#2 — M&A in the AI startup space
You may refer to my first article on Introduction to AI if interested. In this article, I would focus on the following points: 1) M&A activity in AI 2) The AI buyer landscape and precedent transaction analysis
M&A activity in AI
We clearly see an uptick in deals over the last 3 years which ties up with what we have read and understood so far. To me this chart highlights 2 important issues:
1. The M&A market currently is extremely hot. The last 5 quarters have cumulatively resulted in more deals than since the evolution of AI M&A. Some skeptics label the current market as similar to that of the dot com bubble in 2000–02, but I believe this is a genuine value based boom. Well the question remains if the valuations are justified, but the way I see it, this is clearly the incumbents poaching for tech and talent. I wouldn’t be surprised if this graph grew in number of deals (not multiples) on an exponential basis over the next 5–10 years to come. Are the acquirers over-paying? one will really never know!
2. Strikingly, it was only until early 2013 that incumbents started acquiring AI ventures, highlighting how much of a nascent market it was back in 2012 and before. On a personal level, in 2012, when I graduated from a technological institute in India, AI was clearly not a buzzword. Today, interestingly when I speak to my brother who graduated from the same university and who currently works as a data scientist, he claims most of the stuff that they deal with or look upto is AI tech! The last 3 years have been phenomenal with respect to market sentiment towards AI.
The AI buyer landscape and precedent transactions
I have selected a handful of these transactions to try to illustrate the different aspects seen in the AI M&A space.
a) Google / Kaggle (US)
Target description: Crowdsourced platform for predictive modelling and analytics competitions on which companies and researchers post their data and statisticians and data miners compete to produce the best models
Deal rationale: This is Google’s inorganic way to acquire a target customer base of c.0.1m developers and companies that are actively involved with AI and cross-sell its cloud and library offerings. Eventually Kaggle would serve as the data science service of Google.
Transaction details: Undisclosed
b) Google / Deep Mind (UK)
Target description: Deep learning venture known for the AlphaGo program that defeated a human professional Go player for the first time. Elon Musk was also invested in DeepMind.
Deal rationale: Google bet facebook to this acquisition. DeepMind is one of its kind of a unsupervised deep learning AI (roughly means that the AI is not trained for any data, the algorithm is created by itself). This is one of the most complex aspects in AI, and is Google’s bet on a promising early stage venture. Applications of DeepMind can be limitless but here is an interesting function it is doing for Google. Auto-controlling air temperatures at google data-centres could increase energy savings by roughly 40%. As usage of the various google applications increases, data centres heat up and the AI predicts this without any programmed data.
Transaction details: EV (enterprise value): $500m
c) Apple / Realface (Israel)
Target description: Facial recognition company
Deal rationale: Logically, this would be implemented in the future product roll-outs of apple. Video based identity recognition is the next step starting from the age-old hard key and then to the number lock and the current finger print technology. Video recognition in its crude version have been seen in a few smartphones but Realface has achieved high levels of accuracy in this.
Transaction details: EV: $2m
d) Twitter / MagicPony (UK)
Target description: Uses neural networks to improve images especially in applications that include virtual reality / augmented reality (VR / AR)
Deal rationale: This is Twitter’s bet on video content and increasing the video usage by its customers. One of the challenges in video based sharing is the lack of quality in visuals as online videos are often compressed to low-res to enable streaming. MagicPony is capable of modifying these low-res, pixelated video into a much more clearer version through AI. So today if I upload a video of a football game using my 5MP camera on a gloomy London evening, the video gets uploaded the same way, but the AI will identify the contents of the video, understand it’s a football and render its pixels accordingly. It will also understand that there are humans running around with hands, legs etc and will improve the quality of the video. Video is again critical to our future interaction especially with 4G and + just around the corner, this acquisition is of utmost strategic value to Twitter and its video play.
