Please enjoy ARK Disrupt Issue 143. This blog series is based on ARK Brainstorming, a weekly discussion between our CEO, Director of Research, thematic analysts, ARK’s theme developers, thought leaders, and investors. It is designed to present you with the most recent innovation takeaways and to keep you engaged in an ongoing discussion on investing in disruptive innovation. To read the previous issue, click here.
1. SpaceX Has Signed the World’s First Private Passenger to Fly Around the Moon
Follow @skorusARK on Twitter
On Monday night SpaceX announced that Japan’s Yusaku Maezawa, a billionaire and avid art collector, will be the first customer for a trip around the Moon on the BFR (Big Falcon Rocket). As a result, Maezawa has launched the #dearMoon project and plans to take six to eight artists on the flight to inspire their work. ARK estimates that Maezawa is paying SpaceX roughly $300 million for the ride, which will help defray some of the BFR’s development costs. The mission is scheduled to take place in 2023.
Elon Musk has estimated that the cost to develop the BFR will range from $2 billion to $10 billion, with $5 billion his best guess. For context, the Falcon 1 rocket cost $90 million to develop and the Falcon 9 and Dragon combined, $696 million. According to the NASA cost model, the Falcon 9 would have cost $4 billion had it been developed by the government, or $1.7 billion commercially. Based on Musk’s $5 billion estimate and NASA’s analysis of the Falcon 9, if the US government were to sponsor development of the BFR, its costs probably would mount to as much as $50 billion.
Maezawa’s commissioning of artists is an interesting proposition. In 2017 he bought a Basquiat piece for $110.5 million. What will 6-8 pieces of art be worth if created in or inspired by space? Another golden age for space, for art as well?
2. Auto Insurers Should Fear, Autonomous Driving Networks Appear Near!
ARK’s research on Mobility-As-A-Service indicates that by 2030 the vast majority of miles traveled will be in autonomous vehicles. If this mobile transformation begins, an interesting question comes to mind: how will the auto insurance industry fare when autonomous cars dominate?
Today, State FarmSTFGX, GeicoBRK.A, AllstateALL and ProgressivePGR dominate the market and are competing to get a piece of the $220B+ in annual auto insurance premiums. Yet, the majority of them is struggling to turn a profit. In 2016, the average combined loss ratio of the 10 largest insurance companies in the US deteriorated to 107.1% from 99.7% in 2010. Among the forces behind the losses have been higher accident fatalities, rising litigation costs and more frequent catastrophic weather events.1 To stem the losses, insurers have increased premiums, as shown below.
Meanwhile, autonomous vehicles are getting ready to hit the road, adding a host of new problems to the insurance industry. According to ARK’s research, the biggest threats to auto insurance companies will not be rising fatalities or extreme weather events, but a decline in the market. During the next five to ten years, as current and potential drivers increasingly opt for autonomous networks, average insurance rates should fall, as shown in the graph above. In fact, the earliest adopters of autonomous vehicles probably may be drivers in the 16-24 age range who pay the highest premiums, as shown below. If autonomous vehicles go mainstream, insurers will lose their highest paying customers, driving the average rate of insurance down dramatically. By 2030, as even more drivers adopt autonomous vehicles, not only will average auto premiums fall back to levels last seen in 2011-2012, but the $220B+ auto premium market will be cut roughly in half, as shown in the second graph below.
3. ARK Gains Insights from the Intelligent Transportation Systems (ITS) World Congress 2018
Follow @tashaARK on Twitter
This week, I spoke at the ITS (Intelligent Transportation Systems) World Congress in Copenhagen about ARK’s forecasts for the future of cost effective point-to-point mobility via autonomous taxis. The event focused on Europe’s progress as global transportation shifts toward autonomous electric mobility and the role that regulators and policy leaders might play in helping the transition.
Europe is behind2 the US and China in developing autonomous vehicle technology, as testing and mapping laws vary from country to country. Tests often are limited to specific streets and routes, suggesting that European automakers are collecting fewer real-world miles than their American and Chinese competitors. Moreover, startups have had difficulty raising capital in Europe.
While traditional European automakers have enjoyed great success manufacturing gas powered vehicles, their hardware-centric, budget-conscious, and conservative mindsets could hold them back in the autonomous technology space. Their success will be determined by software, data, and algorithms, and a dramatic change in their approach to research and development.
Meanwhile, in the US WaymoGOOGL plans to commercialize its autonomous taxi service in Phoenix, perhaps pressuring European auto manufacturers and regulators to move forward faster than otherwise would be the case.
4. First Trained to Identify Dogs, A Computer Algorithm Detects Cancer
Follow @wintonARK on Twitter
Researchers at New York University (NYU) have created a software model that identifies lung cancer from patient biopsies. Adapted from an open source model that Google introduced in 2015 for image classification, the model not only diagnoses cancer with the same accuracy as medical professionals but also distinguishes among different types of cancer at a superhuman rate. In line with best practices, the NYU researchers first “pre-trained” the model by teaching it how to identify dogs, as well as 1,500 other categories of objects that are available among millions in the open source Imagenet database, before training it on the specific sub-domain of interest, in this case hundreds of thousands of publicly available images in the cancer genome atlas.
