Anthropic investor denies AI bubble but warns of hardware waste

This article summarizes an episode of 20VC’s video series featuring Anjney Midha, founding investor in Anthropic.

Anjney Midha, founding investor in Anthropic/ Photo credit: Anjney Midha

Anjney Midha, founder of AMP and founding investor in Anthropic, argues that business leaders misunderstand AI. Obsessing over software features and hype hides a brutal reality: AI dominance is a war over hardware, infrastructure, and geopolitics.

Testing software in the real world

To build this physical infrastructure, companies must first change how they gather information. AI stops improving when developers rely solely on collecting text from the internet, because current digital information is not enough.

Early attempts to use AI for hard science exposed a severe lack of foundational facts in the training data.

Midha says, “We benchmarked  Claude and Gemini, and surprise: they were terrible at scientific analysis. They were missing a lot of the physics and chemistry data you need to reason about the physical world.”

Overcoming this lack of scientific data means creating systems that learn directly from actual environments. Staying ahead of competitors requires making software interact with physical objects and scoring the results in a closed loop.

“We have LLMs that predict new materials, new superconductors,” Midha explains. “Robots synthesize them, and we have physical machines validate whether they match the predictions. Then we pipe that verification data back into our training run.”

How company structure slows down progress

However, establishing these physical testing loops brings new organizational challenges. Constant real-world updates requires flawless execution. Weak teams, restrictive spending plans, and pushy corporate boards will break the process before the technology itself hits a limit.

Among these organizational challenges, the daily work environment plays a massive role. Midha sees restrictive office culture as the biggest limit on speed.

“Culture actually might be the most important bottleneck,” he argues. “Algorithmic innovation is a function of culture because if you have the right culture, you attract the best researchers who just want to solve the problem.”

Prioritizing the mission over immediate profit
Maintaining this mission-driven culture often means ignoring traditional business advice. True innovation requires financial models that regular businesses would reject.

Midha says their structure allows them to prioritize impact over immediate returns. “We’re giving away most of our compute at cost. A shareholder would say that’s billions in infrastructure given away, but we think it’s the right thing for humanity.”

The false idea of a hardware bubble

Countries control where data lives

Protecting new software from theft