This was our 17th year at Mobile World Congress (MWC) and our 12th “El Clasico” Amadeus Mobile Dinner, where CEOs, investors, entrepreneurs and industry insiders speak freely and unreported. Here I have highlighted some themes from the conference and discussions over dinner.
At MWC 2015, Mobile was “The Edge of Innovation”, in 2016 “Mobile is Everything” and in 2017 Mobile is.. “The Next Element” (like oxygen – get it?). I’m not sure the GSMA marketing creatives realise how they are pushing the metaphor every year closer to a logical dead end. What can come next?
The mobile operators’ community existential crisis was exposed on centre stage this year. It took a few new faces. The Chairman of GSMA is now Sunil Bharti Mittal, founder of Bharti Airtel and the first representative from an emerging market carrier to assume the role – a sign of the times.
The top seven service providers using operator networks to reach their customers are hence worth the same as the entire mobile industry in 2015 (and this has probably gone down further since then – I will ask Sunil next time I see him on the golf course in New Delhi). The combined market cap of the top seven telecom companies is $1,000B – similar to Google and Facebook.
We also witnessed a carrier that has decided to do something serious about it. The brave Russian operator – formerly known as Vimpelcom – went from fear to actions and decided to kill its first born – comms – and “be truly free”. No doubt McKinsey et al will quickly tout this case study around the mobile carriers’ community boards.
Key themes this year included, Car as a Service, Industry 4.0 and Robotics, 5G and Machine Learning. These themes will no doubt be covered in other reports on MWC. I have chosen to focus on some interesting topics that were less visible but will have a significant impact on portfolio companies and deal flow:
- Mobile financial inclusion – a new stack
- IoT – which way will the wind blow?
- AI applied – food for thought (and vice versa)
Mobile financial inclusion – a new stack
Something truly disruptive is happening in India where the biggest economic experiment – probably since the single currency in Europe – is taking place. Nandan Nikelani, co-founder and ex-CEO of Infosys and one of the key contributors of this initiative called it a “Whatsapp moment” in Indian banking. This probably understates the disruption in the Indian financial services industry. The story is best told using a chronology:
- July 2014: The former Unique Identification Authority of India (UIDAI) Chairman, Nandan Nikelani, convinces President Modi to back the ‘Aadhaar scheme’. The UIDAI was set up in 2009 and mandated to assign a 12-digit unique identification or ‘Aadhaar number’ to all residents in India. Aadhaar was designed to avoid duplicates and fakes and, more importantly, to be authenticated online cost effectively and quickly. By September 2010, the first Aadhaar numbers had been issued but for the first five years the programme had a slow start.
- August 2014: In his Independence Day Speech, President Modi launches the Jan-Dhan Yojana (People Money Scheme) whereby, for the first time, account holders will gain access to bank accounts with no minimum balance or fees, with insurance cover and access to overdraft facilities. Mobile banking transactions will also be available using mobile phones via a National Unified USSD Platform (NUUP) adopted by all banks and operators. Each bank account is linked to an Aadhaar number.
- May 2016: The Modi government, via the National Payment Corporation of India, launches the Unified Payment Interface (UPI) on open API platform, allowing the transfer of funds from bank accounts by using account number, Aadhaar number and mobile phone number. The UPI is integrated with the NUUP.
- November 2016: President Modi announced the demonetisation of all 500 and 1,000 Rupee notes – representing 86% of the total value in bank notes in circulation in the Indian economy. (BTW, this was anticipated in one of the most prescient April Fool’s day joke in history).
- December 2016: President Modi announces the launch of the Bharat Interface for Money (BHIM) a free mobile app to transfer money between account holders (even if the receiver does not have the app) and to make e-payments.
- December 2016: Smartphone users in India reach close to 300M – close to 30% of the subscriber base. This is expected to grow to nearly 500M users in the next five years.
