Meta (Facebook) is collaborating with Intel to design and develop Intel's first dedicated artificial intelligence chip for datacenter workloads.

Intel, the chip giant, has emerged as a latecomer to the artificial intelligence market, where Nvidia, maker of graphics chips, has become the key hardware platform for recent developments in deep learning—an important branch of AI that involves training computers using huge piles of data. However, Intel is preparing to fight back later this year with its first chip dedicated to running AI operations, and the company has secured a major partner for this launch: Facebook. "We are thrilled to have Facebook in close collaboration sharing their technical insights as we bring this new generation of AI hardware to market," Intel CEO Brian Krzanich wrote in a blog post. Facebook represents an important player in AI through both research, run by deep learning pioneer Yann LeCun, and the application of the latest AI techniques across its social network.

Intel's upcoming neural network processor, or NNP—formerly codenamed Lake Crest—will be made available initially to only a limited set of partners like Facebook to gather feedback for a second version due out sometime next year. "It's the first chip product in the world specific to neural networks," said Naveen Rao, a corporate vice president of the AI unit at Intel. According to Rao, the chip will work well for both inference, which involves simply running algorithms, and training AI models—the more computationally intensive part where Nvidia currently dominates. Like Nvidia's graphics processors, which have thousands of tiny computers operating in parallel, Intel said the NNP would also perform well with computing in parallel using a number of new chip designs.

Because the chip project began when Rao founded AI startup Nervana in 2014, which Intel acquired for $400 million last year, the first chip will be manufactured by Taiwan Semiconductor Manufacturing Company, not Intel. Intel plans to take over manufacturing with the second version and make the hardware more available. Intel will release performance benchmarks for its AI chip later this year.

While Intel continues to dominate an overwhelming share of 99% of the market for central processing units in the data center, Nvidia's data center business is booming. In Nvidia's most recent financial quarter reported in August, its data center business revenue was $416 million, up 175% year-over-year, while Intel's data center business was much larger with $4.4 billion in revenue reported in July, but grew only 9% from the previous year. As a result, Nvidia's stock has nearly tripled in the past 12 months, while Intel's stock is up less than 6% over the same period.

In this new era of AI, Intel has acquired a number of companies to expand beyond its general-purpose CPU, most notably acquiring Altera, maker of field-programmable gate arrays, for $16.7 billion in 2015. The company is also working on a neuromorphic chip that attempts to function like a real brain, though that project is more experimental and years away from commercialization. "The strength of Intel is our broad portfolio," Rao claimed. "There isn't a one size fits all." Nvidia has similarly moved into AI specialization with its latest graphics chip architecture, Volta, which contains so-called Tensor Cores—specialized computers designed to run the mathematical operations used in deep learning software.