Ideas
Self-driving cars are now mainstream
Alex says self-driving cars were a longtime pipe dream, but the answer is now yes—they are working, and in the Bay Area his friends use them day in and day out. He views self-driving cars as having become a normal part of life, implying the technology has reached mainstream adoption.
Zipline is scaling autonomous delivery rapidly
Zipline has a rare vertically integrated autonomous delivery platform: it designs and builds the aircraft, ground infrastructure, air traffic management, service centers, and customer app. It has flown over 135 million commercial autonomous miles with zero safety incidents, expanded to 5,000 hospitals and health facilities, launched Platform 2 at scale in the U.S., and is growing around 15% week over week while launching a new Walmart Super Center weekly. Keller says the business is capital intensive, but top investors are backing it, and Zipline is accelerating metro launches with the goal of national scale.
Drone delivery economics beat car delivery
Drone delivery and automated logistics have fundamentally better economics than car-based instant delivery. Using a 4,000-lb gas vehicle and human driver to deliver a 5-lb package is inefficient; a 50-lb electric robot/drone can deliver in 3-5 minutes at half the cost with zero emissions. Keller expects costs to fall about 30% per year while car-delivery costs rise, and says this will become one of the biggest markets on Earth over the next 3-5 years.
World models are next AI frontier
World models are the next frontier of generative AI. Unlike text or video models that take one prompt and return an output, world models build a representation of an environment and simulate actions continuously in real time, allowing users to navigate, converse, or control robots. This enables simulation of rare edge cases and creates broad capabilities across robotics, autonomy, avatars, and interactive worlds. World models will need more compute than language models and are improving rapidly, with Runway's GWM1 as an early general example.
Runway leads world models via video foundation
Runway has a differentiated position in world models because it has spent seven years pre-training its own base video models, culminating in Gen 4.5, which it post-trains into GWM1 for continuous world generation. Competitors post-training open-source video models lack control over pre-training and are limited. Runway's model generalizes across robotics, avatars, and interactive worlds, and has attracted strong interest across different audiences. The company is compute-constrained but working with Nvidia and CoreWeave to scale.
World models boost robotics and autonomy
World models are particularly valuable for robotics and autonomy because real-world policy training data is slow, expensive, unsafe, and scarce for edge cases such as accidents. World models can generate rare scenarios and evaluate policies in simulation before deployment, making robots and autonomous vehicles much more robust. Anastasis also believes video models may eventually become the policies themselves, directly controlling robots or cars.
Generative video will power real-time software
Real-time video is Runway's second pillar. Anastasis expects generative video models to generate software experiences on the fly, going beyond vibe coding, with avatars as the easiest starting point and VR as an eventual destination. Over time, a lot of software could be driven by generative video models, creating a large platform opportunity.
This This Week in Startups video, published January 21, 2026,
features Alex, Keller Clifton, Anastasis Germanidis
discussing Autonomous vehicles, Zipline, Drone delivery, Autonomous logistics, AI-SECTOR, Runway, ROBO, Generative video.
7 trade ideas extracted by AI with direction and confidence scoring.
Speakers:
Alex,
Keller Clifton,
Anastasis Germanidis
· Tickers:
Autonomous vehicles,
Zipline,
Drone delivery,
Autonomous logistics,
AI-SECTOR,
Runway,
ROBO,
Generative video
