Mistral board member and a16z VC Anjney Midha says DeepSeek won’t stop AI’s GPU hunger | TechCrunch
Jan 31, 2025 - techcrunch.com
Andreessen Horowitz general partner Anjney Midha highlights DeepSeek's advancements in AI with its open-source reasoning model, R1, which offers industry-standard performance at a lower cost. Despite these improvements, Midha argues that significant investment in AI infrastructure remains crucial, as seen with Mistral's billion-dollar funding and OpenAI's potential $40 billion raise. He emphasizes the competitive edge of open-source models like Mistral, which benefit from community contributions, unlike closed-source rivals that bear all costs. Midha also underscores the ongoing demand for GPUs, driven by AI model training and inference needs, as evidenced by a16z's overbooked Oxygen GPU sharing program.
Midha advocates for Western nations to adopt Western AI models to ensure "infrastructure independence" from Chinese models, which may pose data and censorship concerns. While some companies have blocked DeepSeek, others, like Intel's former CEO Pat Gelsinger, are building AI services on DeepSeek R1. Midha humorously requests any surplus GPUs from those reconsidering their data center plans due to DeepSeek's efficiency.
Key takeaways:
DeepSeek's R1 model offers industry-standard performance at a lower cost, but AI companies will still invest heavily in GPUs and data centers to maximize compute power.
Mistral remains competitive with larger rivals like OpenAI due to its open-source approach, which allows access to free technical labor and more compute resources.
A16z's Oxygen GPU sharing program is overbooked, highlighting the ongoing high demand for GPUs in AI model training and inference.
DeepSeek's advancements prompt discussions on "infrastructure independence," with a focus on Western nations using Western AI models to avoid reliance on Chinese models.