Transaction details: EV: $150m; Employees: 11 PhD (EV / PhD: 13.6x); Goodwill on Twitter’s BS: (47%, which is quite common in tech M&A deals)
e) Ford / Argo AI (US)
Target description: Founded by former Google and Uber leaders, Argo AI develops AI tech for autonomous driving
Deal rationale: This is Detroit’s biggest investment in autonomous driving. Ford is investing $1 billion during the next five years in Argo AI, combining Ford’s autonomous vehicle development expertise with Argo AI’s robotics experience and startup speed on AI software — all to further advance autonomous vehicles. Argo AI will include roboticists and engineers from inside and outside of Ford working to develop a new software platform for Ford’s fully autonomous vehicle coming in 2021; through their equity participation, Argo AI employees will share in the startup’s growth. Self driving tech led by Tesla is one of the most talked about space for AI applications. In December 2015, Google spun-out its self-driving car business into a new company called Waymo. The company, which falls under the Alphabet umbrella, will create self-driving cars for running errands and commuting. Similarly GM has purchased software company Cruise Automation and Ford itself has invested $180m in code company Pivotal
Transaction details: EV: Undisclosed. Structured as an equity share for current owners of Argo AI with Ford’s commitment to invest $1bn over next 5 yearts
f) Amazon / Angel.ai (US)
Target description: Chat bot service that operates as a digital personal assistant, using AI to fulfil requests. This startup was among the first bot ventures to raise VC backing from a major investor
Deal rationale: Acqui-hiring CEO and investing in conversational commerce was at the core of this deal. Interestingly it was in the same week that Google acquired API.ai, a company centered on machine learning and the creation of chatbots and personal assistants. Mr.Hadzaad the CEO has now joined Amazon’s team.
Transaction details: EV: Undisclosed. Acqui-hiring of CEO
g) Uber / Otto (US)
Target description: Otto focuses on self-driving technology that could be fitted into trucks that are already on the road. Thus, Otto didn’t want to develop its own self-driving vehicles, but focuses on producing a self-driving kit based on AI
Otto fitted perfectly into Uber’s strategy as the company didn’t want to become a car manufacturer. Instead, Uber has been looking at partnerships with existing car manufacturers, such as Volvo, in order to turn their cars into self-driving cars using Uber’s proprietary technology.
Transaction details: EV: $680million (share transaction); Employees: 91 (EV / Employee: $7.5m/employee and EV / PhD would be further north)
Having read through these 7 transactions that were hand-picked, following are the possible key takeaways from the analysis:
1. Size: Incumbents are actively looking to bolt-on their AI arsenal and therefore all companies irrespective of their size fall under their radar — from acqui-hiring, small bolt-on (Apple / Realface) to a large M&A (Google / Deepmind).
2. Rationale: DeepMind was a mature firm and Google had to shell out $500m to acquire the company. This is precisely what incumbents want to avoid and hence nascent tech will be acquired either for talent (as in Amazon / Angel.ai) or its wider scope (as in Apple / Realface).
3. Industry: The Uber and Ford AI investments are examples of M&A outside the realm of pure technology companies. Automotive industry and self driving technology is one of the leaders of AI utilisation. But this also leads us to think that soon AI will go beyond tech and auto and will to expand to other industries like healthcare, oil and gas, manufacturing, agriculture etc; Having said that there is also a possibility that AI products would be offered as a service or a as a platform for other industries on a licence / subscription model by the current incumbents. It wouldn’t be of surprise if lets say Halliburton subscribes / licences a Google AI offering at it’s drilling services to identify new reserves as opposed to trying to develop this in-house or acquire in the market.
4. Valuations: As most of the transactions are early stage, detailed financials are often undisclosed to determine valuations. However as startups are often valued using the VC method, often, cash flow forecasts, bottom line etc; do not really make sense. Especially for an AI tech VC, understandably the upside can be limitless. Imagine an app like Realface being used to rollout in the next edition of iPhone8 or Angel.ai being leveraged to enhance customer experience at Amazon. These can have extremely significant implications on the parent company. Therefore in any such deal, it will be interesting to see how much of value is created for the buyer with high certainty in the near term and then use that benchmark to arrive at a reasonable valuation. A lot around value determination for these early stage AI startups will depend on the buyer and its overall strategy for the acquired venture. This makes the analysis challenging and equally interesting. Another benchmark for valuation is the talent metric. We saw the MagicPony deal acquired at c.$13.6m / PhD. This is again a good benchmark for lets say Facebook to consider when they are looking to boost up their video enhancement AI portfolio through M&A.