All of the above API-based services have come to be known as the “India Tech Stack”. Over the past two and a half years the results have been staggering. I am not sure the world has caught on to the mobile financial inclusion revolution going on in India:
- 15/12/2016: 1.06B Aadhaar numbers have been issued and almost 100% of them authenticated, representing 85% of the Indian population in 2015
- 30/12/2016: 97% of the demonetised notes have been deposited in bank accounts
- 1/2/2017: 270M Jan-Dhan bank accounts have been opened and $10B deposited in them. In one week alone in Aug 2014, 18M of these were opened (a historical record) and during the week of demonetisation, the balance on these accounts rose by $4B.
- 21/2/2017: Two months since its launch, the BHIM app has been downloaded 17M times and achieved a 4.0 rating on the Android store.
- 31/3/2017: The government launches Aadhaar Pay, an app for small merchants to receive payments from consumers’ Aadhaar-linked accounts via a biometric reader and their Aadhaar number. The biometric device is currently priced at Rs 2,000 ($30) but the cost will come down significantly as orders rise.
A large number of online financial services will be launched on top of the “India Tech Stack”. Two early applications using the stack impressed us. The first is the use of an Aadhaar- and biometric- enabled attendance system to monitor absenteeism among government employees. The second is the use of the same Aadhaar-enabled biometric authentication by Reliance Jio’s agents to recruit new subscribers and validate them quickly before issuing SIM connections. By leveraging the India Tech Stack, Jio was able to attract and validate 100M new subscribers in 170 days – quite an achievement.
The true potential of the Indian stack, as my Bangalore-based partner Bhavani explained, is that “lenders can now originate much smaller ticket loans because the process of e-KYC is seamless, transparent and fast, removing the manual intervention which meant that the loans had to be of a certain threshold size to be viable. For people familiar with India, where there is a penchant for filling in forms in triplicate (a legacy of the British Raj’s civil service), this is a new way of doing business. Just as the decision of the US government to open up GPS spawned maps and on demand services, so too the India stack could provide new and unimagined services in the realm of fintech.”
The digitisation of much of the Indian economy is going to make the transaction data of 100s of millions of consumers visible for the first time. This could form a “new moat” for fintech start-ups in India.
So far, the Aadhaar programme has cost the Indian government $1.2B. Despite the short term economic turmoil, particularly due to demonetisation, in 10 to 20 years’ time there is a good chance that we will look back at this series of events and recognise it as a piece of inspired government, application of technology and ultimately, a small investment compared to the economic benefits.
Internet of Things (IoT) – which way will the wind blow?
Two recent landmark transactions confirmed how this market is entering a “strategic hot” phase, usually characterised by an excess of hyperbole and shortage of revenues. The first is Softbank’s acquisition of ARM for $ 31.4B and the second, the more strategically significant, acquisition of Machina Research for an undisclosed sum by Gartner (rumoured to be less than $31.4B). Machina Research, a boutique market analysis firm, carved out a great niche as a specialist covering IoT. You know a market has hit the peak of the Gartner’s notorious hype cycle when Gartner itself acquires the sector specialist research firm – a moment in the development of a tech sector where profits from market research significantly exceed the same sector’s aggregate profits (my definition, not Gartner’s).
At the show, ARM launched its ‘Cordio-N’ portfolio of NarrowBand IoT (NB-IoT) technology. NB-IoT is the 3GPP standard-based (i.e. with the GSMA license of approval) solution for IoT connectivity. According to their website, “ARM Cordio-N IP provides on-chip radio connectivity solutions for Internet of Things SoCs implementing LPWAN connectivity, using NarrowBand IoT. Designed for optimal efficiency with the the ARM Cortex-M processor family, the Cordio-N IP provides everything from antenna through to L3 protocol software including all the software layers of the NB-IoT standard, PHY and RF, along with test and debug tools.” As I read this while listening to Masayoshi Son’s (Softbank’s CEO) announcement of the Cordio-N launch during his keynote, I could almost hear the CEOs of the various non-3GPP standard-based startups calling their investment bankers to kick-start a ‘strategic fund raising’, otherwise known as ‘dual track’ or, on trading floors, as “sell, sell, sell”.
