Profile of Cerebras, which made the world's largest chip by using a “wafer-scale” approach that offers one possibility for AI chips to keep up with Moore's law
In the race to accelerate A.I., the Silicon Valley company Cerebras has landed on an unusual strategy: go big.
New YorkerMatthew Hutson
Context & Ripple Effects
Cerebras spent four years going from a little-known Los Altos startup that had raised $112M to the owner of the world's largest chip — a mousepad-sized device unveiled in 2019 with 400K cores, 1.2T transistors and 18GB of on-chip SRAM carved from an entire silicon wafer. The New Yorker profile lands after Cerebras turned that chip into a shipping product: the CS-1 system, already deployed at Argonne National Lab for basic research.
The piece matters because it frames wafer-scale integration as one credible path for AI chips to keep pace as Moore's law slows — a bet that was validated when Cerebras later trained a 20B-parameter language model on a single device, a record no clustered system could claim at that point.
First-order effects
Argonne National Lab gets AI training throughput from a single CS-1 appliance rather than a cluster, collapsing the coordination overhead that multi-chip systems carry.
Cerebras stakes its commercial identity on one extreme architecture, making every customer win or loss a referendum on wafer-scale versus conventional GPU clusters.
Second-order effects
AMD's move to integrate its Helios server racks with Cerebras wafers for simultaneous operation pulls wafer-scale silicon into the standard x86 server ecosystem, giving buyers a hybrid option instead of an all-or-nothing platform choice.
Cerebras's reported Q1 revenue of $193.4M (up 94% YoY) with net losses narrowing 41% to $14M suggests the niche can fund itself while incumbents respond — pricing pressure now comes from a vendor selling whole-wafer capacity rather than per-chip margins.
Third-order effects
If the pattern holds, AI accelerators bifurcate structurally: dense clusters of small commodity chips versus monolithic wafer-scale devices, with buyers choosing based on model size and latency tolerance rather than a single dominant architecture.
OpenAI routing its ultrafast API tier through Cerebras signals that hyperscale AI workloads are willing to diversify beyond incumbent GPU suppliers — the precondition for a genuinely competitive accelerator market as Moore's law flattens.
The trend: As transistor scaling slows, AI compute is splitting between distributed clusters of small chips and monolithic wafer-scale designs like Cerebras's, each competing to absorb the training and inference demand Moore's law can no longer serve alone.
.@NewYorker looks back at our history & how fearless engineering enabled us to overcome the impossible, creating the world's first wafer-scale engine. It powers our leading CS-2 system - addressing the most important workload of this decade. Read here: https://www.newyorker.com/.…
Truth! “To be totally honest with you, for me, ignorance was an advantage,” Vishra said. “I don't know that, if I'd understood how difficult it was going to be to do what they did, I would have had the guts to invest.” 😂😂 https://www.newyorker.com/...
#AI for chip design is here, even for the world's largest computer chip! In a recent interview with Matthew Hutson from @NewYorker Synopsys Chairman and co-CEO Aart de Geus explains how AI software enables @CerebrasSystems' mega-chip design to “go bigger.” https://snps.social/...…
People in A.I. circles speak of the singularity—a point at which technology will begin improving itself at a rate beyond human control. With the development of the world's largest computer chip, is it safe to say that we are living in that moment? https://nyer.cm/oP3Ee6q
“A typical computer chip is the size of a fingernail. Cerebras's is the size of a dinner plate. It is the largest computer chip in the world.” https://www.newyorker.com/...
“Companies have tried to build mega-chips in the past and failed; Cerebras's plan amounted to a bet that surmounting the engineering challenges would be possible, and worth it.” @TheNewYorker's deep dive on the history & innovation of Cerebras: https://www.newyorker.com/...
'The computer looks like what you'd get if you bought a dehumidifier at the moma store; in a rack in a data center, it takes up the same space as fifteen of the pizza-box-size machines powered by G.P.U.s.' https://www.newyorker.com/...
In the evolution of computers, the big brains have arrived. The new computer chips specialized for artificial intelligence are as large as a table. They are designed in Silicon Valley, and manufactured by TSMC. @GuodongNian @SammyHassan_ https://www.newyorker.com/...
A typical large computer chip might draw 350 watts of power, but Cerebras's giant chip draws 15 kilowatts—enough to run a small house. “Nobody ever delivered that much power to a chip,” the company's co-founder said. https://nyer.cm/cSMesmv