Royalties When Our Partners Ship Chips That Contain ARM IP Highly Profitable and Cash Generative

Making progress vs strategy Ian Thornton, Head of Investor Relations

Arm is a subsidiary of

Arm refresher

H1 update – Increasing revenues and investments

Progress vs strategy

• Arm in servers

• AI at the edge

• Autonomous vehicles

33 ©© Arm 2017 2017 Arm Limited Chip design – then and now 1961 Today

Four transistors Two billion transistors One engineer Thousands of engineers

Main processor for running the operating system, applications and user interface

Graphics processor for generating images

Accelerators for frequently-used compute workloads, e.g. image processing, encryption, vision

Radio controllers for mobile, wifi, Bluetooth, GPS

Hardware controllers for the display, memory, image sensors, power supply, etc

Input/Output interfaces for USB, Ethernet, etc

ARM develops intellectual property (IP) blocks which are used in silicon chips Our partners combine ARM IP with their own IP to create complete chip designs We earn license fees when we deliver ARM IP to our partners and royalties when our partners ship chips that contain ARM IP Highly profitable and cash generative

mbed Cloud

Artificial Intelligence in Autonomous Augmented Hyperscale cloud every device machines reality and connectivity

Security and Privacy

Arm develops technology that is licensed to semiconductor companies Arm receives an upfront license fee and a royalty on every chip that contains its technology Business Development

1) Arm licenses 2) Partners develop technology to chips and ship chip Partners them to OEMs Technology SemiCo Arm License Fee Partner

1010 ©© Arm 2017 2017 Arm Limited Q2 Licensing: 26 is the low-end of the normal range

Average Historic licensing = 20 60

Q1 50 Q3 Q2 Q3

Q3 Q2 Q1 11 40 Q1 Q4 Average = 8 Q2

30 Q4 7 Average = 5 Q4 Average Q3 Q2 = 4 Q4 Q1 Q1 4 4 Q2 Average 20 = 1 Year 13 14 15 16 17 0 Classic Cortex-A Cortex-R Cortex-M Mali

DesignStart Pro DesignStart Eval Launched end June 2017 Launched end June 2017 License-fee free; royalty bearing No fees. Evaluation only. Optimised for FPGA

Average Historic licensing = 20 60

Q1 50 Q3 Q2 Q3

Q3 Q2 Q1 40 Q1 Q4 Average = 8 Q2

30 Q4 7 Average = 5 Q4 Average Q3 = 4 Q4 Q1 Q1 4 4 Q2 Average 20 = 1 Year 13 14 15 16 17 0 Classic Cortex-A Cortex-R Cortex-M Mali

1515 ©© Arm 2017 2017 Arm Limited Industry growth vs Arm growth (for Q2 2017) $190bn $80bn $0.5bn Overall Relevant Arm Semiconductor Semiconductor Royalty Industry Industry Revenue +29% +21% +19% +16% +11% +9%

Value Volume Value Volume Value Volume ($) # chips ($) # chips ($) # chips Strong growth in memory Strong growth in MCUs; Arm gaining share in embedded and graphics for data centre slower growth in apps Royalty rate increasing for mobile 16 © 2017 Arm Limited Is declining average royalty revenue per chip a concern? Average royalty per chip 10.0c

8.0c

6.0c

4.0c

2.0c

0.0c 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 Drivers of average • Chip related: Average sales price of the chip; Arm content per chip • Arm related: Average royalty rate per processor; increased content multipliers Second • Customer mix related: Which customers are winning share • Market mix related: Which markets are growing fastest Biggest impact on average over the long term

8.0c

6.0c

4.0c

2.0c

0.0c 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017

Cortex-M Cortex-A 8.0bn 8.0bn

7.0bn 7.0bn

6.0bn 6.0bn

5.0bn 5.0bn

4.0bn 4.0bn

3.0bn 3.0bn

2.0bn 2.0bn

1.0bn 1.0bn

1919 ©© Arm 2017 2017 Arm Limited Investing in engineering to increase productive output

Arm recruitment is running at around +1,100 2x previous run rate +586 +608 +488 Focus on retaining quality as well as +388 +329 quantity; and on cultural and organisational integration

Will need to sustain this run rate for next 2-3 years

(£10m) £493m Future cost increases are expected to be £20m consistent with increases in headcount £41m

£49m

£25m

£66m £301m

* Share-based compensation was previously included in IFRS “other costs” H1 2016 24% increase Increased Reclass Bonus and Impact of Unwinding H1 2017 in headcount IT, facilities of share- replacement weaker bad debt Costs and other based comp scheme sterling provision Costs 21 © 2017 Arm Limited investments comp* Revenues, profits and profitability Investment strategy

H1 £1,400m only 45% Over the past 10 years Arm's Revenue Total Costs revenues grew faster than costs 40% £1,200m IFRS EBIT IFRS Margin 35% £1,000m 30% H1 2017 vs H1 2016 £800m 25% • Revenues +17% 20% £600m • Headcount +24% 15% £400m • Costs +64% 10% £200m • Profits -39% 5% • IFRS Margin * 18% vs 35% £0m 0% 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 H1 2017 Fiscal years * Excluding impact of currency fluctuations as USD/GBP went from 1.27 to 1.32 This contributed to £37m of non-cash costs due to revaluation of long-term contracts. Including these IFRS Margin would be 12%.

