Beyond the Chip: Andrew Feldman and the Future of AI Infrastructure at TechCrunch Disrupt 2026

The relentless pursuit of more capable artificial intelligence has triggered a seismic shift in the technology sector, fundamentally altering how we think about compute, power, and infrastructure. As AI models grow increasingly sophisticated, the appetite for high-performance hardware has reached a fever pitch, leading many industry observers to ask a defining question: How much longer can this trajectory of exponential scaling actually continue?

At the upcoming TechCrunch Disrupt 2026 conference, scheduled for October 13-15 at the Moscone West in San Francisco, this critical tension will take center stage. Andrew Feldman, the CEO and co-founder of Cerebras Systems, is slated to headline a high-profile session titled "Can AI Keep Scaling?" During his appearance, Feldman will dissect the immense pressures mounting on the global AI infrastructure landscape and discuss how his company is challenging the industry’s reliance on conventional chip architectures.

Challenging the Conventional AI Chip

Cerebras Systems has spent the better part of a decade operating with a contrarian philosophy. Since its founding in 2015, the company has operated under the assumption that the traditional path of linking thousands of individual GPUs together is not the only—or even the most efficient—way to handle the massive, complex calculations required by modern large language models.

Feldman, a veteran of the computing infrastructure world, brought a wealth of experience to this challenge. Before launching Cerebras, he co-founded SeaMicro, a startup that pioneered energy-efficient microservers and was subsequently acquired by AMD in 2012. His background in network infrastructure and server design—including leadership roles at Force10 Networks and Riverstone Networks—provided him with a unique vantage point on the bottlenecks inherent in standard data center design.

At Cerebras, the goal was audacious: to bring "wafer-scale computing" to the mainstream. Rather than following the conventional semiconductor process of cutting a silicon wafer into hundreds of individual chips, which must then be packaged and networked together, Cerebras engineers designed a processor built on the entire wafer itself. This massive, monolithic piece of silicon acts as a single, ultra-fast engine designed specifically for the rigorous demands of deep learning and AI training.

This unconventional approach has gained significant momentum recently. In May 2026, the company successfully completed a $5.5 billion IPO, signaling a massive vote of confidence from the market. Furthermore, Cerebras has secured a multiyear, high-stakes partnership with OpenAI, committing to deploy 750 megawatts of specialized compute power between 2026 and 2028. This partnership, alongside the launch of the company’s latest generation of hardware—the CS-4—places Cerebras in the heart of the debate over how to sustain the AI boom.

Scaling AI Means Scaling the Infrastructure Behind It

While processor speed and architectural design are critical, Feldman has long argued that they are only one part of a much larger, more precarious equation. The reality of modern AI development is that a faster chip is useless without the physical foundation to support it. As AI systems become more energy-intensive, the true bottleneck is increasingly shifting toward the physical limits of our power grids, cooling technologies, and industrial manufacturing capacity.

Cerebras is currently navigating these challenges on a massive scale. As of August 2026, the company reported that it has more than 600 megawatts of data center capacity either currently operational or under contract to be delivered by the end of 2027. This level of growth requires an extraordinary logistical effort, and the company has signaled that it is aggressively scaling its manufacturing capabilities by more than tenfold throughout 2026 to keep pace with demand.

Geographic expansion is also a priority. Recognizing that the demand for AI compute is a global phenomenon, Cerebras has initiated plans to bring its first European data center capacity online later this year, with an ambitious target of reaching 200 megawatts of capacity in the region by the end of 2027.

These initiatives underscore a reality that many software-focused startups often overlook: building the future of AI is, at its core, an industrial challenge. For the thousands of founders, venture capitalists, and corporate leaders who will gather at TechCrunch Disrupt, Feldman’s session is designed to offer a sobering look at what it actually takes to provide the "fuel" for the AI revolution. By framing these challenges in terms of energy consumption, thermal management, and raw manufacturing output, Feldman aims to provide clarity on where the industry is heading and, more importantly, where it might hit a wall.

What Happens When Hardware Reaches Its Limits?

The central question remains: what happens when the current paradigm of AI hardware reaches its physical or economic limits? Conventional approaches are currently consuming massive amounts of power and space, and there is growing concern that the energy requirements of next-generation models may outstrip the ability of current infrastructure to provide reliable, cost-effective support.

Feldman’s appearance at Disrupt is not merely a product showcase; it is a strategic inquiry into the viability of the AI industry’s current growth trajectory. He intends to explore the consequences of the current hardware arms race and whether alternative approaches, such as those championed by Cerebras, can provide the necessary efficiency to keep the scaling trend alive.

For those in the audience, this is an opportunity to learn from a founder who has spent over a decade betting against the grain. Whether one is building a foundation model, investing in the next generation of AI startups, or managing the IT strategy for a global enterprise, the underlying infrastructure of AI is becoming the primary constraint on success.

TechCrunch Disrupt 2026 expects to host over 10,000 attendees, including founders, investors, and operators from across the global technology ecosystem. With more than 250 speakers across six stages, the event is positioned as a hub for the most pressing conversations in the industry. Beyond the main stage, the event will feature over 300 exhibiting startups, alongside various roundtables and breakout sessions that facilitate the kind of networking and dealmaking that often defines the industry’s next big shift.

As the industry grapples with the high costs of innovation, the conversation around compute efficiency and physical infrastructure is moving from the back office to the boardroom. Andrew Feldman’s session represents a critical touchpoint in this transition. By addressing the logistical and physical realities of AI, he invites participants to look past the software headlines and consider the silicon and power foundations that make the modern intelligence age possible.

For those interested in the future of the field, the insights shared on the Disrupt stage will likely prove essential for understanding how the next phase of the AI cycle will play out. As the conference approaches, the broader industry waits to see whether the promise of continued scaling will hold firm or if the infrastructure constraints of 2026 will force a fundamental redesign of how we build the future of technology. Attendees are encouraged to register and engage with these themes directly, as the lessons learned at this year’s event may well dictate the competitive landscape for years to come.