Cerebras CEO pushes back on AI doomsday warnings

We have reached a weird point in the cycle where the people building the most powerful technology in human history are also the ones telling us it might kill everyone. It is a strange marketing strategy, and Andrew Feldman, the CEO of Cerebras, has finally had enough of it. At a recent industry gathering, he laid out a perspective that founders need to hear: the existential dread is becoming a distraction from the actual work.

The Problem with Doomsday Marketing

For the last two years, we have seen a parade of AI researchers and big-tech CEOs signing open letters and testifying before government bodies about the potential for artificial intelligence to trigger a global catastrophe. They talk about rogue agents, biological warfare, and the end of work. Feldman’s take is refreshingly blunt. He suggests that if these leaders truly believed the technology was that dangerous, they probably shouldn't be building it in the first place.

From a founder’s perspective, this doomsday talk often feels like a regulatory moat. If you convince the world that AI is a nuclear-level threat, you ensure that only the massive incumbents with the budget for massive compliance departments are allowed to play. Feldman is pushing back against this narrative because it creates a climate of fear that stifles the practical, beneficial applications we are actually capable of building today.

Hardware Realism vs. Software Fantasy

Cerebras is in the business of building massive chips. They deal with the physical reality of compute—silicon, power, and throughput. When you are focused on the hardware layer, the conversation is about efficiency and solving hard problems in medicine, physics, and engineering. It is not about sci-fi scenarios. Feldman’s frustration stems from the fact that the industry is drowning out these tangible wins with hypothetical nightmares.

For builders, this is a signal to refocus. The market is currently split between people chasing AGI (Artificial General Intelligence) and people building tools that work. The AGI crowd loves the doomsday talk because it makes their projects feel more significant. But for those of us trying to use AI to improve supply chains, diagnose diseases, or write better code, the fear-mongering is just noise.

The Innovation Tax

Every time a major CEO goes on a news program to talk about the 'extinction risk' of AI, they are essentially asking for more oversight. While some oversight is healthy, the current trajectory is leading toward a massive innovation tax. Small teams and startups are the ones who suffer when the regulatory environment is built around the fear of a robotic uprising rather than the reality of data privacy or algorithmic bias.

Feldman argues that the industry’s focus on these far-off risks is a disservice to the potential benefits. We are looking at a technology that can drastically accelerate drug discovery and energy efficiency. By focusing on the 1% chance of a disaster, we are slowing down the 99% chance of significant human progress. This is the 'builder-first' mindset: identify the risk, mitigate it, and keep moving. You don't stop the train because someone thinks the tracks might end in a hundred miles.

Why Founders Should Ignore the Hype

If you are building in the AI space right now, you are likely feeling the pressure to have an 'AI Safety' stance. While safety is important, it shouldn't be the core of your identity unless you are specifically a safety researcher. Feldman’s message is a reminder that our job as founders is to create value. The hardware is getting faster, the models are getting more efficient, and the real-world utility is there for the taking.

The skepticism we see from leaders like Feldman isn't a denial of risk; it's a denial of the theater. The doomsday narrative has become a performance. It allows companies to look responsible while they continue to scale as fast as possible. As builders, we should be wary of any narrative that suggests the technology is too powerful for the average person to handle.

The Practical Path Forward

So, where does that leave the rest of us? We need to look at the compute. The bottleneck for AI isn't a lack of fear; it's a lack of accessible, high-speed hardware and clean data. Cerebras is pushing the envelope on the hardware side because they believe the upside is worth the effort. They aren't waiting for a permission slip from a safety committee that is worried about a computer gaining consciousness.

We should be focusing on the following areas instead of the apocalypse:

• Efficiency: How do we get more output from less power?
• Vertical Integration: How do we build models that solve specific industry problems rather than general-purpose toys?
• Data Integrity: How do we ensure the models we build are based on high-quality, non-hallucinatory information?

These are the problems that Feldman and other hardware-centric leaders are worried about. They are the problems that actually matter for the next five years of development.

Takeaway for Builders

Stop apologizing for building powerful tools. The existential dread coming from the top of the food chain is often a strategic move to consolidate power. If you are building something that helps people or solves a technical debt, you are doing the real work. Don't let the doomsday marketing distract you from the fact that we are currently in the most productive era for software and hardware integration in history. The best way to prove the doomsdayers wrong is to build something so useful that the benefits become undeniable.

Read the original at TechCrunch AI →