In a recent analysis, Goldman Sachs has drawn attention to the surging investments by major tech giants aimed at advancing generative AI technologies. The investment banking firm, however, cautions that sustainable business models in AI have yet to solidify.
Goldman Sachs forecasts an expenditure nearing $1 trillion over the next few years, dedicated to critical AI infrastructure such as data centers, semiconductors, and grid enhancements. Despite these substantial financial commitments, concerns linger regarding the uncertain financial returns, even with the development of potential breakthrough applications.
Challenges in Infrastructure and Resource Constraints
Countries at the forefront of AI innovation, including the United States, are confronting critical challenges such as hardware shortages and constraints on power supply. These issues may necessitate a comprehensive overhaul of national infrastructure to meet the escalating demands posed by AI technologies.
Jim Covello, head of global equity research at Goldman Sachs, poses a fundamental question: “What $1 trillion problem will AI solve?” He points out the stark contrast between AI and earlier technological revolutions, highlighting that while the internet facilitated cost-effective solutions like e-commerce, AI replaces low-wage jobs with costly technologies.
Uncertainty in Cost Dynamics and Market Perception
The complexities involved in AI chip manufacturing, compounded by Nvidia’s market dominance, raise doubts about the natural decline of costs in the AI sector. Covello argues that market complacency regarding cost reductions is a significant risk factor.
MIT economist Daron Acemoglu echoes this sentiment, estimating that only a quarter of AI-related tasks will prove cost-effective for automation over the next decade. He projects marginal impacts on US productivity and GDP growth, emphasizing the challenges in achieving widespread economic benefits from AI advancements.
Covello questions the sustainability of the “build it, they will come” approach in AI investments. He warns that if significant AI applications fail to materialize within the next 12-18 months, investor enthusiasm may wane. Nevertheless, he underscores that robust corporate profitability allows for continued experimentation with high-risk, high-reward AI projects, suggesting that AI infrastructure spending is unlikely to decrease in the near future.
Eric Sheridan, a senior equity research analyst, offers a contrasting viewpoint, suggesting that current AI investments should not be directly compared to previous tech investment cycles. He emphasizes that current spending levels by cloud computing firms on AI initiatives are proportionate to past investments that revolutionized enterprise and consumer computing.
Kash Rangan, another Goldman Sachs analyst, outlines the stages of AI development—from infrastructure to platforms and finally applications. He acknowledges that the quest for a transformative AI application is ongoing and underscores the time-intensive nature of this developmental phase.
Impact on Power Grid and Infrastructure
Carly Davenport, senior US utilities equity research analyst, anticipates a significant surge in electricity demand driven by data centers, projecting a compound annual growth rate of 2.4% from 2022 to 2030. This growth necessitates substantial increases in power infrastructure capacity to meet burgeoning data center demands, particularly in states like Virginia, where data center power consumption surged to 2.2GW in 2023.
Hongcen Wei, commodities strategist, highlights Virginia’s role as a data center hub, stressing that such hubs are central to the substantial increase in US power demand expected in the coming years. He warns of the strain this places on existing power grids and the challenges utilities face in adapting to rapid technological advancements.
Former Microsoft VP of energy Brian Janous underscores the unpreparedness of utilities for the swift pace of AI technological advancements. He contrasts the rapid evolution of AI technology, such as the development from ChatGPT 3.5 to ChatGPT 4.0 in just six months, with the prolonged timelines required to build supporting power infrastructures.
Goldman Sachs editor Allison Nathan maintains cautious optimism despite the identified challenges. She suggests that while uncertainties loom over AI investments, the potential for significant advancements in AI applications or the persistence of speculative bubbles offers avenues for continued growth in the AI sector.
