Google and Microsoft: The Race For AI Chips

Management Technology

The Race For AI Chips

The semiconductor industry is doing well, with sales of memory and chips growing in 2017. But are current manufacturers slow to adapt to fast developing new technologies like artificial intelligence, virtual reality, and the Internet of Things? Computer chips are still not powerful enough to process AI algorithms and VR. IoT requires more processor power to manage vast networks of connected devices. Demand will find satisfaction as new players enter the semiconductor industry, some of them known for achievements in other fields.

Google and Microsoft join battle

Google and Microsoft are both developing their own chips to tackle the challenges of AI and VR. Apple, too, is developing proprietary chips to power mobile devices and is probably working on a more advanced AI-chip technology.

Harry Shum, in charge of Microsoft’s new AI and Research Group, recently unveiled a new chip created for the HoloLens augmented reality goggles. The chip is designed to run deep learning software able to take advantage of recent developments in speech and image recognition. Microsoft state that since they were unable to power their VR device with any existing processor, they had to research and create one in house.

Google announced their deep learning chip in late 2016. The tensor processing unit or TPU should improve deep learning processes in the company’s cloud. Demand for technologies like speech and image recognition is skyrocketing and Google decided to invest into the new TPU instead of into 15 new datacentres. TPU technology is now available to Google’s cloud computing clients, turning the company into a major chipmaker. Meanwhile, Microsoft plans to offer its chip to clients of its cloud as early as next year.

Traditional chipmakers under fire

Established semiconductor manufacturers like Intel and Nvidia are set to witness growing pressure from chipmakers developing AI chips and processors designed specifically for VR and IoT applications. Though trailing established manufacturers in some specific fields, behemoths like Google, Apple, and Microsoft have vast resources to create their own chips. There are other rising stars, however. Ex-Google engineers who were part of the secretive unit that developed TPU have formed their own venture, Groq, to work on a new deep learning chip. Other startups, from stealth mode to ones that have a few funding rounds behind them, are also working in the field.

Worldwide Semiconductor Capital Spending and Equipment Spending Forecast, 2015-2020 (Millions of Dollars)
2016 2017 2018 2019 2020
Semiconductor Capital Spending ($M)  67,994.0  69,936.6  73,613.5  78,355.6  75,799.3
Growth (%) 5.1 2.9 5.3 6.4 -3.3
Wafer-Level Manufacturing Equipment ($M) 35,864.40 38,005.40 38,488.70 41,779.70 39,827.00
Growth (%) 7.9 6 1.3 8.6 -4.7
Wafer Fab Equipment ($M)  34,033.2  35,978.6  36,241.1  39,272.8  37,250.4
Growth (%) 8.1 5.7 0.7 8.4 -5.1
Wafer-Level Packaging and Assembly Equipment ($M) 1,831.20 2,026.80 2,247.60 2,506.90 2,567.70
Growth (%) 3.9 10.7 10.9 11.5 2.8
Source: Gartner (January 2017)

While those startups provide cloud companies, data centres, and other innovative business with brand new solutions, companies like Intel and Nvidia are trying to retain their market shares by selling what they already have. In developing deep learning capabilities, they appear to bet on extending current abilities incrementally, rather than starting from zero. Both companies, and Intel especially, are also acquiring innovative startups to boost AI development. Google and Microsoft, incidentally, are also aggressively buying innovative tech startups.

Though sales are flourishing, chip making is experiencing a flux, with established manufacturers attempting to slow adoption of brand new technologies and a good number of startups aiming for major breakthroughs. Players like Google and Microsoft, positioned somewhere in between, are gradually accumulating the technology and experience that will allow them to become mainstream deep learning chipmakers. Apple is also in the race, though their secretiveness obscures the real thrust of their AI chip research. The clever money says that AI, VR, and deep learning technologies are set to revamp the industry in a major way sooner rather than later.

By Kiril V. Kirilov

Christian Buckley

The Evolution of SharePoint Customization

When I started working with SharePoint back in 2005, deploying WSS 2.0 followed and then SharePoint Portal Server 2003 for a large client, the concept of "customization" as we define it today was not really ...
Anita Raj

A Winning Data Strategy Series Part 2: Data, an Asset, or a Liability?

Data, an Asset, or a Liability? This is the second piece of a 5-part series on plugging the obvious but overlooked gaps in achieving digital success through a refined data strategy. You can read the ...
Rick Braddy

The Secrets to Achieving Cloud File Storage Performance Goals

Storage Performance with Cost Reduction By 2025, according to Gartner, 80 percent of enterprises will shut down their traditional data centers. As of 2019, 10 percent have already shifted their data centers and storage to ...
Virtana

Episode 8: Managing Cloud Strategy During the Chaos of 2020, Plus an Outlook for 2021

An Interview with Kash Shaikh, CEO of Virtana Companies are wrestling with the idea of moving to the cloud, staying on-prem or finding a hybrid solution. Kash Shaikh, the new CEO of Virtana, looks at ...
Sangeeta Chhabra

Leverage DaaS To Solve Challenges of Remote Work

Solve Challenges of Remote Work In the past one year of Coronavirus (COVID-19), we have seen it all: An ailing world economy, dramatic changes in the political structure of many countries, and, most prominently, a ...
Johan

Why the digital infrastructure is a matter of national interest!

Digital Infrastructure National Interest When the Internet was born, it promised a form of democracy and guarantee that everybody could be part and setup their company to contribute and make the Internet better. Today - ...