Cost is one of the biggest problems facing AI companies. Hyperscalers like Microsoft, Amazon and Google are spending tens of thousands of dollars per quarter to build increasingly larger data centers. Even as the model becomes more efficient, consumer demand is increasing, and new methods rely on more processing power to improve capacity.
Suleyman says Semafor Microsoft AI is already working on the next version of some of the world's largest data centers, featuring Nvidia's next-generation chips known as the GB-200s.
“We have a five-year roadmap that we're investing in after the quarter, so I think that's going to continue.”
Suleyman helped co-found DeepMind, an AI company that was acquired by Google in 2014. After leaving Deepmind, he founded the Fecflection AI. He was hired by Microsoft last year and launched the AI division with many inflection teams.
The Microsoft move created tension in its relationship with Openai, which had agreed to run only AI models in Microsoft's data centers. The launch of its own high-end AI model could create even more competition between the two partners.
Suleyman downplayed the possible Rift. “Our goal is to deepen our partnerships and make sure we're collaborating with Openai for years to come,” he said. “I hope it continues. I'm sure it will.”
Suleyman has been a long-time AI security advocate and recently wrote an essay warning about the dangers of “seemingly conscious AI” that mimics human emotions.
He said Microsoft aims to remove the qualities that make the model appear to have feelings and goals in the “post-training” process.
“We really have to answer questions as sculptors. What is our intention?” he said. “The purpose of my essay is to draw attention to the fact that blind imitation of all human abilities poses material risks. Now is the time to actively think about those risks.”
