Frontier Chips for Frontier AI Labs, with Walter Goodwin, Founder/CEO of Fractile
Walter Goodwin, founder and CEO of AI chip startup Fractile, joins Sarah Guo to discuss the technical and commercial landscape of frontier AI inference silicon. Goodwin breaks down why Fractile shifted away from SRAM toward high-bandwidth DRAM, how AI is compressing chip design cycles, and why frontier labs face existential risks by relying purely on proprietary first-party ASICs.
- Hyperscaler ASICs like Google TPU, Meta MTIA, Microsoft Maia, and OpenAI Jalapeno rarely offer differentiated computational performance over Nvidia or AMD. Read →
- Over the past 20 years, raw compute FLOPs scaled 1,000,000x, while memory bandwidth scaled only 40x. Read →
- Physical fabrication enforces a strict floor of 3 to 5 months from tape-out to physical chip delivery, even when running expensive super hot lots at foundries. Read →
- Big tech companies rely heavily on ASIC houses like Broadcom, a $2 trillion company, to handle the physical layout and packaging of custom chips. Read →
- Fractile spent its first two years building on-chip SRAM chips like Groq and Cerebras before abandoning the architecture entirely. Read →