83
Bringing AI to the device: the US industry has set its sights on the burgeoning edge-AI market
as illustrated with recent mergers.
Notable among emerging trends in AI based on different
technologies are neuromorphic and quantum computing.
The biggest demand for semiconductors comes from outside the EU. In particular China
represents more than 50% of today’s semiconductor market and as such, Chinese demand
significantly supports and drives the growth of the semiconductor industry worldwide. The
Covid-19 outbreak has sped up the digital transformation and led to short-term shifts in
the demand across sectors. On the one hand, the demand for cloud/data centre services,
“working from home products” and medical devices has increased, on the other, lock-downs and
loss of income reduced the demand for cars, mobile phones and consumer electronics.
The development and fabrication of chips has been increasingly subject to massive
subsidies. In order to reduce its dependence on imports, notably from chips made in Taiwan, and
more generally from US-developed technology which can be subject to unilateral US export
control restrictions, China has set itself the ambition of reaching 70% autonomy in chip-making
by 2025 with the revenues of its home-grown chip industry growing to USD 305 billion by 2030.
To this end it has allocated USD 150 billion over 10 years under the “Made in China 2025” plan
to build up research excellence and manufacturing capacity. Such measures have served to
support local industrial ecosystems. In the wake of the COVID-19 crisis, the US has responded
to China’s bid for independence with a USD 25 billion plan over 5 years to support investments
in R&D and manufacturing, to reduce its dependencies on other parts of the globe, notably
Taiwan.
Meanwhile, the South Korean government announced a plan to spend KRW 1 trillion
(USD 869 million) to develop AI chips through 2029.
The ICT ministry plans to put the next-
generation chip on the Korean New Deal project list for KRW 20 trillion (USD 17,38 billion)
Intel, leader in CPUs, acquired Altera (USD 17B) for their FPGAs and are now introducing their discrete GPUs,
such as Xe, also for AI in data centres.
Nvidia, leading in programmable GPUs and AI, is planning to acquire ARM (USD 40B). GPUs, initially
conceived to process images, are now being widely used for AI acceleration to speed up processes such as neural
network and machine learning tasks. ARM has the IP needed to excel in CPUs (and low-power AI at the edge).
AMD, the challenger in both CPUs and GPUs, has acquired Xilinx (EUR 35B). The acquisition will reinforce its
position as a credible alternative for the datacentre/AI market. Xilinx also offers SmartNIC for network
acceleration, which for AMD can be another asset for data centres in view of Intel’s acquisition of Infiniband and
Nvidia’s acquisition of Mellanox. Intel acquired Altera end of 2015; the announcements by Nvidia and by AMD
were in 2020.
Worth noting is the fundamental difference between ARM and Intel (i.e. X86) approaches to CPUs/processing
units and hence their different application markets. The former focuses on low-power consumption and the latter
on advanced performance, which in turns leads to more power consumption and higher use of silicon space. The
low-power approach of ARM has allowed ARM to offer the best balance of performance and energy efficiency for
mobile devices including chips for cell phones and laptops.
Accenture (2020) Semiconductor Companies: Business Resilience in the Wake of COVID-19 A guide to the
disruptive impacts & practical actions for semiconductor companies to take, retrieved from
https://www.accenture.com/_acnmedia/PDF-126/Accenture-High-Tech-Covid-19-SEMICONDUCTOR-Final.pdf
In fact the plan earmarks 12 B$ to attract TSMC to build a 5nm fabrication facility in Arizona with construction
planned for 2021 and production for 2024. At the invitation of Japanese government TSMC is planning to set up a
joint advanced packaging and testing plant in Japan. https://focustaiwan.tw/business/202101050014
https://techcrunch.com/2020/10/13/south-korea-pushes-for-ai-semiconductors-as-global-demand-grows/
https://en.yna.co.kr/view/AEN20201012007500320