Goldman Sachs anticipates a 50% increase in data center capacity by 2027, driven by expanding AI workloads. This surge is expected to double the sector’s energy consumption by 2030. Despite high expectations, there are concerns that AI’s rapid growth may not match the current enthusiasm.
The tech industry is on high alert as AI demand prompts substantial investments amid fears of being outpaced. Goldman Sachs’ Eric Sheridan describes this as a frenzy of activity with companies preparing for both defense and offense.
Currently, data centers operate at around 62 gigawatts of capacity, with AI accounting for an emerging 13%. Projections indicate AI will occupy 28% of capacity by 2027. While traditional and cloud workloads continue to grow, AI’s accelerated rise shifts the balance in this expanding market.
Investment trends align with studies from Omdia, noting infrastructure gains rivaling mid-scale economies. For instance, Amazon’s annual investments approximate Costa Rica’s GDP.
The implications for the semiconductor industry are vast, with potential revenues doubling to over $1 trillion by 2030. Counterpoint Research highlights that AI servers will be crucial to this boom, emphasizing advancements pushed by hyperscalers.
Significant hardware changes are expected in AI training, with systems evolving to house 576 GPUs per rack, consuming energy equivalent to 500 homes.
Looking ahead to 2030, global data center power consumption is predicted to rise by 165%, reaching up to 4% of total electricity usage worldwide. Renewable energy is set to cover 40% of the new demand, while natural gas will handle 60%, potentially raising emissions by 0.6% globally.
While the outlook remains positive, Goldman Sachs advisors maintain caution, watching for underperformance risks and the commoditization of AI models.
Goldman Sachs outlines a base growth scenario of 17% CAGR in data center capacity, with variations from 14% to 20% depending on AI’s market traction. Analysts and industry leaders, including OpenAI’s Sam Altman, acknowledge the presence of an AI bubble, urging caution in forecasting AI-driven demand.