a16z Podcast

Three Startups Reinventing Critical Infrastructure

Aug 5, 2026
Listen Now

Summary

This episode spotlights three startups rebuilding foundational pieces of critical infrastructure: ocean robotics, critical minerals processing, and next-generation nuclear power. Ulysses is developing autonomous surface and underwater robots for infrastructure inspection, conservation, and defense, with a broader vision of making the ocean a more accessible frontier. Mariana Materials argues that mining and refining must be treated as a vertically integrated system, and that software, plant automation, and reinforcement learning can make mineral extraction and processing far more efficient. The episode also highlights how geopolitical pressure, supply-chain fragility, and a new generation of technical founders are making these once-hard industrial bets newly viable. Across all three companies, the conversation centers on how deep-tech startups can modernize essential systems that were previously slow-moving, underinvested, or strategically constrained.

Key Takeaways

  • 1Autonomous ocean robotics could open up major opportunities in inspection, conservation, and defense.
  • 2The ocean has remained an underexplored frontier partly because public ambition and technical momentum faded.
  • 3Critical minerals require a vertically integrated approach because mining, refining, and deployment are tightly coupled.
  • 4Reinforcement learning and automation can materially improve refinery and plant performance.
  • 5Geopolitics, supply-chain fragility, and new technical talent are making hard infrastructure startups more investable.
  • 6Copper and other minerals face a growing supply challenge driven by rising demand, declining ore grades, and permitting bottlenecks.

Notable Quotes

""We build autonomous robots. That solved that problem.""

""There's like single digit thousand autonomous underwater vehicles or underwater drones made every year.""

""The ore's will come to the surface, they'll be less than 1%, less definitely less than 5% concentration unless you have less word, world class deposit.""

""They were able to reduce energy consumption by 30%.""

""We're just going to need an insane amount of copper.""

""If you're exploring over more than a five-acre parcel, you have to submit like a plan of record or plan of operations that needs to be approved by the BLM before you can start to expand and explore over a larger piece of land.""

""And so by July 4th, several companies will have reactors built that go critical that are fundamentally new designs, completely new and from scratch, but it hasn't happened yet.""

Episode questions

What are Ulysses’ main use cases for underwater robots?

They span conservation, commercial infrastructure inspection, and defense. Examples include reef mapping, underwater seeding/replanting, inspecting wind platforms and subsea cables, and supporting tasks like mine removal and surveillance.

Why does the speaker believe ocean exploration lagged behind space?

He argues the ocean lost cultural momentum after the 1960s and became more classified and less publicly inspiring. As a result, fewer founders and workers were drawn into building ocean tech.

Why does Marianna Materials think mining is a software-and-operations problem, not just a physical extraction problem?

Because every mine has different ore grades, impurities, and process conditions, the flow sheet must be custom-designed and constantly adjusted. The company believes software, automation, and RL can help operators run these systems more efficiently.

Why does the founder think commissioning refineries is so slow today?

The transcript says changes in one part of the circuit may not show up for 24-48 hours, creating high latency and slow iteration. That makes it difficult to reach spec quickly, and some Western refineries reportedly take two to four years versus much faster timelines elsewhere.