#203: Chips, Automation, EVs, NGS, Fintech
1. Nvidia Announces Its Next Generation Self Driving Chip
Last month Nvidia announced Orin—the successor to its Xavier chip for autonomous vehicles. Capable of 200 trillion operations per second, Orin is almost seven times faster than Xavier and will ship in a variety of configurations ranging from one to four chips.
Orin is roughly 40% faster than Tesla’s Full Self Driving (FSD) but won’t ship in production cars until 2022, the main reason Tesla chose to design its own AI chip. Had it waited for a third-party chip supplier, Tesla’s self-driving program would have been delayed for at least three years.
Nvidia’s Orin and Tesla’s FSD highlight how horizontal and vertical business models can impact performance and delivery timelines. Orin is faster and more feature rich but needsmore complex specifications to support dozens of customers with varying requirements. Tesla’s chip is much simpler because it has only one customer—its internal software team—and as a result will ship one generation faster than comparable third-party suppliers.
Like ARM and Apple in the mobile market, both Nvidia and Tesla can be winners in autonomous driving. We believe the losers will be traditional OEMs as the automotive value chain shifts from engines and gearboxes to silicon and software.
2. As Margin Pressure Mounts, Automation Could Save Food Delivery Providers
Recently food delivery firms have been consolidating, a trend likely to accelerate as Google threatens them with disintermediation. Recently Google enabled users to order and pay for restaurant food in Google Search, Maps, and Assistant, skipping delivery provider apps altogether. As a result, Google is likely to pressure delivery company margins even more than recently has been the case.
To survive, food delivery platforms are likely to integrate with dark kitchens, robotic food preparation, and autonomous delivery. This week, for example, Michigan-based startup Refraction AI began offering food delivery in Ann Arbor using rolling robots to undercut the fees that GrubHub and other food delivery services charge restaurants. Refraction’s co-founder explains, “We’re not spending Bentley money to drive around your tacos.”
Across the logistics space, ARK expects purpose-built autonomous vehicles on the ground and in the skies to enhance the productivity and consumer delight associated with food delivery.
3. Tesla’s Manufacturing Productivity Has Surpassed That of Traditional Auto Manufacturers
In 2017, Tesla needed nearly 12 months to ramp Model 3 production to 3,000 vehicles per week, as shown here. At the time, bears were convinced that Tesla never would scale manufacturing, relegating it to a niche existence.
Two years and nearly 460,000 Model 3s later, Tesla built its Shanghai Gigafactory in 12 months and ramped manufacturing to 3,000 vehicles per week in just 3 months. For context, Toyota took four years to build a manufacturing plant in Texas with initial capacity similar to that of Tesla’s Shanghai Gigafactory.
Decades of manufacturing no longer seem to be a competitive moat in the automotive market. Quite the contrary, traditional auto manufacturers still are struggling to produce electric vehicles that can compete with the first Model S that Tesla produced in 2012.
4. Oxford Nanopore Technologies (ONT) is Gearing Up for Clinical Adoption
While investors were focusing on Illumina’s (ILMN) acquisition of Pacific Biosciences (PACB), Oxford Nanopore Technologies (“ONT”) quietly announced key hardware and software upgrades during its Nanopore Community Meeting (“NCM”) in December. Prior to these upgrades, clinical diagnostic providers had shunned ONT sequencers because of their lower accuracy, frequent update requirements, and lower sample throughput.
During the NCM, ONT announced a new nanopore architecture. These new nanopores measure voltage instead of changes in electric current, aiming to improve the accuracy of reading the difficult stretches of DNA involved in oncology. ONT also discussed chemistry improvements that could increase throughput to 7TB of data per run, slightly outperforming Illumina’s NovaSeq 6000. Finally, it announced Q-Line, a special class of ISO-regulated instruments built for diagnostic sequencing tasks.
While preparing to penetrate the clinical sequencing market once again, ONT has had a history of over-promising and under-delivering on an “Illumina-Killer” for roughly a decade. If successful in any way, however, ONT will introduce more competition into the sequencing space, perhaps incentivizing Illumina to accelerate its own cost declines and increasing access to whole human genome sequencing.