According to BlockBeats, citing Anthropic investor Deedy, AI chip startups are pursuing six distinct approaches to challenge Nvidia's GPU supremacy by addressing fundamental data movement inefficiencies. These routes include eliminating DRAM (Groq, acquired by Nvidia for $20 billion), interconnects (Cerebras, valued at $48 billion), compute-memory separation (d-Matrix), server-centric architecture (Majestic), generality (Etched, Taalas, MatX), and lithography costs (Substrate). Etched's valuation doubled to $10.3 billion this week, underscoring rapid capital repricing in the sector.
The shift reflects a fundamental strategic pivot: next-generation AI chip competition centers on data movement efficiency rather than raw compute peak. Nvidia's acquisition of Groq signals the company recognizes this transition. Among 18 major startups, private companies hold ~$58 billion in paper value while public market capitalization totals ~$48 billion, spanning inference, training, systems, foundry, and lithography segments.