business, where the assortment of items is huge, and each customer’s order is unique. “I used to joke about the law of material-handling equipment, which was, Five plus five equals seven,” Steiner says.

Constant spending on improvements, meanwhile, was eating up cash. By 2007, Ocado’s revenues were $338 million, and galloping ahead at more than 20% annually. But the company was also running annual operating losses of more than $45 million—and spending tens of millions more to improve its CFC and purchase delivery trucks. It had more than $130 million of debt. At one point, John Lewis wrote its investment in the young online grocer down to zero.

Around this time, Paul Clarke, a software consultant with experience running tech startups, received a call from an Ocado recruiter. “I said, ‘Look, I’m really sorry, but I don’t want to work in retail,’” recalls Clarke, a lanky 60-year-old with the professorial demeanor of the Oxford PhD he once considered becoming. But when he toured Ocado’s warehouse, Clarke was impressed by its scale and complexity. Hatfield was a giant automation puzzle—exactly the sort of engineering problem he enjoyed cracking. “I fell in love,” he says.

Clarke signed on for a one-year gig, tasked with improving the system that controlled the flow of goods along Hatfield’s fast--moving conveyor belts. Ocado’s operation was so complex, Clarke says, that the only way to reengineer it was to build a series of digital “twins”—in essence, real-time software simulations of the operation. This allowed Clarke and his team to experiment with improved configurations before implementing them in the real warehouse, avoiding costly trial and error. Before long, Steiner says, the twins helped wring new efficiencies from equipment—making five plus five equal 12.

In the summer of 2010, Ocado went public on the London Stock Exchange, in a listing that valued the company at 937 million pounds ($1.4 billion at the time). That was more than many analysts thought the money-losing grocer was worth, and its shares fell 10% on their first day of trading. That skepticism would continue to haunt Ocado: Over the next decade, its shares would frequently have the dubious distinction of the being among the market’s most shorted, a Greek chorus of hedge fund managers cheering for its failure.

Over the following year, though, Ocado eked out its first small operating profit. Around the same time, grocery consultants, investment banks, and, eventually, huge packaged goods companies—Procter & Gamble, Unilever, Nestlé, and Coca-Cola—began quietly asking to tour the company’s fulfillment centers. Steiner’s instinct was to refuse. “We were quite secretive,” he recalls. But he soon realized that while other companies might glean a few tips by touring the CFCs, they couldn’t replicate the integrated system of software, hardware, warehouse workers, and delivery drivers Ocado had built over a decade. In its 2012 annual report, Ocado for the first time made monetizing its intellectual property a strategic plank.

Among those who visited Ocado was the U.K. grocer Morrisons, which had revenues far exceeding Ocado’s but didn’t have an online offering. The two companies reached a deal that sold half the capacity of Ocado’s newest CFC to Morrisons, with Ocado agreeing to manage the facility and a delivery fleet on Morrisons’ behalf. When Morrisons.com launched in January 2014, it was the first evidence that Ocado could put its platform to work for other grocers.

Clarke, who by then had been promoted to chief technology officer, had an even more ambitious version of that platform in mind. The day after Ocado launched Morrisons, Clarke gathered his staff for an off-site in a local hotel. He congratulated them on having gotten Morrisons up and running on schedule and in just six months—a significant technical achievement. What he said next made many in the room gasp. “We’re going to rewrite the whole technology stack from scratch,” he told them. At first, this largely meant software improvements. But, by mid-2015, Ocado had begun developing the army of robots that would eventually staff its “hives.”

The robots, designed by Ocado in conjunction with U.K. robotics company Tharsus, are controlled by an internal 4G network with more base stations packed into less space than pretty much anywhere else on the planet. The network enables each robot to communicate with the software controlling it 10 times per second. At Erith, the hive generates four terabytes of data every day, all of which is fed back into a digital twin to refine the system.

The robots allow Ocado’s newest fulfillment centers to pick 200 items per hour of labor time—and mean they can move a typical order from inbound supply truck into the hive, and then have it picked, packed, and loaded on a van for delivery in 15 minutes or less. Meanwhile, the modular design of the hive itself means it can easily be replicated and sized to fit new locations.

Complementing the hardware is new software—lots of it, from cloud-based mobile apps to artificial intelligence. This integrated package, along with the engineering support to maintain and upgrade it, is what Ocado now offers to the world’s grocers.

