← Home
Invest Like the Best · August 25, 2026 · 1h 18m

Neil Movva - Making AI 10x Cheaper - [Invest Like the Best, EP.488]

Neil Movva
Show Notes

Neil Movva, founder of Sail, joins Invest Like the Best to chart the future of AI inference. Sail is a 'token factory' built for a world dominated by long-running autonomous agents, where cost per token matters far more than real-time latency. The conversation tours the full stack of intelligence — chips, software, and power — exploring Movva's 'scavenger strategy' for buying overlooked compute and energy, why the premium frontier labs charge for being months ahead may collapse, and his contrarian bet against NVIDIA's dominance as the market shifts from training to persistent, cost-sensitive inference workloads.

This summary was generated from show notes and public descriptions, not from a full transcript review. Details may contain inaccuracies.

Novel

•
The Growing Irrelevance of the Frontier Model Premium01:12:42
The premium pricing that frontier AI labs charge for being three to six months ahead in model capability is unsustainable as models commoditize and the market shifts to optimizing cost and reliability for specific agentic tasks.
ClaudeChatGPTGemini
•
The Contrarian Case Against NVIDIA for Inference01:12:42
As the AI industry transitions from a training-dominated capex cycle to a world of persistent, cost-optimized inference, NVIDIA's general-purpose GPU architecture will be increasingly challenged by specialized, lower-cost chips designed specifically for running models.
ClaudeChatGPTGemini

Highlights

•
The Rise of the Token Factory00:03:22
Sail is building a 'token factory' — an inference company optimized for a future where AI agents run autonomously for hours or days, prioritizing radical cost reduction over low latency.
ClaudeChatGPTGemini
•
Long-Running Agents Will Dominate Future Workloads00:05:32
The next phase of AI will be defined by 'long-running agents' performing tasks like deep research and persistent cybersecurity over extended periods, shifting the primary infrastructure demand from speed to cost-efficiency.
ClaudeChatGPTGemini
•
The Full Stack of Intelligence: Chips, Software, and Power00:15:12
To achieve a 10x reduction in the cost of AI tokens, a company must simultaneously reinvent the chip architecture, the low-level software stack, and the power sourcing strategy.
ClaudeChatGPTGemini

Misc

✧Sail is explicitly not an AI model company — it is an inference company designed to produce tokens as cheaply as possible for long-running agent workloads.
✧Movva actively seeks out chips and power sources that other, latency-obsessed buyers reject.
✧The core bet is that a large future market exists for slower, dramatically cheaper token generation where the AI works in the background for hours or days.
Was this useful?