The Neon Show Why OpenAI Has No Moat & LLM Prices Will Collapse | Avinash Lakshman, Weilliptic
Sep 18, 2026
Avinash Lakshman, the engineer behind Amazon Dynamo, Apache Cassandra, and Hedvig, explores the infrastructure shaping autonomous AI. He discusses why open-weight models could commoditize LLMs, inference consolidation, Python’s data-center problem, Rust and WASM security, sovereign multi-cloud compute, immutable audit trails, and how a tiny team can build systems for thousands of AI agents.
AI Snips
Chapters
Transcript
Episode notes
Inference Providers Will Consolidate
- Avinash Lakshman expects inference providers to consolidate as specialized offerings become standardized and larger companies absorb smaller players.
- He compares the pattern with search, where many providers eventually narrowed to a handful of dominant platforms.
Python Could Become AI Infrastructure’s Bottleneck
- Agent proliferation could create a hidden data-center crisis because Python consumes far more power than compiled alternatives such as Rust.
- Avinash estimates ten million agents could require 100 megawatts in Python versus roughly 10 megawatts in Rust.
Build Efficient Sandboxed Agents
- Build agents with efficient compiled languages and sandbox them to control power use, security exposure, and unintended behavior.
- Avinash recommends Rust alongside WebAssembly, which provides isolation and guardrails for agent execution.