Amazon and Walmart are no strangers to robotics. Amazon has begun using Roomba-like robots from Kiva, a Massachusetts company it acquired in 2012 for $775 million, to move large stacks of pallets around its fulfillment centers. Last year, it also acquired Canvas, a Boulder, Colo., startup whose computer vision systems allow the warehouse robots it builds to work in crowded conditions alongside people. Walmart has deployed thousands of robots from Bossa Nova Robotics, a Pittsburgh company, to keep track of stock in its stores. It has also created a pilot, fully automated warehouse in New Hampshire that serves its order-online-and-customer-pickup grocery business.

Pure-play supermarkets have been far slower to automate. But in June 2017, a major move by Amazon gave Ocado’s modernization sales pitch a Saturn V–size boost. That was when the Everything Store spent $13.7 billion to buy upscale grocer Whole Foods, which had 500 stores worldwide. The deal stoked grocers’ fears that Amazon would decimate them as it had so many retailers in other categories—and the trickle of interest in Ocado’s technology became a torrent.

In November 2017, Ocado announced a deal with French retailer Groupe Casino to supply the technology for its e-commerce in France. Two months later, it partnered with Sobeys, which operates 1,500 stores under a variety of brand names across Canada. “It’s the only profitable e-commerce model at scale that I’ve seen,” Sarah Joyce, Sobeys senior vice president for e-commerce, says of Ocado.

Several other deals followed, including with ICA, a Swedish company that operates 1,300 groceries; with Coles, in Australia; and with Aeon, Asia’s largest supermarket chain, in Japan. But the biggest of them all was the strategic partnership Ocado reached in May 2018 with Kroger. The American giant took a 5% share in Ocado and gained exclusive U.S. rights to its technology; Ocado committed to building about 20 CFCs for Kroger. The British company’s shares soared 44% on the day the partnership became public.

Rodney McMullen, Kroger’s CEO, says he had been watching Ocado for a decade, meeting periodically with its top executives. Kroger implemented a store pick–based -e-commerce operation after a 2013 merger, but McMullen says it became unwieldy as it grew. The struggle to keep both online and in-store customers happy was driving Kroger toward towards the same sort of automated fulfilment centers Ocado had. McMullen admits to keeping a close eye on Amazon and Walmart, but he insists the Ocado partnership is not about what competition might do. “We didn’t see a path where we could accelerate to where Ocado is in a year or two,” McMullen says.

It’s a common refrain among Ocado’s customers: They lack the resources to replicate Ocado’s technology. “We are a big company, but we are not a technology company,” says Anders Svensson, the CEO of ICA Sweden. Ocado, in contrast, employs more than 1,800 software engineers and 600 hardware specialists. That’s far less than Amazon or Google, but it’s a lot for a grocer.

The slew of licensing deals pushed many investors to abandon their skepticism: It’s been a long time since Ocado was a heavily shorted stock. “I used to be a big bear, but I became a big bull after the Kroger deal,” says Bruno Monteyne, an analyst who covers European retailers for the equity research firm Bernstein Research. Still, those deals aren’t adding much to the bottom line—because Ocado receives money only after the automated CFCs are built.

Ocado currently operates six CFCs to support its U.K. grocery operation; it aims to run at least 50 worldwide within the decade. Its first CFC outside Britain, built for France’s Groupe Casino, went live on March 26. Another, for Sobeys, outside Toronto, should open by June. And its first center for Kroger is scheduled to come online in Monroe, Ohio, in the first half of 2021.

Ocado’s partners are responsible for acquiring land, building external structures, providing a delivery fleet, and hiring workers. But Ocado has to build the hives, supply the robots and software, and provide training and on-site engineering support. It costs Ocado $40 million to $45 million in “peak cash outflow” for each average-size CFC, Steiner says. Only after construction does Ocado collect a fee based on the warehouses’ available capacity. In its most recent fiscal year, just 6% of Ocado’s revenue came from licensing.

Sherri Malek, an equity analyst at RBC Capital Markets, says Ocado won’t see positive free cash flow from its licensing until at least 2022. Meanwhile, Ocado’s heavy investment has led to ballooning losses—worsened by a catastrophic fire that gutted one of its CFCs in early 2019.

