Dusting off the crystal ball
2020 is set to be the most important year yet when it comes to the evolution of the nascent AI accelerator chip industry. Here are a few ideas about what we might see happening this year, in no particular order…
The acquisition-fest will begin
Will Intel’s acquisition of Israeli AI accelerator startup Habana Labs last month kick-start a surge of acquisitions? This market is absolutely teeming with chip startups, many of whom are reaching a level of maturity where they are revealing their architectures and starting to produce measurable results. As established semiconductor companies start to appreciate the importance of the AI accelerators, and the range of vertical markets AI will encroach on, will some of them look to jump-start their strategies with acquisitions? I suspect they will.
Some companies won’t make it
With dozens of startups at the stage where first products are being marketed and results are being unveiled, the opposite effect also applies. I spoke with Geoff Tate, CEO of Flex Logix, recently and he quoted Warren Buffett: “When the tide goes out, you can see who’s been swimming naked.”
Not all the startups we see in the market today will be successful. 2020 will likely start to separate the winners from the losers, beginning a serious culling of the herd as the hype begins to deflate and commercial realities set in.
Facebook will launch its own AI chip
It was widely noted back in spring 2018 that Facebook advertised for chip designers, and widely speculated that the company was working on its own AI accelerator. By my estimation, we should expect to see the results of that project in the second half of 2020.
Up to now, Facebook is one of the only hyperscalers not to have its own custom AI accelerator silicon for the data center; Google, Amazon, Alibaba and Baidu all have in-house designs (Microsoft is another notable exception). Could a Facebook AI chip be destined for its data centers?
Or could it be for something else entirely? In his keynote speech at NeurIPS last month, Facebook chief AI scientist Yann LeCun showed slides claiming “of course Facebook is working on its own [deep learning] hardware for [augmented reality] glasses”. LeCun described AR glasses as a “killer app” for embedded deep learning hardware. Based on these comments, I believe it’s more likely we’ll see a Facebook chip for AI at the edge this year.
Nvidia’s grip on the industry will weaken
As a natural consequence of other technologies becoming more mature, I believe that Nvidia’s grip on the AI accelerator market will begin to weaken. Don’t get me wrong, they will remain a force to be reckoned with, but competition is heating up rapidly. For example, Intel has strengthened its data center offering with the acquisition of Habana Labs and has a ready-made route to market with its existing data center customers. Aside from custom chips, GPUs also face strong competition from FPGAs in markets such as automotive and other edge applications.
Battle of the global superpowers will continue
I think we will continue to see and hear a lot of discussion about who is winning the global race for leadership in the AI chip market. The Chinese government has invested massive sums in various Chinese AI startups and is also a major customer for a lot of them. China’s AI chip industry is reportedly flourishing as a result. Meanwhile, Silicon Valley has its fair share of companies pursuing this market, both new and established. Whether it’s the US or China who dominates 2020 (Europe does not look like a contender at this point), don’t expect this battle to be resolved any time soon.