2323 ©© Arm 2017 2017 Arm Limited Arm in servers

Server Processor TAM ($) $25bn Standard enterprise $20bn

$15bn High-performance computing $10bn

$5bn Cloud $0bn 2016 2017 2018 2019 2020 2021 Cloud HPC Standard Server ARM focused on Cloud/HPC

Arm technology Chip Software End user systems

Beta First generation Testing Arm v8.0-A No commercial silicon Proof of concepts support Start date: 2008 2010 2012 2014 Arm technology Chip Software End user systems Arm-based chips available from Commercially Production Arm v8.4-A multiple vendors available deployments Latest status: 2017 2017 2017 2017

Arm v7-A Arm v8.0-A Arm v8.2-A Arm v8.4-A

Cortex-A15 Cortex-A72 Future Future Cortex-A9 Cortex-A57 Cortex-A75 Future Cortex-A7 Cortex-A53 Cortex-A55 Future Cortex-A5 Cortex-A35 Future Future Architecture released 2004 2011 2015 2017 Multi-core processing 64-bit computing Improved multimedia To be announced Multimedia acceleration Improved multi-core Improved multi-core New Capability Improved multimedia Improved error management Acceleration for AI/ML Acceleration for virtualisation

Performance Performance per thread

https://www.qualcomm.com/media/documents/qualcomm-centriq-2400-media-deck.

All the major open source operating systems are now commercially available for Arm-based servers

Mobile Automotive Robotics Drones

Home, surveillance IoT & analytics VR/MR Shipping & logistics

Security Privacy Latency Bandwidth Power and Cost

Need multiple types of processors to handle this DSP DSP DSP

Fixed function works when requirements known in advance, algorithm is well understood, and CPU big CPU big FPGA high performance needed CPU CPU DSP Little Little Programmable cores are essential GPU GPU Acc

Machine Learning applications

Domain-specific libraries + ML frameworks: TensorFlow, Caffe, MXNet, etc.

Tools and libraries from Arm

CPUs + GPUs + ISPs + CV engines DSP FPGA ACC

Redesigned memory Advanced compute New single cluster design Greater flexibility sub-system capabilities

– Audi at CES Asia

F-35 24

Windows 10 60

US Army future combat system 63

Facebook 65

Luxury car in 2010 (IVI) 100

Luxury car in 2016 (+ADAS) 150

Luxury car in 2020 (L3 autonomy) 300~

All Google services 2000 Million lines of code

Cockpit Connected Gateways ~50,000 DMIPS ~20,000 DMIPS Audio Visual, Maps, Traffic, Toll LTE 5G, WiFi, Bluetooth High-end smartphone payment, Google services connecting to CAN FD, LIN, Flexray, Body Electronics 30,000-50,000 DMIPS Rear entertainment, Voice recognition, Ethernet <10,000 DMIPS Main applications processor, Gesture control, Cluster and HUD HVAC, Lighting, Doors, Electric WiFi, modem, sensors, etc. seat, Windows, Mirrors, Cameras, Seat belt, Air bag, BCM

Semi Autonomous ~350,000 DMIPS e-Powertrain Level 3 autonomy, Radar / image processing, Collision avoidance, Pre- ~15,000 DMIPS Main Motor control, crash, Cruise control, Lane departure, Transmission, Engine control, Parking Chassis ~15,000 DMIPS Generator/E-water pump EPS, ABS/EBS, Active VDC, EPB Battery management

Advanced HIGHLY AUTOMATED AUTONOMOUS • Start of fully automated • Several control functions • All-round collision avoidance driving • Collision Avoidance Steering • Limited automated driving • High speed all-round collision (Low speed) • Ride sharing avoidance • Advanced camera systems • Camera systems with 4k • Car sharing • CAN FD (10Mbps) • Ethernet bus (1Gbps) • Connected vehicle to vehicle • Sensor fusion • Interactive

Relative to 2016 20X performance 40-50X performance 100X performance Vehicles 10X Data rate 100X Data rate 400x Data rate

Scalable processing solutions Powered Powered Powered Powered Powered Powered SoC SoC SoC SoC SoC SoC

New Audi A8 44 © 2017 Arm Limited zFAS using NVIDIA Tegra K1 - based on Quad core Arm Cortex-A15 *As a production car (autonomy level 3) 44 Denso licenses first ever Arm processor (Cortex-R52) Targeting advanced driver assistance and safety critical systems

Cortex-R52: Highest performance real- Driver assistance time processor

First processor with embedded virtualisation to simplify software consolidation

Designed for functional safety systems Safety critical