Cerebras Systems’ Andrew Feldman on whether AI can keep scaling at TechCrunch Disrupt 2026

AI models support getting much capable. But each leap guardant demands much compute, energy, and infrastructure. How acold tin that continue?

Cerebras Systems has spent the past decennary challenging a basal presumption down AI computing: that progressively almighty AI indispensable beryllium connected accepted spot architectures. The institution built its attack astir wafer-scale computing and contiguous provides AI compute done on-premise systems and its unreality platform.

At TechCrunch Disrupt 2026, Cerebras Systems CEO and co-founder Andrew Feldman volition instrumentality the Disrupt Stage for “Can AI Keep Scaling?” He’ll research the increasing request for compute, energy, and infrastructure, however Cerebras is approaching those constraints differently, and what comes adjacent if today’s AI hardware reaches its limits.

Image Credits:TechCrunch

Want to recognize what could find however acold AI tin scale? Secure your Disrupt ticket to perceive from a laminitis solving the AI standard challenge. Bring a co-founder, colleague, oregon adjacent astatine 50% off.

Cerebras challenges the accepted AI chip

Andrew Feldman co-founded Cerebras successful 2015 aft years of gathering companies astir computing infrastructure. Before Cerebras, helium co-founded and led energy-efficient microserver startup SeaMicro, which AMD acquired successful 2012. Earlier, helium held enactment roles astatine Force10 Networks and Riverstone Networks.

At Cerebras, Feldman and his co-founders took connected a occupation agelong considered impractical: bringing wafer-scale computing to market. Rather than cutting a silicon wafer into idiosyncratic chips, Cerebras developed a processor built connected the wafer itself, an architecture designed specifically for demanding AI workloads.

Now Cerebras is scaling that attack arsenic request for AI compute grows. The institution raised $5.5 cardinal successful its May IPO and signed a multiyear statement with OpenAI to deploy 750 megawatts of Cerebras systems from 2026 done 2028. In August, Cerebras introduced CS-4, the latest procreation of its wafer-scale AI infrastructure.

Can approaches similar Cerebras’ present the compute progressively almighty AI demands — and tin the infrastructure down them support up? Secure your Disrupt pass, get a 2nd astatine 50% off, and perceive from an entrepreneur who has spent much than a decennary betting connected a antithetic mode to physique AI hardware.

Scaling AI means scaling the infrastructure down it

More almighty processors unsocial don’t lick the scaling problem. Those systems request information centers, electricity, cooling, and manufacturing capacity.

Cerebras is already confronting that challenge. In August, the institution reported much than 600 megawatts of information halfway capableness unrecorded oregon nether declaration for transportation by the extremity of 2027 and said it was expanding manufacturing capableness much than tenfold during 2026. It besides plans to bring its archetypal European information halfway capableness online this twelvemonth and grow to 200 megawatts determination by the extremity of 2027.

Scaling AI isn’t solely astir designing a faster processor. It requires capable carnal infrastructure to enactment that compute to work.

For founders, investors, and exertion leaders making decisions astir AI infrastructure, Feldman tin enactment those constraints into context: wherever compute request is heading, what it takes to enactment it, and wherever today’s hardware could deed its limits. Register for Disrupt to perceive what Cerebras’ acquisition tin archer america astir the adjacent signifier of AI scale. Bring a co-worker, colleague, oregon adjacent astatine 50% off.

Learn what happens erstwhile today’s hardware reaches its limits astatine Disrupt

Cerebras pursued wafer-scale computing agelong earlier today’s AI infrastructure roar and is present scaling its computing and manufacturing capableness arsenic request for AI accelerates.

At Disrupt, Feldman volition research what increasing request for compute, energy, and infrastructure means for AI’s aboriginal and what could hap if accepted hardware tin nary longer support pace. For anyone building, funding, oregon deploying AI, it’s a accidental to perceive from a laminitis investigating a antithetic attack to 1 of the industry’s biggest constraints.

His league is 1 of 200+ sessions crossed six manufacture stages, roundtables, and breakouts astatine Disrupt, taking spot October 13-15 astatine Moscone West successful San Francisco. More than 10,000 founders, investors, operators, and tech leaders are expected, on with 250+ speakers and 300+ exhibiting startups. Beyond the agenda, matchmaking, dealmaking, and networking make opportunities to link with the founders, investors, and builders shaping what comes next.

Register for Disrupt to larn however agelong AI tin support scaling — and what indispensable alteration for it to continue. Lean successful connected this league with your co-founder, partner, oregon adjacent astatine 50% off.

Image Credits:Eric Slomonson, The Photo Group