As Valuations for Self-Driving Technology Craters, Amazon Buys Zoox

Amazon recently announced its acquisition of Zoox, a developer of Autonomous Passenger Vehicle (APV) technology and an aspiring developer of driverless taxis, for a reported US$1.2 billion. This was markedly less than the company’s 2018 valuation of US$3.2 billion, but since 2018 the optimism about the self-driving car opportunity has fluctuated. Waymo, the market leader in testing, once had an astounding US$175 billion valuation, but this has since plummeted to US$30 billion (17% of the initial valuation). Cruise Automation, General Motors’ (GM) main horse in the Artificial Intelligence (AI) race and second place to Waymo, is valued at US$18 billion. Cruise Automation suggested that unmanned automotive technology represented no less than an US$8 trillion opportunity, larger in fact than the entire Japanese economy. Most of this value (US$5 trillion) is focused on ride-hailing.

Suffice it to say these projections are rosy in the extreme, and this market has so far failed to live up to the enormous expectations placed upon it.

Zoox has faced lawsuits from Tesla due to former workers allegedly taking documents as they moved between the two developers. This has been a regular feature of the APV space, with Uber having to pay Waymo for similar infringements. Zoox also made waves by enticing 14 employees from Apple’s secretive Titan project. This is common jockeying in Silicon Valley, but very distinct from the far less ruthless atmosphere in robot hubs like Pittsburgh or Boston. This “playing to the edge” is based on the widespread assumption that the prize for the first real achiever really is US$8 trillion. Thus, there are feverish attempts to take staff and Intellectual Property (IP) from competitors and intensely guard it.

However, while APV technology is exciting and presents considerable opportunities, there are important things to consider:

• Is APV Technology Scalable? Deploying APVs and driverless taxis will not scale like digital platforms. Even Uber doesn’t scale like Facebook or Google and makes obscene losses every year. Every major ride-hailing company makes losses and their low prices are subsidized by investors. Likewise, the process of commercializing this technology is long and arduous, with considerable regulatory hurdles to overcome even when ignoring the technical pitfalls.
• Is Autonomous Ride-Hailing Viable? Automating ride-hailing would theoretically make it profitable, but what are the chances of deploying this in the next 10 years? Uber sees it as essential to profitability, but it will take a decade to scale. Waymo, the market leader, initially wanted to develop autonomous ride-hailing, but has since suggested it will not attempt this, given the lack of profitability from its small taxi service in Arizona. Anything less than Level 5 autonomy (which is a long time away) will not make ride-hailing profitable.
• Is It Really Winner Take All? While there are official taboos about copycatting In Silicon Valley, in other jurisdictions like China there is no such taboo. Like many markets, there is a good chance that the most innovative companies will lose out to those that can copycat and optimize. Autonomous driving will therefore not follow the trajectory of digital markets like search engines, where one company dominates by getting there first. More likely, advances in Silicon Valley will be quickly replicated elsewhere by local companies that are better placed to offer services to their respective jurisdictions.
• Is This AI Technology Overrated? The APV market has been hit by increased skepticism and the timeframes for rollout of Level 5 autonomy have been extended to 10 or 15 years from now. This continual shifting of projections suggests no one really knows when or to what extent the economy will be impacted. Some companies, including Starsky Robotics, have already stopped operating, including Starsky Robotics, whose former CEO noted that the technology used to determine road safety was limited and that, contrary to predictions, improvements in data lead to greater expenditure for new data to improve the system. In essence, the better your algorithm, the more it costs you to refine it further.  

So, with all this in mind, why is Amazon paying so much money to acquire a technology with no real business model? Well, it has been investing a lot in APVs for a while, with US$350 million having been given to software developer Aurora Innovation. Amazon has also made significant plays in both robotics and the automotive space. In 2019, it placed a US$5 billion order for 100,000 fully electric Rivian trucks. Though these will not be showcasing Level 5 autonomy, they could be upgraded with Level 3 and 4, and having Zoox in-house could help Amazon facilitate gradual automation.

