Inside Benchmark's AI Bets | Eric Vishria

Summary

Benchmark General Partner Eric Vishria discusses how AI is reshaping software and infrastructure, drawing parallels to the cloud era. He explains why running AI inference is surprisingly hard (giving Fireworks a moat), how the competitive frontier for SaaS has shifted, why energy is a critical bottleneck for intelligence, and why he believes multiple giant winners will emerge across AI infrastructure, chips, and applications. He also shares lessons from Benchmark's investment in Cerebras, the state of robotics, and his venture investing philosophy.

• AI inference is extremely difficult to run efficiently, creating room for specialist providers like Fireworks.
• The cloud era showed that AWS did not eat everything; an oligopoly and many hundred-billion-dollar independents emerged, and the same pattern is likely in AI.
• Legacy SaaS companies must invert their approach, because sticking to the old plan destroys equity value as the competitive frontier changes.
• Enterprise AI adoption is faster than cloud adoption was, with blue-chip companies eager to experiment and a desire for 'AI sherpas' to guide them.
• Energy is the key bottleneck for compute; China is bringing on far more capacity, and underinvestment in the US could lead to more expensive intelligence.
• Cerebras's wafer-scale chip architecture addresses AI's communication-bound problem, making it a rare hardware winner.
• Robotics is following an LLM-like training pipeline with vertical integration, and early demonstrations like laundry folding suggest real progress.
• Benchmark is expanding into growth investing because larger outcomes are creating cash-on-cash opportunities beyond early stage.