What are the top operators and VCs saying about robotics?

A consensus is forming among operators and VCs that the primary bottleneck in robotics has shifted decisively from hardware to software [1, 4, 13, 23]. Experts assert that hardware has been sufficiently capable for years, and the central challenge is now developing "physical intelligence" through AI foundation models [13, 15]. This paradigm shift is fueled by the convergence of commoditized hardware, advances in edge compute, and the increasing generalizability of AI . A critical component of this shift involves a new approach to data collection, moving away from slow, unscalable teleoperation towards massive datasets of "sensorized human data," primarily egocentric video . This new data paradigm is seen as the key to unlocking a neural scaling law for physical dexterity, with some predicting teleoperation will become a negligible data source in the **next year or two** [6, 14].

The central unsolved problem in the field is the absence of predictable scaling laws, which in the world of LLMs connect investment in data and compute to model capability [1, 9]. Establishing a similar formula for robotics would transform the field from a high-risk research endeavor into a more predictable engineering problem, providing the conviction needed for massive capital investment . A key technical breakthrough is the ability to transfer knowledge from pre-trained vision-language models (VLMs) to robots by fine-tuning them with a small amount of robotics data, enabling generalization from internet-scale information [1, 8]. However, while current models show impressive generalization on complex tasks, their performance is still considered **"grad student" level**; achieving deployment-ready reliability remains the next major hurdle . The core difficulty lies not in modeling the robot's own movements, but in modeling its interaction with the highly variable and unstructured physical world .

This technological shift is creating a bifurcated outlook on the investment landscape. One optimistic view predicts a "Cambrian explosion" of vertical robotics companies, as the required capital and expertise to start a new venture are decreasing rapidly [17, 18, 19, 22, 27]. This could create a vibrant ecosystem of specialized startups building on generalist AI platforms, presenting "picks and shovels" opportunities in areas like simulation [7, 10, 22]. The potential market expansion is considered vast, with one investor speculating that functional robotics could expand the total addressable market for venture capital by a **factor of 100** . Some believe the industry is on the cusp of a "GPT-3 moment," with large-scale deployments imminent [21, 29].

However, this optimism is tempered by a more sober perspective from experienced operators. Adam Bry of Skydio warns that many recent robotics investments will likely fail because the physical world imposes slower, more capital-intensive learning cycles that do not align with typical software investment playbooks [11, 25]. Founders are cautioned to focus on substance over hype, as achieving extreme reliability in physical products requires mastering complex manufacturing and operational hurdles with long development timelines [20, 25]. This tension is reflected in timelines, where some researchers are more pessimistic than entrepreneurs . Ultimately, the biggest risk cited by some is not competition, but the fundamental scientific challenge—the possibility that the problem of generalist physical intelligence may not be solvable with current methods .