Amazon might use Zoox in a few ways:

• Supplement Robotics and Unmanned Logistics (The Most Likely Option): Amazon has an enormous, almost unprecedented, pool of talent dedicated to mobile robotics. Beyond Kiva Systems, it has acquired vision-based perception company CANVAS and mobile delivery robot maker Dispatch. It has also funded autonomous Forklift developer Balyo. With Zoox being added, Amazon has more engineering talent for robot navigation than any other company. Zoox’s self-driving software may not be deployed on the road in any real commercial capacity, but it could be invaluable for automating industrial field vehicles like tractors or semi/yard trucks. This could either be through developing internal products at Amazon or leasing the platform to outside robot developers or Original Equipment Manufacturers (OEMs).
• Develop Ride-Hailing with Some Autonomy: As mentioned previously, the technology doesn’t allow for this at scale, and it will be a loss-maker for years. But then, Amazon is much better placed to develop a loss-making ride-hailing service than Uber is. The latter needs constant outside subsidy, while Amazon can use up the profits of Amazon Web Services (AWS) to nourish continual losses in ride-hailing. Even if robot taxi services are limited to a few test locations, they can eventually be scaled up to supervised autonomy for ride-hailing. There is also the possibility that Level 5 is just in reach and that the bloated market, while oversaturated, is the precursor to more tangible offerings. If so, it is an open question whether Zoox fits as a pioneer or as part of the overhyped ecosystem. Depending on the answer, Amazon may have either bought an “also-ran” in the self-driving space for a bargain or struck gold just as the market looked to be losing steam.
• Develop Teleoperated Freight: Amazon has invested in both Rivian and Aurora Innovation. The former is developing electric freight trucks while the latter has shifted to automating semi-trailer trucks over APVs. While APVs have to navigate complex environments, trucks often drive on straight highways that are less dense. The purchase of Zoox will likely translate into Amazon developing a presence in automated long-distance freight. This will be not be fully autonomous anytime soon, but it doesn’t have to be. Teleoperated freight would, by itself, significantly improve worker retention and alleviate the labor shortage in long-distance freight across the United States.

Ultimately, Amazon bought an APV developer on the cheap following a massive depreciation in the valuation of the market. There is now no pressure on Zoox to make immediate returns, so onlookers can expect it to be melded into Amazon’s wider plans to dominate the mobile robotics space.

APV as a business proposition has been put on ice for the foreseeable future. The suspension of test driving during COVID-19 was a major blow and, given the redundancy of the testing, interest in this space will dry up. Energy is not dissipating, but is rather going into last-mile delivery and long-distance freight.

Of these two opportunities, last-mile delivery is much more readily available, and this is what Amazon’s robotics team will be rolling out next. They have been developing their Scout delivery robot and expanded testing to two new jurisdictions: Atlanta, Georgia and Franklin, Tennessee. The Scout is building on the success of another last-mile delivery developer, Starship Technologies, which has deployed robot delivery services across a number of jurisdictions (including campuses and small cities) and is regularly used for grocery and takeaway services. ABI Research expects Amazon to develop last-mile delivery for both smaller payload robots (for pavements) and larger payload automated pods (for particular neighborhoods) and use Zoox to power technology for the latter.

There is also the question about whether this technology stays in-house. Amazon has rolled out testing for both drone and robot delivery but may also use its considerable resources in mobile robot navigation to create robot-related services for upcoming developers. This would be complimented by offerings from AWS, which is creating cloud-based tools for developers and end users to create, manage, and use robots as a service.

Elon Musk referred to Amazon as copycats after its Zoox acquisition became public. This is partially true; Amazon’s robotics journey has not been defined by innovation, but rather by implementation. That said, it is implementation on a grand scale. Whether the buying of Zoox will manifest in new uses of robotics is up for debate, but it will help cement the e-commerce giant’s dominance in deploying mobile robots throughout the supply chain.