Internet platform giants are all interested in custom silicon. Impatient, and with deep pockets, they're getting more prone to going the do-it-yourself route.
Their dominance in penetrating sectors like healthcare is a good example of the general direction of the FANG group, whereby they gradually grow their presence in the sector, with a view to ultimately owning it. Earlier this month, the UK’s national health service, NHS, announced it is collaborating with Amazon to provide ‘reliable’ health information through voice-assisted technology — in other words through Alexa. It said that by 2020, half of all searches are expected to be made through voice-assisted technology.
Could this be a precursor to Amazon bidding to run the UK health service at some time in the near future? Or could it be Google? While Google has already run into trouble on data protection with its partner, the University of Chicago (where a lawsuit alleges their data handling didn’t protect data that could identify patients), the company still has traction in this sector: Google Health has grand plans as a health company, according to its new head, Dr. David Feinberg.
Taking this pattern of behavior across to semiconductors, we’ve already seen the inroads the FANGs are making to develop their own chips, such as Facebook recruiting a chip team. A lot of the chip development by the FANGs is currently driven by the need to process extremely high volumes of data in the pursuit of incorporating artificial intelligence (AI) accelerators, and the need to have their own specific architectures to improve their deep learning capabilities in silicon.
One semiconductor industry executive told us that customers are all interested in custom silicon, since customization is the only way to get the performance they need. The difference with the FANGs is the incredibly deep pockets they have and the really aggressive goals they set themselves. This is significant, in that in the past few consumer electronics companies were able to afford this.
In an interview earlier this year, Yann LeCun, VP and chief AI scientist at Facebook, said they needed more options than that provided by the de facto supplier in the industry, Nvidia. He said that current GPU vendors make assumptions about what the customer needs, but they would like to ideally have hardware based on their own specific assumptions.
The participation of the FANGs in semiconductors is likely to come from both hiring their own teams as well as acquisitions. Apple for example is reportedly currently in talks to buy Intel’s modem chip business for $1 billion. This not only serves Apple’s needs, but also helps Intel to divest itself of a business unit that has not been so successful. Previously, Apple also bought Dialog Semiconductor’s power management business for $600 million last year as well as the widely rumored acquisition of French multi touchless sensor company Fogale Sensation for €100m four years ago.
There’s no doubt that the semiconductor industry is aware of the gradual spread of FANG’s tentacles across the ecosystem of chip development. That’s why we often hear semiconductor industry executives talk about the need to move up the value chain, and the industry talk about developing domain specific architectures (John Hennessy will be doing the keynote about the rise of the latter at the AI Hardware Summit in September 2019).
NXP Semiconductors’ CTO, Lars Reger, for example has told us how new sets of customers increasingly may have domain specific knowledge for their sectors, such as smart homes or smart mobility, but not at all in semiconductors. The lack of silicon knowledge is why he said NXP started offering reference designs for different industry verticals.
This trend will only increase, where non-tech companies with no semiconductor background will be looking to implement ‘smart’ technology to improve processes or performance in their own industry. An example of a company commissioning their own chip cited by SiFive was a real estate owner looking to manage 40 million square meters of real estate with cameras and sensors, for which they hope to deploy 30 million chips, and enable a smart real estate management system.
In effect we are seeing a top-down approach of FANG companies trying to build their own teams either directly or through acquisition to develop domain expertise. This is not just limited to the FANGs — Tesla is a good example of a new wave of companies looking to develop their own chips, and no doubt there will be other traditional companies also looking to create solutions that involve some level of customized chip.
Conversely there’s a ‘bottom-up’ approach of the chip industry looking to create greater relevance by moving up the value chain. There was already a trend for semiconductor intellectual property (IP) companies to become more system-level companies, but with the new focus on internet of things (IoT), sensors with more edge processing capability, and serving many different industry verticals, the chip companies are being forced to offer solutions that are both domain specific but can be optimized for performance.
Where the FANGs meet the chip industry in the middle, that’s where we may see consolidation of the chip industry, with fewer smaller and medium sized players as they get acquired by the FANGs. The larger players will still have a role, albeit a different one, as they try and become system companies rather than just chip companies.