Groq CEO Jonathan Ross says ChatGPT suffers from two problems

The smartest way is to create themed note systems. Ask ChatGPT to organize notes by categories like work ideas, pitch drafts, interview points, product observations or weekend plans. It can rewrite messy inputs into smooth paragraphs, making your notes ready to reuse without extra editing.

Jonathan Ross, the founder and chief executive of Groq and a former senior engineer at Google, has offered a pointed critique of OpenAI’s ChatGPT, identifying what he believes are the two biggest friction points holding back advanced AI use. Ross, who is set to join Nvidia after a reported $20 billion technology licensing agreement, argues that the future of AI hinges on speed and scale, not feature parity or hardware imitation.

Speaking in a recent discussion shared online, Ross made it clear that his criticism is not about capability but about experience. He described ChatGPT as impressive, yet constrained by infrastructure choices that slow down power users and researchers. According to Ross, the first major issue is latency, particularly during deep or complex research tasks. When queries take several minutes to return results, it breaks the natural flow of inquiry and limits how users can build on previous answers.

Ross explained that waiting ten minutes for a response effectively stalls human curiosity. During that time, users cannot ask follow-up questions or explore adjacent ideas. For researchers and developers, that delay compounds quickly, turning what should be an interactive process into a stop-start experience that feels inefficient and frustrating.

The second problem Ross highlighted is what he calls the rate-limit barrier. These limits, designed to prevent system overload and ensure fair access, restrict how frequently users can send requests within a given timeframe. While understandable from an operational standpoint, Ross sees them as another obstacle for intensive workflows. For users trying to iterate rapidly or explore large problem spaces, rate limits slow momentum and cap productivity.

Together, these two issues point to a deeper concern Ross has with the current AI ecosystem. In his view, the industry is bottlenecked not by model intelligence, but by the speed at which those models can respond and scale under heavy demand. That is where Groq’s philosophy diverges sharply from traditional approaches.

Ross has consistently argued against direct competition with established hardware leaders. He does not see value in trying to outdo Nvidia by copying its GPU-based model. Instead, he believes competition of that kind wastes research and development resources. In his words, replicating what already exists diverts attention from solving problems that have not yet been addressed.

For Ross, the goal is real-time human-AI collaboration. That requires inference speeds fast enough to feel conversational and continuous, rather than batch-based and delayed. Groq’s language processing unit architecture is designed around that premise, focusing on predictable, low-latency output rather than sheer throughput alone.

His move to Nvidia signals a broader convergence in the AI hardware landscape. As demand for large-scale AI systems accelerates, even the most successful platforms are running into physical and economic limits around compute, memory and power. Ross’s critique of ChatGPT reflects a growing industry-wide realisation that the next competitive advantage will come from responsiveness and efficiency, not just smarter models.

While OpenAI continues to push the boundaries of model capability, voices like Ross’s suggest that users will increasingly judge AI systems by how fluidly they can think alongside humans. In that context, latency and rate limits are no longer minor inconveniences. They are strategic constraints that could shape which platforms dominate the next phase of AI adoption.