For someone waking up after four years and seeing chip stocks soar—what’s happening, and where does Cerebras fit?
AI has the potential to transform the entire economy, much like electricity or the internet. It threatens to upend a 25-year status quo in digital markets by reshaping how enterprise software is built and delivered, from Salesforce to every major application. That scale of change creates once-in-a-generation opportunities—similar to how the smartphone wave let ARM and Apple capture compute leadership, or how Cisco emerged in enterprise infrastructure in the late 1990s.
That’s why public and private valuations are surging—because the underlying opportunity is “mind-bogglingly large.” In that context, we raised $1.1 billion at an $8.1 billion valuation: AI’s rise isn’t an abstract story; it’s creating immediate demand for new compute. Cerebras exists to meet that demand with purpose-built systems for AI, rather than relying on architectures designed for other workloads.
You just announced a $1.1 billion Series G. How does that position you on the path to an IPO, and what must you prove next?
We raised the round in a few weeks, led by Fidelity with participation from “world-class” investors like T. Rowe, Tiger Global, and Valor—raising the most money at the highest valuation from the best investors. Late-stage rounds before an IPO are common; this capital lets us sustain extraordinary growth while preparing for public markets without slowing our execution.
The IPO path remains arduous—there’s regulatory oversight and shifting market sentiment. Earlier this year, excitement was muted; today, it’s surging. Our aim is unchanged: reach the public markets in the first part of next year. This financing doesn’t distract from that goal; it strengthens our ability to get there while continuing to scale.
Some say there’s a bubble and capacity could be overbuilt. How are you guarding against that?
Bubbles are defined by absent earnings. That’s not what we’re seeing: the largest public companies in AI infrastructure are delivering extraordinary revenue and profit growth. We’ve simply never seen companies this big grow this fast. That reflects AI’s economic potential, which may be closer to electricity in its transformative power than anything in the last half-century.
From our vantage point, demand vastly exceeds our capacity. We’re not “building and hoping”; we are chasing to keep up. It doesn’t feel like a bubble—it feels like the very beginning of the AI revolution. As that beginning gives way to broader deployment, the gap between demand and supply is likely to remain a central dynamic.
Why do some analysts persist with bubble narratives when evidence points the other way?
“Bad news sells,” and some forecasters “predict nine of the last three recessions.” If you call rain every day, you’ll be right eventually, but you’ll miss years of sunshine. In true bubbles, earnings don’t support valuations—the housing crisis being a prime example. Today in AI, there’s tremendous hunger for what’s being built, with value being produced “left and right.”
Economic revolutions are never perfectly smooth, and there will be bumps. But the reflexive “sky is falling” stance often comes from the sidelines. The builders deploying capital, growing companies, and serving customers see durable demand and real returns, not froth detached from fundamentals.
What is the Wafer-Scale Engine 3, and what trade-offs did you make—especially around efficiency and sustainability? How do you see modular/chiplet trends?
Our WSE-3 is the largest chip ever built—about the size of a dinner plate, 56× larger than the next-largest device. AI presents unique computational challenges; by building much bigger, we keep far more data on-chip and move it far less. The result: up to 20× faster performance than alternatives, at a fraction of the power for equivalent compute. That directly addresses two fundamentals of modern AI—speed and energy efficiency.
Getting here took years of innovation and roughly half a billion dollars in R&D. Once we solved the wafer-scale challenges, the performance-per-watt advantages became clear. Users don’t want to wait for answers, and data centers want more compute per unit of power—WSE-3 was designed for both. Rather than retrofitting a graphics architecture, we pursued a clean, AI-first path to deliver the fastest solution, period.
What does your production and deployment roadmap look like over the next one to two years?
We manufacture in the U.S. and are immediately doubling manufacturing capacity. Over the last ten months, we built five U.S. data centers; more are coming online in North America and Europe. We’re scaling manufacturing, expanding data-center footprint, and investing aggressively in the engineering programs that produced our current platform.
The fresh capital accelerates all of this. The aim is straightforward: meet overwhelming demand with more systems, more sites, and more capacity, without compromising the innovation cadence that differentiates us.
What systemic constraints worry you most—power grids, cooling water, HBM/packaging, talent—and is the U.S. positioned to meet AI’s demand?
Decades of neglect have left our power infrastructure less than ideal, and distributed governance complicates coherent policy: federal intent requires state and local alignment. The U.S. has plenty of power—often in the wrong places. Moving electricity is expensive and lossy; regions rich in generation may lack fiber and telco where workloads live. We’re remedying this as a nation, but upgrading the grid and adopting longer-term planning horizons is urgent.
We’re also short on world-class AI practitioners and researchers. Historically, U.S. universities attracted the best global talent—an extraordinary advantage. If policy makes that harder, we must improve K-12 STEM and universities to “mint” extraordinary domestic talent. Power infrastructure and human capital are the two structural limitations on AI’s transformation.
Do current U.S. policies have AI on the right footing—industrial strategy, alliances, capacity enablement?
We’re vastly better off with the current approach than the previous one. Not every policy will please everyone—“I don’t agree with everything my wife does either”—but AI is a top priority, with thoughtful organization and recognition that we must provide AI capacity to allies. Keeping AI top-of-mind and resourcing the whole ecosystem—hardware (NVIDIA, AMD, Cerebras), software (OpenAI, Anthropic), and the application layer—is essential.
This is an ecosystem race, not a single-company contest. The policies in place are helping nourish that ecosystem so it can flourish at home and with partners abroad.
What’s truly at stake in the AI race with China? How big could AI become in the economy?
AI will be a massive economic block—potentially larger than oil and gas, which is about 14% of U.S. GDP today. It’s tightly linked to robotics—AI is how we will train robots—and to adjacent military technologies like airborne and subsea drones. These are foundational capabilities that will change how we live and work and how nations gain advantage and wage war.
While collaboration with China defined earlier decades, we’re now in a race. We’ve seen glimpses in Ukraine, where off-the-shelf drones have forced a strategic rethink. Maintaining leadership in AI and robotics is critical if the U.S. wishes to maintain global leadership.
Take us back to the founding thesis. What did you see in 2015 that’s become reality today?
After selling SeaMicro to AMD in 2012, our founding team reunited in 2015 with two goals on a whiteboard: build something important and work together. We saw AI on the horizon as the biggest wave in technology—bigger than we expected, as it turns out. Traditional systems excelled at numerical calculation but struggled to understand images and language; AI would flip that script.
We believed AI required a new kind of computer, not a remodel of graphics silicon. Starting with a clean sheet, we pursued speed and low power via a bigger chip, solving problems no one had solved in compute. That insight proved right. It required pioneering invention, thousands of patent claims, and close collaboration with partners willing to bet on unprecedented scale—from 800 mm² devices to a 46,000 mm² wafer-scale engine. Those partners have grown with us as the vision has paid off.
Bottom line: where is this all headed in the near term?
AI demand is real, earnings-backed, and still early. We’re doubling manufacturing, adding data centers, and scaling teams to meet “overwhelming” customer pull. The constraints are power and people; both are solvable with focus.
Cerebras was built for this moment: an AI-first architecture that delivers faster answers at lower power, at a time when speed and efficiency define competitive advantage. The opportunity ahead remains, in every sense, generation-defining.