The NB-IoT standard is based on a technology originally developed by local Cambridge company Neul (“cloud” in Gaelic) acquired by Huawei; probably the best $25M they have ever spent on cloud technology. I am sure that there have been a few boozy dinners in Kista and Shenzhen (probably charged to Mr. P. Jacobs’ QCOM corporate credit card where Ericsson and Huawei have held discussions about the benefit of a new PHY chip for NB-IoT. Whilst this was happening the brain that designed the technologies (Bluetooth and NB-IoT) which have wirelessly connected, and will likely connect, the largest number of devices in history may have been sailing his beautiful “Awellina of Sweden” along some wonderfully rugged and cold coasts.
Although Ericsson’s forecasts of “cellular connected IoT devices” continue to drop (from 20B to just over 1B) even faster than its share price (from 100 SEK per share two years ago to 60 SEK today) we are starting to see some volumes ramping. Applications span from connected trash bins to endangered species. Nevertheless 90% of the volume remains in energy smart meters, cars and security/safety systems. Telit and Sierra Wireless are doing well with their IoT modules in the first two market segments and have reached combined revenues of close to $1B (at $10 a module I estimate this represents 100M connected units a year). On the other hand, Gemalto, the largest manufacturer of SIM cards, seems to be establishing an early leadership position in the safety/security space thanks to its strong digital security heritage and acquisitions of Centurion Wireless and SensorLogic more than five years ago (a long gestation period in this market).
Despite the fact that probably more than 80% of connected IoT devices are non-cellular, the fast adoption of eSIMs (also known as soft SIM, a new SIM standard that allows re-programability and remote activation, giving customers the ability to switch carriers without getting a new SIM card) are a game changer according to McKinsey. The introduction of ‘network slicing’, a mechanism used by operators to support multiple ‘virtual’ networks and IoT applications behind the air interface across the fixed part of the mobile operator’s network, both backhaul and core (originally unveiled by Ericsson a few years ago), is also providing an impetus to this market. As a result, the large players in the industry are staking out their strategic positions on the IoT chessboard. The two main players are Intel and ARM.
At the moment there is no operating system (OS) for the IoT space. As most applications are vertically integrated (from microcontrollers to hardware modules and all the way to applications) the market is dominated by highly fragmented real time operating systems (RTOS).
Enter Intel – stage left. Intel decided to make a $884M ‘gift’ to IoT. In 2009, Intel-acquired Wind River Systems was the market leader in embedded systems software (most of today’s cellular IoT applications are newly connected industrial embedded systems applications of old). In 2015 Intel ‘donated’ Wind River’s highly valuable IP to the IoT community by making it open source and turning this IP into the core of Zephyr (‘zephyr’ being a light or west wind – thanks to my partner Alex for spotting the subtle reference). Intel’s intention was for Zephyr to become the open source ‘Linux of microcontrollers’ (thanks George for the term) and consolidate the currently fragmented RTOS landscape, so that there will be one OS and many IoT cloud device and data management players on top of it (Google, AWS, Cisco/Jasper, Alibaba, MSFT, etc). This way, Intel will be able to optimise its Atom processors for Zephyr and have front line access to the data fumes coming out of this OS or ‘defragmentation layer’ in technology strategists’ lingo (read ‘chokepoint’ for you and I) – helping Intel to build a data-based “new moat” in the IoT space.
Enter ARM – stage right. What Masayoshi Son forgot to mention in his keynote was that ARM has started quietly marketing to its developer community the Mbed Cloud product (“the scalable solution for IoT device management”) expected to launch this year. The product comes from ARM’s incubation business team whose job it is “to create new solutions to enable the Internet of Things”. The not yet obvious ‘join the dots’ strategy that ARM might pursue is to offer the first complete chip-to-cloud platform solution to its Cordio-N licensees (most of whom depend on ARM’s architecture having designed their product lines on ARM’s processors). Mbed Cloud could represent a ‘defragmentation layer’ (chokepoint) for the fragmented landscape of multiple cloud services providers (Google, AWS, as above).
