Amazon has built a cloud computing business that has swelled to many times the size of its competitors. Now it is taking some time to test new ideas, like transferring trillions of bytes of data in a shipping container and plugging off-beat computer chips into its data centers.

On Wednesday, Amazon Web Services, the company’s cloud computing unit, said that it planned to offer a new service powered by field-programmable gate arrays, or FPGAs, in the latest sign that cloud companies are increasingly lured to the flexibility of configurable chips.

At an event in Las Vegas, the company said that it was using chips from Xilinx, one of the biggest makers of FPGAs. The chips can be quickly reprogrammed for various specialized jobs, ranging from searching through financial data to analyzing encryption algorithms. The Xilinx chips are built with circuits measuring just 16 nanometers long.

The FPGA’s promise is that it’s a jack and a master of all trades. Engineers can program them to handle tricky computing jobs like speech and image recognition better than general-purpose computers sold by Intel and the graphics chips devised by Nvidia. They also contrast with application-specific chips – like the machine learning chips that Google devised for its data centers – which put all their energy into a single task.

“We are giving you the ability to design your own logic, simulate and verify it using cloud-based tools,” wrote Jeff Barr, the chief evangelist for Amazon Web Services, in a blog post. It is “a business model that is more akin to that used for every other type of software,” he added.

Programming an FPGA is still extremely tricky, though. It can take months to plot the routes that electrical signals take over the chip – a fact that has slowed the FPGA’s shift into the mainstream market. But over the last year, the technology’s future has brightened and some big cloud companies are diving into it.

For instance, FPGAs are the keystone in Microsoft’s Project Catapult, an initiative to improve how fast its cloud computing servers finish complicated tasks like machine learning. In October, China’s search engine company Baidu said that it was using Xilinx chips to power artificial intelligence services. Nimbix, a supercomputing cloud company, recently said it would offer more powerful services based on Xilinx chips.

​Meanwhile, Xilinx has worked to distinguish its technology from competitors like Intel, which bought Altera last year for $16.7 billion. Last month, the chipmaker announced its Reconfigurable Acceleration Stack, which includes libraries, frameworks, and developer boards for working with FGPAs. The chipmaker says that the tools will vastly improve an FPGA’s computer efficiency, making them better at handling machine learning programs.