Intel, pivoting away from the personal computer business it pioneered, has purchased Habana Labs, a maker of advanced microchips, to boost its share of the artificial intelligence market.
Faced with declining sales of chips for PCs and laptops, Intel paid $2 billion for the Israel-based HabanaOpens a new window , which specializes in making microprocessors that excel at carrying out AI tasks in the cloud, such as recognizing images and training machines to speak.
Large cloud providers like Amazon Web Services and Microsoft Azure are battling each other to offer AI services and the technology requires super-fast microprocessors. Intel is competing with rivals such as Nvidia to make the fast, high-capacity chips for cloud data centers.
The future of computing
At stake is the shape of compute and storage over the coming decade. Entrepreneurs could be held back from creating successful businesses if they lack the infrastructure to use AI to hone their offers. And in this environment, artificial intelligence is already becoming a competitive advantage.
Technology companies realize they will fail if they lack the ability to recommend products, refine target audiences, run voice assistants, analyze big data and create other services that their customers are coming to expect.
The cloud providers, meanwhile, are racing to offer AI-as-a-service to create the new products they need from cloud-based systems. AI-as-a-service will save them from building their own AI computing in-house, saving money and potentially allowing access to more sophisticated technology than they could build themselves.
Amazon is creating for its Amazon Web Services subsidiary an AI chip called the AWS Inferentia. Google is building the TPU chip to power artificial intelligence apps in the Google Cloud and Microsoft is offering its Azure customers access to AI chipsOpens a new window made by Graphcore.
Intel, the world’s biggest computer chip maker, could lose out without a strong product in this market.
In 2016, it boughtOpens a new window the AI chip specialist Nervana and launched the NNP-I chip. The chip specializes in making inferences about data – one of the key functions of AI. Another recently-launched Nervana chip is the NNP-T, which specializes in training algorithms about data patterns.
AI skeptics plentiful
Many believe Habana has outpaced Intel in creating AI chips for data centers. In January 2018, the start-up launched the Goya inference chip, and in spring this year the Gaudi training processor. The new products pose the question of whether the acquisition will allow Intel to axe the NNP chips because they’re similar to Habana’s offerings.
Not in doubt is the opportunity that Intel sees in artificial intelligence technology. The company expectsOpens a new window the silicon AI market to be worth $25 billion by 2024 and AI chips to account for $3.5 billion of its revenues this year.
However, some analysts remain skepticalOpens a new window about cloud AI. They argue that many companies will be reluctant to upload into the cloud the huge amounts of data required for AI processing.
Their doubts surfaced in part because of security concerns and privacy issues connected with moving around vast amounts of personal data on the Internet. And there is the question of bandwidth: Huge amounts are needed to upload those vast quantities of data. The capabilities of cloud AI are fairly limited at the moment, some analysts point out, restricted to categorizing images and understanding speech.
The more advanced abilities of the technology, such as natural language processing and unsupervised learning in which artificial intelligence uncovers patterns in data without being prompted, could be some way off. For those and other reasons, some think that the AI cloud will be a temporary fad.
Intel, though, believes AI chips for the cloud will be a major money maker. The company wants to be at the heart of the cloud AI boom in the 2020s just as it powered the personal computing revolution that began in the 1970s.