Chris Kelly, Facebook’s former chief privacy officer, said Tuesday that the next phase of the artificial intelligence boom will focus on improving efficiency.
As major AI players race to churn out the infrastructure needed to support AI workloads, Kelly told CNBC’s “Squawk Box” the industry needs to streamline power-hungry builds.
“Our brains run on 20 watts. We don’t need gigawatt power centers to reason,” Kelly says. “I think finding efficiencies will be one of the key things that the big AI players will focus on.”
Kelly, who was also Facebook’s general counsel, added that AI winners will emerge from companies that can make breakthroughs in reducing data center costs.
The data center market will accumulate more than $61 billion in infrastructure deals in 2025 as hyperscalers enter a global construction boom, according to S&P Global.
OpenAI alone has more than $1.4 trillion in AI commitments over the next few years, including large-scale partnerships with GPU leaders. Nvidia and infrastructure giant oracle and coreweave.
But the data center frenzy is raising concerns about where the power to support these increases will come from, at a time when power grids are already strained.
In September, Nvidia and OpenAI announced a project that would include at least 10 gigawatts of data centers, roughly equivalent to the annual electricity consumption of 8 million U.S. homes.
10 gigawatts is also about the same amount as New York City’s summer peak demand in 2024, according to New York Independent Operators.
Cost concerns were further heightened when DeepSeek launched a free, open source, large-scale language model in December 2024 for less than $6 million. The company claims this is significantly cheaper than its US competitors.
Kelly said he expects “more Chinese companies to emerge,” especially after President Donald Trump recently approved the sale of Nvidia’s H200 chip to China.
Open source models, especially outside of China, will give people access to “base-level computing” and generative and agentic AI, Kelly added.
