Ex-NVIDIA Engineer: Why AI Is About to Get 1000x Cheaper

Summary

Neil Movva, co-founder of Sail Research, explains his strategy to build a 'token factory' that makes AI inference radically cheaper across software, chips, data centers, and power. He argues the next phase is long-horizon background agents rather than low-latency chatbots, which changes chip architecture, data-center design, and energy economics. The discussion covers Nvidia, AMD, memory/HBM, semiconductors broadly, and unorthodox power sources such as solar and wind. In his view, intelligence is becoming a commodity and there will always be demand for cheaper, more abundant inference.

• Sail Research is a token factory aiming to provide the lowest cost per token using open-source models and long-running agents.
• Neil expects agentic inference to shift from low-latency chat toward background workloads running for hours or days, potentially reaching 90/10 in favor of background use.
• He is bullish on Nvidia short-term, sees AMD as underappreciated, and believes HBM/memory is a persistent underpriced bottleneck.
• He argues semiconductor strength is supported by real, non-speculative inference demand rather than a dot-com-style buildout.
• His data-center strategy favors distributed 1MW sites with lower uptime, cheap chips, and underutilized solar/wind power rather than concentrated gigawatt facilities.
• Open-source dynamics and distillation make AI capability diffusion inevitable, and he expects custom, abundant intelligence to become widely available.