Revisit Our Tech Shift Thesis

(modified from Gerd Leonhard’s Image on Flickr)

Two years ago, we published “Major Tech Shifts Drive Investments” as a high-level explainer of Tsingyuan Ventures’ software investment theses. At Tsingyuan Ventures, we invest in early stage technology startups that can potentially generate big returns. These types of startups typically happen alongside major technological shifts in the world. The shifts are necessary as they create windows of opportunity for startups to challenge/disrupt much larger incumbents of the respective sectors. In the blog post we explored 5 areas of technology shifts.

This post revisits our thinking from two years ago to explore our subsequent investments in that context, and whether the shift is still one that we’re betting on.

1. Enterprise Software Shifts to Public Cloud

With the move to the public cloud, traditional firewalls and traditional VPNs are losing their grip. We have made three bets: Stellar Cyber and Banyan Ops to address these two aspects of the platform shift.

• Stellar Cyber: built by an experienced serial entrepreneur, Stellar Cyber starts out to address the new enterprise security challenges associated with modern enterprises’ move to the hybrid cloud model. This shift shifted the attack surface and the severity of attacks. Stellar Cyber is purpose-built for the data collection and analytics for the distributed, heterogeneous environment. Leveraging machine learning algorithms, it can take full advantage of the data collected while adapting to changing anomaly patterns a few steps faster.
• Banyan Security: With the shift to public and private cloud, every employee and every service becomes remote. The point to point nature of VPN becomes an obsolete mode of access control or protection. Banyan Security builds a transparent, secure service mesh that allows simple access to internal services (either by employees, contractors, partners, or by other services). It is ideally suited for a containerized or micro-service heavy environment (which is more and more common). Banyan Security is going to be the new VPN.
• Openprise: This investment is indirectly related to the shift described. Openprise is a “data orchestration platform” that allows data to be seamlessly passed from SaaS application to SaaS application (in the pre-SaaS world, much of this work would be done by a costly and time consuming Informatica custom project). Openprise is one of the enablers for the deconstruction of the enterprise stack.

We continue to be bullish on this trend. All three of the companies that we bet on in this space are experiencing significant demand, and have the opportunity to be global players. We have reviewed a large number of other companies in this space but haven’t pulled the trigger on more investments. We have a high threshold for the team and technology due to the “winner take all” nature of this space.

2. Device Intelligence: Shift to Edge Computing

As devices (camera and sensors) become more capable in collecting large amount of data, this is a bet that computing on the edge (near/at the senor) becomes more attractive in cost, responsiveness, reliability and flexibility. We have made several bets — Turing Video, Biscuit and IntEngine — in this shift:

• Turing Video: Turing is primarily a bet on computer vision (as described in the next section) but also a bet in edge computing as real-time detection and management based on the richness and voluminous of video data are best computed on the edge. See more discussion below.
• Biscuit: Biscuit is a building IoT company. Biscuit devices include a large number of sensors (motion, particles, temperature, light, etc) and utility controllers. When integrated with their software, large building owners can better analyze and improve utilization, air quality, and other measures for commercial building management.
• IntEngine: an ASIC design company focused on low-power, custom designed AI acceleration chips, IntEngine is betting on the rapid rising demand of intelligent appliances. Instead of relying on cloud-based AI such as Alexa/Google/Siri, its low-cost, low-power consumption AI chips can be easily integrated into air conditioners, elevators, lightings, fans and many other appliances to first add voice command capabilities, and later vision-based intelligence as well.

We believe this area will continue to evolve, driven by sensor computing (such as Sony’s IMX500 sensor) and ever powerful mobile devices. Going forward, its impact as a standalone shift may continue to evolve now against the bigger backdrop of networking/connectivity infrastructure improvements in 5G, wifi6 and starlink etc. We are currently looking at a variety of companies — from new chip designs to more efficient neural networks optimized for edge computing.

3. AI: Deep Learning in NLP and Computer Vision

In this case, we have bet big, and we remain bullish. Deep learning in NLP and Computer Vision are strongly represented in our portfolio, including:

• Subtle Medical: Computer Vision applied to medical imaging. Subtle Medical won the race to be the first company approved by the FDA for AI-enhanced images for both PET and MRI scans. Their product allows hospitals and clinics to dramatically reduce the time required for high quality images (freeing up the capacity for their most expensive and important machines), and requiring much less of the radioactive tracers used for current technologies
• Turing Video: Computer Vision applied to building operations. This has been one of the more interesting companies in our portfolio — the company started as a computer vision platform to assist security guards, but quickly expanded to many use cases for building operations (analyzing everything from in-store shopper counts and demographics, environmental quality (temperature, air quality, light quality, etc), and operational efficiency. Recently, the company has expanded its offering to include temperature sensors that are necessary for re-opening post-COVID
• WeRide: self-driving car company (discussed below under “shift to AVs/ EVs”) that leverage computer vision extensively in its perception models, sensor fusion and HD mapping.
• Otter.ai: leading automated transcription service, tightly integrated with Zoom. We originally invested in Otter as a reflection of our interest in deep-learning applied to NLP, and for Otter’s potential as a platform. The COVID crisis has caused Zoom (and Otter) usage to skyrocket, as both video conferencing and automated transcription have become core productivity tools for “the new normal”.
• DeepScribe: Doctors are overwhelmed by paperwork. Automated note taking tools and voice dictation are helpful, but not sufficient. An ideal tool for a doctor has to have a very high degree of accuracy, understand medical terminology, and be well integrated into EHRs. Doctors frequently turn to scribe services, which provide people (often nurses or medical students) who transcribe notes, extract diagnoses and prescriptions, and enter into the EHR system. It is helpful to doctors, but expensive and hard to scale. We invested in DeepScribe because they were implementing a novel statistical technique, developed at Berkeley, that could more efficiently extract medically important information from natural conversations. They have also seen a major uptick in usage during the pandemic, since the automation used for conversation is also useful for telemedicine more generally.
• DeepHow: Our first Detroit-based investment. DeepHow automates the process of creating high quality instructional videos, initially aimed at skilled trades working at large enterprises (for example, how a Siemens mechanic should maintain every piece of equipment in one of their facilities). Because of the use of NLP and object recognition, DeepHow can easily index videos, making it easy to find relevant information, slice the video into sections, auto-caption in multiple languages, etc. Their technology is currently aimed at a narrow use case, but one can imagine this technology being applied more generally to the giant category of “how to” videos.

4. Computing Shifts from CPU to GPU

While we still stand behind our assertion of the limitations of CPU due to the unattainable thermo profiles of higher clock speed, the use case of GPU remain limited due to the lag and algorithmic challenge in converting larger number of computing problems into the SIMD (single instruction multiple data) model GPU is designed for. It continues to perform well in certain AI tasks but the transition to general tasks such as Databases has remained slow.

In high performance computing, compute units (compute cores), memory and network form the most important trios. Having large numbers of fast compute cores in the case of GPUs have pushed the bottleneck to storage/memory and data buses.

In fact the flexibility of out-of-order computing, random access and convergence of storage and memory enabled by “big memory computing” is really taking off, led by another Tsingyuan portfolio company Memverge. Memverge took a bet on the emerging standard of high performance non-volatile memory architecture called 3D Xpoint with significant improvements in write-latency, caching, and other key computing metrics. MemVerge recognized that the race was on to create the virtualization platform that would allow software to take advantage of the new architecture.

Overall, we still believe in the general trend of high performance computing that can power AI as a service. The evolution nature of compute units from CPUs marching towards many cores, the rise of lower-power mobile CPU/GPU cores, and other in-memory innovations have reshaped this area.

5. Shift to EV and Autonomous Driving

In the past two years we have bet on a number of companies in the energy tech space. In energy tech, the general theme has been around faster charging battery technologies. The explosion of battery demand driven by EVs plays into our strengths: the industry is dominated by the US and China, and the innovative companies are classically deep tech (PhDs in chemistry, physics, and nanomaterials).

• Nimbus Energy: disruptive solid state energy storage technology allowing a battery that can be safely charged extremely rapidly with practically no limit on recharge cycles. One key industrial use is as a supplemental battery for EVs
• GRU Energy Labs: developer of a novel anode technology that will allow faster charging and higher capacity for Li-ion batteries
• Next-ion: developer of an aerogel membrane that allows batteries to be safely charged at very high speeds without thermal runaway (catastrophic explosions).
• WeRide: Weride continues to lead in the China market in terms of commercial autonomous taxi services in Guangzhou. Compared to its US counterparts, they also have observed that the intensity of observations Chinese metro area is about 30X higher than typical US suburban areas where most US based autonomous vehicles operate (this is critical for training and improving the models that are being used for autonomous operation). We believe that this is yet another advantage in speeding up the learning process.

We continue to be bullish on the opportunity for battery technologies, especially for companies that are able to operate cross-border.

On the other hand, the last 6 months has seen a cooling off of investments in self-driving. Since our initial bet on autonomous driving company Weride, this sector has seen increased competition as well as more realistic expectations as technology starts to mature. We continue to see automobile electrification happening at a fast pace in China. And we continue to believe autonomous driving will be adopted at scale in China first as well.

Where does this leave us:

Our theses ended up driving a significant portion of our software and materials science investments in the past two years. All of those aspects of our portfolio are fairly well-positioned, but the “deep-learning in NLP and Computer Vision” cluster seems particularly strong.

Many of these theses remain relevant going forward. Notably, we have been slowing down our investment pace in self-driving, as we already have a number of important investments in the space, and we believe that the industry now has to absorb some of the hype that drove rapid investments in the past.

We will next write about some of the tech shifts that should drive our next two years. Here is the link!