In summer 2016, chip giant Intel (INTC 1.74%) announced it would acquire a company called Nervana to accelerate its efforts in the small-but-fast-growing market for artificial intelligence and machine learning processors.
With Nervana's technology, Intel hopes to more effectively compete against technologies like NVIDIA's graphics processing units (GPUs), which are more efficient than Intel's general purpose central processing units (CPUs) for specific machine learning tasks.
Image source: Intel.
On Intel's most recent earnings conference call, CEO Brian Krzanich offered some insight into when investors should expect sales of Nervana chips to impact its financial results. Let's dig into his comments.
Shipping soon, real revenue late next year
Krzanich said during the company's third-quarter earnings call that the company intends to start shipping its first Nervana product, known as Lake Crest, by the end of this year. Don't expect those shipments to yield significant revenue, though. Krzanich indicated the chips that'll be going out later this quarter will be "handed out as test silicon." The reality is that the Nervana chips represent an entirely new platform and potential customers are going to need time to evaluate the capabilities of the chip and figure out how to make sure that their software can take full advantage of those capabilities.
"I expect the first instantiations [of the Nervana technology] to be more people figuring out how to use it," Krzanich said. "As we go through next year and move on beyond that and get the second generation [Nervana chips] and beyond, [revenue] will grow, we believe, quite considerably."
Intel's broader strategy
One thing that Intel seems to be doing right is moving away from the mentality that every workload will eventually run best on the company's general-purpose processors. In the past, Intel executives would often dismiss specialized accelerator technologies and claim that the company's goal is to advance the performance and capabilities of its general-purpose processors by so much that the value proposition of such accelerators would be diminished.
The reality is that there are increasingly popular workloads that can benefit greatly from the speedups that specialized chips can deliver. Moreover, those specialized chips can get better over time with improvements to the chip designs as well as with improvements in chip manufacturing technology, so general purpose processors are constantly chasing a moving target.
It makes sense, then, for Intel to try to identify those workloads and then try to build specialized chip technologies to service those workloads. Specialized chip technologies generally require additional software ecosystem support for their full potential to be realized, so in addition to building compelling hardware, Intel is going to need to work diligently to make it as easy as possible for software developers to invest in Intel's Nervana platform.
Intel also has a lot of flexibility in how it can deliver these products to market. For example, the first versions of the Nervana processors are stand-alone chips, but Intel has talked about building products that integrate its general-purpose CPU technology with its Nervana technology onto a single package.
By offering such integrated products, Intel won't necessarily see a loss of a CPU sale in exchange for a Nervana chip sale. If the company can offer products that integrate both technologies in a way that competitors can't, then Intel could enjoy high data center chip average selling prices. Such a boost in chip average selling prices would ultimately boost the company's revenue and profits, which would be a clear win for Intel's business and its stockholders.