One looming question is whether the coronavirus could thwart Ocado’s expansion. When the pandemic first struck, Ocado encountered trouble obtaining a key part for its robots, because it was made in Wuhan, China, the epicenter of the outbreak. (The company has since found an alternate supplier.) Ocado mostly hires local engineering teams, so travel bans have had little impact on its plans. And grocery and construction workers have been classified as essential in most places, enabling work to continue. Duncan Tatton-Brown, Ocado’s CFO, told reporters in March that the company is sticking to its guidance on the completion of its international CFCs. Still, he acknowledged that if restrictions on movement stayed in place for many months, the construction timeline would suffer.

At the same time, Ocado doesn’t expect all of its pandemic-driven revenue boost to last. Much of its sales bump came from customers buying dry goods and other nonperishables; the company predicts that demand for many of these items will fall below normal levels in the second half of the year, as customers work through their stockpiles. As a result, the company has not raised its full-year sales and earnings forecasts.

Even before the coronavirus, some observers were doubtful that Ocado’s grocer partnerships would pay off. Christopher Mandeville, a food retail and distribution analyst at research firm Jefferies, has criticized the Kroger deal in particular. Other than in a few major cities, he says, population density in the U.S. isn’t high enough to support Ocado’s CFC model. If anything, the pandemic has shown that “store pick” may be a more resilient business model: Stores can staff up to fulfill online orders if a crisis prompts a surge in demand, whereas automated CFCs, designed to operate close to capacity most of the time, aren’t as flexible.

Such doubts haven’t stopped Ocado from raising capital. It issued a $187 million share offering in 2018. In February 2019, it sold 50% of its British e-commerce operation to U.K. retailer Marks & Spencer. The sale simplified Ocado’s proposition to investors, positioning it as more of a pure-play tech platform, while raising $982 million. (Marks & Spencer will also become Ocado’s new wholesale food partner; Ocado’s long-term supply contract with Waitrose comes to an end in late 2020.) Ocado also sold $655 million of convertible bonds in December.

The frequency of Ocado’s fundraising has made some analysts uneasy. But Steiner, the CEO, says the efforts are a sign of strength, not weakness. “The only reason to do a capital raise is because you think you are going to do more [business],” he says. And in February, the world’s most prominent investor offered an indirect vote of confidence in Ocado’s strategy: Warren Buffett’s Berkshire Hathaway disclosed in government filings that it had spent almost $550 million to buy a 2.3% stake in Kroger.

While Ocado breaks ground on CFCs around the world, Clarke, the chief technology officer, is peering around the next technological bend. He has experimented with new robots, including ones with human-like hands that could someday eliminate the need for human pickers altogether. He’s also helped research ones that could grasp delicate items, like fruit, without damaging them. And Ocado has trialed a robot called SecondHands, which travels on wheels but has a humanoid torso, head and arms, to help carry out repairs to the hive and conveyors. Eventually, Clarke says, “the goal is to move to an entirely dark facility”—that is, a CFC with almost no people.

Robots aren’t the only topic on Clarke’s mind. Ocado has made multiple investments in “vertical” farming—indoor experiments in sustainable agriculture. It bought a majority stake in Jones Foods, Europe’s largest vertical farming company, and has put money into a joint venture called Infinite Acres along with a U.S. startup called 80 Acres Urban Agriculture and Dutch company Priva Holdings. Another investment is Karakuri, a British startup creating automated kitchens that can prepare restaurant-style meals for delivery. Clarke says Ocado envisions building an “integrated food machine.” By combining vertical farming, food prep, and delivery in one facility, he explains, “you might be able to go from plant to kitchen table in two hours or less.”

Steiner and Clarke have also begun looking beyond food altogether in search of profitable business lines. Ocado’s expertise in logistics, A.I., robotics, and simulation could be deployed to tackle automated car parks, parcel sorting, baggage handling, rail freight, container ports and modular, configurable buildings. Clarke says the company has filed patents in a number of these areas. Ocado has already created simulations of a car-parking system, Clarke says, and has begun exploring scaled-up versions of its robots for handling freight far heavier than a crate of bananas.

It all may sound like a stretch for a company whose core grocery business is still fighting to prove its staying power. But for those who wonder why Ocado would want to expand into parking or port operations, Steiner has a ready answer: What if Amazon had simply stopped with books? 

A version of this story appears in the May 2020 issue of Fortune with the headline “The grocery robots on the pandemic front lines.”