I can only imagine the business model gymnastics that Dipesh Patel – in charge of ARM’s incubation businesses – and Rene Haas – in charge of ARM’s IP licensing business and existing customer base –had to go through to argue how this strategy could be presented consistently to ARM’s customers and make sense for both groups of ARM; the whole probably over a boozy lunch in some of the wonderful historic pubs in the Cambridgeshire country side (perhaps charged to Mr. P. Jacobs’ QCOM corporate credit card).
If successful, Mbed Cloud could become the data ‘exhaust pipe’ of the IoT value chain giving ARM (i.e. Softbank) almost perfect visibility on this market, enabling them to build a data-based “new moat” in the IoT space. I wonder how many $Bs appeared in Masayoshi Son’s mind when he saw the ‘Cordio-N + Mbed’ slide as he flicked through the sell side ARM advisors’ pitchbook on some beautiful part of the Turkish coast last summer. I cannot wait to see which way the wind will blow in the next scene in this intriguing IoT play though I now know its title: “A New Moat”.
AI applied – food for thought (and vice versa)
Every time I think of AI, an emerging image comes to mind. It is the image below courtesy of Simon Knowles, CTO of Amadeus Capital Partners-backed AI processor company Graphcore.
The picture on the left represents MSR-ResNet – an image recognition neural network trained on one of the largest AI training sets in the world (made up of 26M parameters and 16M activations). As you throw computing power at the training data set and graph the result, the image emerges looking increasingly like an MRI scan of the human brain (right).
With that in mind (….) I decided to spend some time walking the MWC halls (20k steps daily on my iPhone step counter for 4 days) to figure out who is actually making money from AI. What I found is that talk is cheap, the applications are many (from recognising objects in images to predicting crop yields) and it was hard to find anybody who is currently making serious money from it . After walking from Hall 8 to Hall 1 (at the entrance) I found one such company in this comparatively small, non-descript booth (see picture right). It almost seemed as though this company did not want to be there but had been told they had to and were given some free booth material left over from the construction of the Huawei booths (more on this later).
At the Google stand, I heard how YouTube introduced AI algorithms in their personalised search functionality back in 2012 and this has led to a 10x increase in viewership. During the same period, according to Credit Suisse, YouTube’s revenues grew from $4B to $8B. Although I’m sure not all of this revenue growth was due to the application of AI, I was told that is was a significant contributor. I had finally found some real AI money after 80k steps – about $50M a step – not bad.
At the end of MWC, on my way back to the somewhat derelict Ryanair terminal of the Barcelona Prat Airport (practically a future Amazon canteen) I could not help but think how quickly defensibility (i.e. moats) in the technology industry has changed over the past three years. A new moat is quickly emerging – proprietary, very large, training data sets to fine tune AI algorithms.
On the plane – sitting in my comfortable Ryanair emergency landing row so that I could write this – I started thinking what a drawbridge across this new moat could be. The thought reminded me of how recent research by Suzana Herculano-Houzel on “how our brain became remarkable” shows that learning to cook was an “essential requirement for (human) brains to become larger” and hence to make humans the dominant species on our planet (that is until Masayoshi Son starts spending some of his $100B pocket money on robots, in his perennial search for the singularity).
According to Suzana, approximately 2M years ago, the number of neurons in homo sapiens’ brains became capped at 40-50B because it was powered by the energetic uptake from scavenging and hunting and there were only so many hours a day our ancestors could hunt. So how did I get to the 86B neurons firing underneath my bald patch? According to Suzana’s research “the brain needed to secure more energy from the same type and quantity of foodstuffs. As from 1.5 million years ago that is just what our ancestors achieved by cooking their food.” If large training data sets are AI’s “new food”, is AI, then, limited today by the amount of energy that AI chips can turn into 1s and 0s? What could be the equivalent of “cooking”?
One of the answers could lie in the AI algorithmic equivalent of ecological rationality heuristics, the other might be found in quantum computing. I am sure my partner Hermann Hauser is already sitting in pitches about these and will soon be standing in front of his whiteboard, marker in hand, in one of his famous “Let me tell you why I think this is going to work” sessions.
