facebook rss twitter

45 startups now working on AI computer chips

by Mark Tyson on 15 January 2018, 13:31


Quick Link: HEXUS.net/qadput

Add to My Vault: x

There's a new frontier in computer processor development giving the minnows a chance to compete with the likes of Intel, Nvidia and Qualcomm. According to an article this weekend in the New York Times this computer processor market segment targets artificial intelligence. NYT contributor Cade Metz reports that there are now at least 45 start-ups working on AI chips and at least five of them have raised more than $100 million from investors.

For many years processor development has been out of reach for start-ups, as established players built up barriers to competition using both business (partnerships, vertical integration, massive R&D budgets) and legal (copyright etc) frameworks. However the AI processor niche players often tackle processor designs from a completely different angle. As it is such a new field, innovative left-field thinking can have a big impact.

Remember, last week Intel revealed more info about its cutting edge chip designs. While x86 is its cash cow Intel is researching many other data processing ideas such as quantum computing, neuromorphic computing (Loihi), and neural network processors (Nervana). In the NYT article it relates how "Intel scrambled," to buy Nervana in 2016. It paid over $400 million for this 48-person deep-learning startup.

In more recent months raising money for AI computing has become much easier. Investors look at these startup AI companies not only as firms that could actually end up shipping an 'AI product' but as one that will be gobbled up for millions by the likes of Intel, Google, Apple, Microsoft, Facebook etc .

Meanwhile the start-ups working on AI technology aren't solely looking at 'CPU' architectures. With radical new architectures there is a need for new types of memory and interconnects, as well as software.

Though not acquisition related, back in May 2017, Google showed what could be achieved using its transformational Cloud TPU design (an ASIC optimised for machine learning) and software. This shows that in an innovation friendly company you don't have to rely on the acquisition of start-ups to create breakthroughs.

HEXUS Forums :: 1 Comment

Login with Forum Account

Don't have an account? Register today!
Today the EETimes says that AI is Fuelling Next-Gen HBM Demand