Microsoft AI chief cools AGI hype: Expect 5 to 10 years, not current chips

Sam Altman recently claimed on Reddit that AGI could be achieved using existing hardware. Now, Microsoft's top AI boss is pouring cold water on that timeline.

In an interview on The Verge's Decoder podcast, Microsoft AI CEO Mustafa Suleyman said point-blank that current chips like Nvidia GB200s will not be enough to deliver AGI.

Counting down by chip generations

Suleyman, who co-founded DeepMind and Inflection before taking the helm of consumer AI at Microsoft, sees the road to AGI as fundamentally tied to hardware cycles.

He argues that achieving human-level general intelligence will require between two and five more chip generations. With modern hardware cycles taking 18 to 24 months, that puts AGI somewhere between 5 and 10 years away.

He also pointed out that chip progress is becoming increasingly difficult. We are no longer seeing smooth, linear jumps in compute power and cost efficiency, making absolute predictions ungrounded.

Suleyman also made a clear distinction between AGI and superintelligence. For him, the singularity means an exponentially self-improving runaway system, whereas AGI simply means software that can handle general human tasks.

He expects AI to be capable of handling most human knowledge work within the next five to 10 years, though physical labor in robotics remains a much steeper hill to climb.

Addressing skeptics who claim LLMs just regurgitate data, Suleyman argued that interpolating between concepts in high-dimensional space is itself the beginning of genuine creativity.

Why it matters

For enterprise users and developers, this timeline is a reality check. Suleyman is shifting the conversation away from sci-fi panic and toward practical tooling. Instead of waiting for a magical superintelligence to disrupt everything overnight, the near-term focus is firmly on integrating assistants like Copilot into daily knowledge workflows.

Within the industry, it exposes a fascinating philosophical divide between Microsoft and its closest ally, OpenAI. While Altman has incentives to lower the definition of AGI and claim it is arriving tomorrow, Microsoft has to deliver reliable consumer software on realistic hardware budgets.

We think Suleyman's grounded perspective is a smart move. Tying AI progress directly to the physical limits of chip fabrication avoids overpromising and gives teams a realistic blueprint for building useful tools today.