a16z's Martin Casado: China dominates 50% of global robotics deployments — America needs to act now

Summary

Martin Casado, general partner at Andreessen Horowitz leading its infrastructure practice, made the case that China's robotics dominance is a more urgent national security concern than its AI software capabilities, and that the US policy response so far is badly lagging.

China's Robotics Position

Casado's team conducted a market study on global robotics originally scoped to inform a16z's software investing thesis. The findings were stark: China accounted for roughly 50% of all global robotics deployments over the past three years. China has also locked up the robotics hardware supply chain and owns significant positions in what were previously Western industrial robotics leaders, with Casado citing KUKA, the century-old German robotics manufacturer, as now Chinese-owned.

The DJI precedent is the operating analogy. Chinese companies proved willing to subsidize consumer hardware below cost to capture market share and crowd out domestic competitors, a dynamic that effectively ended meaningful American drone manufacturing. Casado and others in the conversation see Unitree, whose humanoid robots are currently listed on Walmart.com for approximately $20,000, as the next iteration of that same playbook, applied to a far more consequential product category.

Where Casado Draws the Policy Line

Casado distinguishes sharply between hardware and software when it comes to import and export controls. On software, including open-source Chinese models such as DeepSeek, he opposes restrictions categorically, arguing they are technically unenforceable and would handicap US researchers given how deeply Chinese academic output is embedded in the global AI ecosystem.

On hardware, his position is the opposite. He supports maintaining ASML lithography machine export restrictions, noting China still struggles with yields below 9-nanometer processes at scale. He also explicitly advocates for import tariffs and controls on Chinese robotics hardware, framing this not as economic protectionism but as a national security necessity. His argument is that the US should not allow dependence on a geopolitical rival for critical physical infrastructure, the same logic that eventually drove Huawei out of Western telecom networks, a process Casado notes took far too long given that concerns were already acute during his time working with the intelligence community in 2001 to 2003.

The Software Advantage the US Should Protect

Casado sees complex enterprise software as a durable American advantage. He challenges anyone to name a major Chinese enterprise software company comparable to Databricks, Snowflake, or Salesforce, and attributes the gap to structural differences in Chinese enterprise buying behavior rather than engineering capability. His view, drawn from running teams in Beijing and Shanghai during his VMware years, is that China's domestic market is large and distinct enough that local companies optimize entirely for it, producing software that the rest of the world cannot easily adopt.

For robotics specifically, Casado argues the US edge in software and in building products with broad global appeal should be decisive, but only if the hardware supply chain question is addressed first. Without domestic capability in robotics components, software leadership alone does not close the gap.

Infrastructure Investing Themes

On the broader a16z infrastructure portfolio, Casado manages a $1.25 billion fund focused on software infrastructure spanning chips, AI models, developer tools, and compute, network, and storage. He is not concerned about hyperscaler competition crowding out startups, noting he cannot identify a single company AWS has successfully put out of business despite copying nearly every product category it enters. His more serious competitive concern is founder-led companies such as Anthropic, OpenAI, Stripe, and Figma, which combine institutional scale with startup hunger.

On AI revenue durability, Casado pushes back on the easy-come-easy-go narrative, pointing to Midjourney, Cursor, OpenAI, and Anthropic as evidence that rapid ramps in this cycle reflect genuine market size and growth rather than fragility. He does acknowledge that AI companies must still develop traditional moats, whether integration lock-in, two-sided network effects, or brand, and that simply reselling inference capacity from Anthropic is not a defensible business. Brand effects, which Casado says he had largely ignored as an infrastructure investor, have re-emerged as a primary adoption driver for the first time since the mid-1990s internet wave, with X now functioning as a primary distribution channel for technical products.