Seemingly intuitively and magically, pre-trained image classification algorithms appear to yield more accurate results faster than medical professionals. Fed with examples of dogs, cats, cars and trees, the deep learning model contains optical filters that are useful for pattern recognition. High-contrast edges or corners can activate the higher layers of the model, while more abstract features like eyes or facial structures trigger the deeper layers. When a model is trained on a subspecialty like cancer, some of the deeper layers devoted to the face-like and other irrelevant structures are overridden, while the higher-level layers with more general-purpose capabilities remain fixed.
Pairing NYU’s research with Facebook’sFB illustrates the power of deep learning when fed by large data resources. In its paper, Facebook demonstrated best-in-class image classification after it pre-trained its model with 1 billion Instagram images annotated by user hashtags. Heretofore, the standard data set has been only 1 million Imagenet images, commercially annotated. Pre-trained with the Instagram images instead of publicly available data, Facebook’s model reduced error rates by more than 22%. Given Instagram’s scale—hundreds of billions of images—Facebook probably has the world’s best pre-training dataset.
Facebook probably could design a superior cancer detection model by combining NYU’s methodology and its pre-trained data set. In other words, one of the strange byproducts of neural nets and software 2.0 could be that companies with the most data will win not only in their expert domains, but across multiple verticals in unanticipated adjacencies.
5. Meituan-Dianping Is China’s Fourth Largest Internet Company
Follow @jwangARK on Twitter
China’s leading food, hotel, and travel marketplace, Meituan-Dianping, made its initial public offering (IPO) debut this week in Hong Kong at a $51 billion valuation, making it China’s fourth largest public tech company. Meituan launched in 2010 as a GrouponGRPN clone but rapidly morphed into a platform for online ticketing, food deliveries, and travel bookings. During the second quarter, Meituan Hotels captured almost half of China’s online hotel bookings, displacing industry incumbents Ctrip and Qunar.
Meituan has grown its food delivery business at an astonishing rate. During the fourth quarter, it delivered 15 million meals per day, more than all of its global competitors combined.3
Meituan is a textbook example of how China’s internet companies are growing and conquering adjacent markets far more quickly than their American counterparts. Its successful IPO potentially sets the stage for the larger IPO of ridesharing company, Didi, in 2019.
6. Good News: The Bitcoin Blockchain Does Not Seem to be Commoditizing
Follow @yassineARK on Twitter
Recently, Bitmain announced plans to roll out a long awaited 7nm Application Specific Integrated Circuit (ASIC), in conjunction with a new generation of mining machines. China-based Bitmain launched the last generation of Bitcoin mining machines, the Antminer S9, in 2016. In general, replacement cycles have extended over time, as shown below:
The primary reason for these extended life cycles is the increasing degree of difficulty associated with leading edge fabrication: successive nodes demand bursts in expertise and capital expenditures. Not many foundries are researching and developing cutting-edge nodes. To date, only three of 28 foundries globally are providing leading edge fabrication, suggesting that the bottleneck is at the foundry level.
Because several mining machine manufacturers already have rolled out a 7nm chip, many question Bitmain’s ability to deliver products on time. History would suggest that new bitcoin mining machines will be 30% less efficient than advertised.
Given the combination of lengthening replacement cycles, new players, and Bitmain’s delayed rollout, bitcoin mining seems to be commoditizing. Consequently, the threat of miner centralization seems to be diminishing, a positive for the Bitcoin technology ecosystem.
7. Paige.AI Secures Priceless Data from MSKCC
Follow @msamyARK on Twitter
Paige.AI, an AI-powered oncology diagnostic company, secured an exclusive license to 60 years of Memorial Sloan Kettering Cancer Center’s (MSKCC’s) pathology notes, 25 million biopsy slides, proprietary computational pathology IP, and patient survival data. Its goal is to eliminate guesswork, helping pathologists to diagnosis cancer, provide the correct treatment, and predict survival outcomes.
Founded by Thomas Fuch, Director of Computational Pathology Lab at MSKCC, Paige.AI has been backed by a number of other top executives at the hospital, including Department Chair, David Klimstra. Many of its staff pathologists, upon whose decades of work and lab notes Paige.Ai may profit, are none too pleased.
That said, Paige.AI is working on game-changing technology, and its data advantage could dissuade other AI companies from entering medical artificial intelligence with a focus on oncology. Paige.AI already has implemented AIRI, the most advanced architecture built for scale-out AI, powered by NvidiaNVDA and productized by Pure Storage. Pre-training algorithms on image data instead of domain-specific data seems to result in superior predictive diagnostic outcomes. Paige.AI is digitizing roughly 200,000 biopsy slides a month, on pace to complete its training platform in little more than a year.
ARK's statements are not an endorsement of any company or a recommendation to buy, sell or hold any security. For a list of all purchases and sales made by ARK for client accounts during the past year that could be considered by the SEC as recommendations, click here. It should not be assumed that recommendations made in the future will be profitable or will equal the performance of the securities in this list. For full disclosures, click here.