Intel’s Strategy for AI Chips is Smart But Still Lags Nvidia, Analysts Say
Intel hosted a conference on Thursday to highlight its latest efforts to sell more microchips used to meet the booming demand for artificial intelligence, machine learning, and related disciplines. The overall strategy made sense to analysts who follow the company, but some of them were concerned that Intel has a ways to go to catch up to competitor Nvidia.
The conference agenda included the announcement by Intel of a strategic partnership with Google across the search giant’s many AI initiatives. Intel is creating versions of its chips optimized for Google’s TensorFlow software that runs machine learning programs and neural networks, for example. Intel also explained how it would integrate a batch of acquisitions it has made in the AI area, such as using machine learning technology acquired with Nervana Systems in August for new chips called Lake Crest and Knights Crest.
Intel needs to find new areas of demand like AI for its chips as its main business of selling chips for PCs continues to shrink. The company largely missed out on the smartphone revolution, so CEO Brian Krzanich is looking to expand Intel’s dominance in chips for data center servers into new markets. Intel’s data center group brought in $4.5 billion of sales last quarter, though only a small portion went to computers used for AI.
Intel’s server chips for machine learning tasks, like translating text into different languages and recognizing objects in pictures, are in fierce competition with chips from companies that sell graphics processing chips, or GPUs, for the same tasks, particularly Nvidia. In its latest quarter, Nvidia’s sales increased 54% year-over-year to $2 billion, with about 12% of the sales going to data centers largely for machine learning tasks.
Get Data Sheet, Fortune’s technology newsletter.
Intel executives didn’t mention Nvidia by name, but they set a lofty goal of achieving 100 times the performance of graphics chips for AI tasks by 2020.
Still, some analysts said Intel lags behind Nvidia (NVDA) in the machine learning market. The company “is clearly playing catch-up to Nvidia, which has championed AI for years,” Barclays analyst Blayne Curtis wrote in a report on Friday. But Curtis said Intel’s improved focus and broad strategy should “bolster longer-term (data center group) growth.”
Bernstein Research analyst Stacy Rasgon echoed the idea that Intel (INTC) was lagging, noting that the company’s presentations only compared new chips’ improved performance to older Intel chips, not competitors’ products.
And by helping make cloud computing so much more powerful, Intel also risks seeing more of its ordinary corporate clients close their own Intel-powered data centers and shift to centers run by Amazon (AMZN) or Google (GOOGL). “The broader question of whether AI applications and workloads will be additive or cannibalistic to the company’s (total addressable market) remains unanswered,” Rasgon wrote.
So far, most of Intel’s learning chip sales have been to the largest data center operators, such as cloud computing giants Amazon, Google and Microsoft. But the chip maker argued that sooner or later, those ordinary companies would be interested in using machine learning chips directly to improve their operations, which would stimulate more demand. UBS analyst Stephen Chin agrees that it is an “untapped opportunity,” but he says the timing is still “certain.”
The market is growing so fast that there may be enough business to reward Intel, Nvidia and Advanced Micro Devices (AMD), another chip maker that sells graphics processing units for AI tasks, Mizuho Securities analyst Vijay Rakesh noted.
“While only 0.1% of all servers are dedicated to AI, the AI market will grow 12x in the next four years,” he wrote on Friday, making it a “positive” for all three players.