Odd Lots: Midha’s Plan to Lower the Price of Compute (Podcast)

Bloomberg’s Odd Lots podcast featured an interview with Omar Midha, a venture capitalist at Lightspeed Venture Partners, on May 29, 2024, where he outlined a strategy to reduce the cost of computing power. Midha proposed that the industry should shift its focus from building larger, more expensive AI models to optimizing existing ones and developing more efficient hardware. He argued that the current trend of escalating compute costs, driven by the demand for ever-larger models, is unsustainable and could stifle innovation. Midha suggested that a more pragmatic approach would involve investing in specialized hardware tailored for specific AI tasks, rather than relying on general-purpose, high-cost GPUs. He also highlighted the potential of open-source models and techniques like quantization and pruning to make AI more accessible and affordable. The venture capitalist pointed to the increasing cost of training and running large language models, citing figures that indicate a significant rise in the expense associated with AI development. Midha believes that by prioritizing efficiency and cost-effectiveness, the AI industry can achieve broader adoption and unlock new applications without the need for continuous, exponential increases in hardware investment. His vision centers on a future where compute power is not a bottleneck but an enabler for a wider range of AI-driven solutions.