BlockBeats news, on July 26, Anthropic investor Deedy pointed out that all startups developing the next generation of AI chips are attempting to challenge NVIDIA’s dominance in various ways, targeting the fundamental issue of data movement. The six differentiated approaches include: eliminating DRAM (Groq, acquired by NVIDIA for $20 billion), eliminating interconnects (Cerebras, already listed with a market cap of approximately $48 billion), eliminating the separation of computation and memory (d-Matrix), eliminating server architectures centered on computation (Majestic), eliminating generality (Etched, Taalas, MatX), and eliminating the $400 million lithography machines (Substrate).
The listings and acquisition prices of Cerebras and Groq have established a pricing benchmark for the entire AI chip sector, while Etched’s valuation doubled to $10.3 billion this week—surging from obscurity in just three weeks—further confirming that capital is rapidly re-evaluating valuations in this space. Among the 18 leading startups, the combined paper value of private companies totals approximately $58 billion, while public market capitalization stands at around $48 billion, covering multiple sub-segments including inference, training/new architectures, systems, and wafer fabrication and lithography.
Each approach challenges the core assumptions of NVIDIA’s GPU architecture—that moving data between computing units and memory consumes both time and energy. Groq eliminates memory hierarchy latency by replacing DRAM with SRAM, Cerebras removes inter-chip interconnect bottlenecks with wafer-scale chips, d-Matrix performs computation directly within memory, and Etched sacrifices generality to design hardware exclusively for the Transformer architecture. The AI chip startup ecosystem has evolved from a single narrative of replacing GPUs to a multi-pronged architectural assault. NVIDIA’s acquisition of Groq signals its deep understanding that the next frontier of competition is no longer peak compute power, but a fundamental reimagining of data movement efficiency.