Maybe one day our AI start ups will pay Data Chefs more than they are currently paying their Chief Data Officers. Food for thought – or is it the other way round?
At the Amadeus Mobile Dinner the conversation was over good food and as lively as usual (and before you ask, I paid for dinner with my own credit card – at this point Mr Jacobs’s was beyond its limit).
The topics of discussion were both varied and interesting. Some guests wondered whether the market shares of the companies at the conference were directly correlated to the size of the LED screens on their booths. On the basis of the following photographic evidence this seems to be the case – with Huawei dominating (my new friend found fixing up the stands of Chinese vendors confirmed that the cost of the Huawei stand was a cool $2M and it took 200 Chinese contract workers to put up).
Others were wondering whether Cisco is sitting on a ticking time-bomb in the form of a product recall thanks to a faulty Intel Atom C2000 processor. On a lighter note, some compared Masayashi Son’s keynote speech to some of the world domination soliloquies of 007 movies villains (cue – white-purring cat robot), while others thought he was more like Gru, the black turtle neck-wearing character in the Minions film, an apparently malevolent genius who wanted to steal the Moon only to be found wanting because of his big heart. Finally others were trading ‘Barca v. Gijon’ tickets (the match finished 6 – 1, reminiscent of the Huawei v. Ericsson score of late) for invitations to the Facebook party.
Mobile trivia of MWC 2017
Who is the individual driving the most important roadmap in technology today? His name is Greg Kroah-Hartman.
Your phone’s OS (Android), the site where you do your Christmas shopping (Amazon), the social network where you store your memories (Facebook), most of the web services you use (Google, Wikipedia, Twitter), the White House web servers, the US Department of Defence, the US Navy (and its nuclear submarines), the US Federal Aviation Administration, US Postal Service, US Federal Courts, the French Parliament, Dutch Police, CERN, the London Stock Exchange, the NYSE, the Chicago Mercantile Exchange, Chi-X – all of these services and institutions rely on a key technology, Linux. Linux has become the fabric of most of the technology systems that drive our daily activities.
The Linux operating system is an open source OS based on the Linux Kernel originally created in 1991 by Linus Torvalds. Since then the Linux kernel has received contributions from 12,000 programmers from more than 1,200 companies. The latest release on 19/2/2017 was 4.10 and there are new version releases every two to three months. Version 2.0 was less than 2M lines of code; version 4.10 is more than 22M lines. A 2011 analysis by David Wheeler estimated the cost of redeveloping the Linux kernel version at that time at $3B.
New changes to the kernel are submitted and reviewed by a select group of version reviewers (subsystems maintainers – in the jargon of the organization) from the Kernel.org, in a very meritocratic and egalitarian way, before making their way into the kernel next version’s release.
As security issues and bugs are found and fixed on major releases (for example 4.9), new branch releases are made, for example 4.9.1, 4.9.2 and so on. In this way each major kernel release becomes more stable as it is maintained and used by the largest applications. Moreover, every four or five releases, a major version release is designated as a stable release. Stable releases are currently maintained for two years. As well as critical and security bug fixes being created for these stable releases, the kernel community guarantees that future releases will be compatible.
The man in charge of maintaining the stable branch and designating the stable release – thereby managing the kernel’s staging system – is Greg Kroah-Hartman. It is not an exaggeration to say that the security, stability and roadmaps of many of the world’s main technology systems indirectly rely on the contribution of Mr. Kroah-Hartman.
And finally, some fun…
The two most impressive DEMOS of the WEEK were
- RippleBuds – turns out you actually speak with your ears (although not as loudly). Ripple is the first in-ear microphone which is combined with an earbud (they cleverly solved the echo issues with software) thereby eliminating ambient noise. No more “please get out of the wind when you speak”. Great demo – funded on kickstarter.
- Lechal – a GPS-connected vibrating device in your shoes’ insoles which literally guides you to your destination by wirelessly linking to a HERE map on their phone app and sending vibrations to your feet signalling which direction to go. Originally built for blind people – brilliant demo, totally intuitive implementation. I think they should tackle ballroom dancing as the next step.