technologyScore 60/100Research
Andrew Feldman: Cerebras achieved 15-18x performance vs. NVIDIA GPU by building large monolithic chip (dinner-plate size) with collocated memory rather than GPU-like architecture
Will Marshall· Planet Labs· Space· 2026-07-19· about Cerebras, NVIDIA (CBRS, NVDA)
“The hard part is moving data from memory to compute. This is the fundamental problem in AI. And we solved it with a way that that very few others had even attempted, which was to build a very big chip and to put memory right next to compute. And by building a big chip, a chip the size of a dinner plate, whereas most chips are the size of a postage stamp, we could use a different type of memory. And by using a different type of memory, a memory that was vastly faster, we opened up all sorts of opportunity. So when OpenAI uses us, we're 15 or 18 times faster than a GPU, right?”
Why it matters
Cerebras demonstrates a fundamentally different architecture wins through memory-compute co-optimization, validating the thesis that monolithic/large-die designs can outperform modular GPU approaches for AI workloads. This challenges NVIDIA's architectural dominance.
Investment implication
NVIDIA faces architectural competition. Customers seeking inference speed (not just throughput) may prefer Cerebras. Advanced packaging (chiplets, UCIe, EMIB) suppliers and memory manufacturers (HBM) that support these designs gain importance. Cerebras capex for wafer supply becomes critical dependency.