“Gen AI: too much spend, too little benefit?”: Key Insights from Goldman Sachs' AI Investment Report

Goldman Sachs' latest report, “Gen AI: too much spend, too little benefit?” examines the surge in $1 trillion in AI investments over the next decade and the likelihood of meaningful ROI. We used Brightwave's advanced analysis engine to distill key insights from the 31-page document in one seamless pass, connecting dots across topics like AI market dynamics, infrastructure investments and impacts on the energy and utilities sectors to create every word of the following report. 

Here's what you need to know to stay ahead:

Key Takeaways

• AI infrastructure is projected to cost more than $1 trillion in the coming years, raising questions about the return on such a massive investment.
• Goldman Sachs projects a 15% increase in U.S. labor productivity and GDP growth from generative AI, assuming widespread adoption.
• MIT's Daron Acemoglu provides a conservative estimate, expecting AI to boost U.S. total factor productivity by only ~0.5% and GDP by ~1% over the next decade.
• Hyperscalers have invested $60 – 80 billion in AI infrastructure, yet a definitive “killer application” for AI has not been identified.
• U.S. electricity demand is expected to rise at a 2.4% compound annual growth rate (CAGR) from 2022 to 2030, with data centers accounting for roughly 90 bp of that growth.

Risks

Rising Power Demand Challenges AI Expansion

• Data centers in the U.S. are expected to more than double their electricity usage by 2030, increasing their share of total U.S. power demand from approximately 3% to 8%. This surge translates into a 15% CAGR in data-center power demand from 2023 to 2030.
• The existing U.S. power grid, which has not seen significant growth in electricity consumption for nearly two decades, may struggle to keep up with the rapid increase in power demand driven by AI data centers.
• The swift advancement of AI technologies is in sharp contrast with the lengthy timelines needed to construct essential power infrastructure. This discrepancy is exacerbated by regulatory delays, interconnection issues and supply chain constraints, leading to grid connection wait times of 40 to 70 months. Consequently, there has been a nearly 30% increase in the backlog of power projects awaiting grid connection over the past year.
• In Europe, AI data centers are expected to consume energy equivalent to the combined current power usage of the Netherlands, Portugal and Greece, posing substantial risks to the stability and capacity of European power grids over the next decade.

Overestimation of AI Capabilities Risks Market Corrections

The economic impact of AI in 2024 has sparked significant debate, with projections varying widely among experts. Daron Acemoglu forecasts a modest ~0.5% increase in productivity and ~1% increase in GDP over the next decade, contrasting sharply with Goldman Sachs' more optimistic estimates of a 9% increase in productivity and 6.1% increase in GDP. Acemoglu's skepticism stems from his belief that generative AI will primarily enhance the efficiency of existing production processes rather than drive transformative changes. He estimates that AI will impact only 4.6% of all tasks within the next ten years, based on the assumption that only 23% of AI-exposed tasks will be cost-effective to automate. This conservative view is supported by studies indicating that AI-related cost savings will be around 27%, leading to a total factor productivity effect of 0.53%.

In contrast, Goldman Sachs' Joseph Briggs is more bullish. He predicts that generative AI will automate 25% of all work tasks, resulting in a 15% cumulative gross upside to U.S. labor productivity. Briggs incorporates the potential for labor reallocation and new task creation into his estimates, which he argues will significantly boost productivity. He points to the historical record of technological innovation driving new opportunities, with 60% of workers today employed in occupations that didn't exist 80 years ago. 

This optimistic outlook is tempered by concerns about the substantial investment required for AI infrastructure, projected to exceed $1 trillion. Jim Covello highlights the high costs and complexity of AI technology, questioning whether it can solve the complex problems necessary to justify such investment. He notes that the cost of AI technology must decline dramatically to make automating tasks affordable, a scenario he finds unlikely given the current market dynamics. Covello also warns that the AI bubble could take a long time to burst, with infrastructure providers continuing to benefit in the meantime.

High AI Costs and Uncertain ROI May Deter Investment

• Hyperscalers like Microsoft, Alphabet and Amazon have spent between $60 billion and $80 billion in incremental capital above regular cloud capex on critical tools for building and training AI models. 
• Cloud computing companies are currently spending more than 30% of their cloud revenues on capex, with the vast majority of incremental dollar growth aimed at AI initiatives. 
• Return on invested capital (ROIC) visibility is low, and the timing of returns is uncertain. Analyst Eric Sheridan compares this to the Web 1.0 tech cycle, noting that returns on capital only became positive several years after initial investments, suggesting that AI's payoff period may extend well beyond the next few years.

Opportunities

AI-Driven Power Demand Surge Catalyzes Investments in Renewable Energy and Grid Modernization

• The U.S. is projected to need an additional 47 gigawatts (GW) of power generation capacity by 2030, with a 60/40 split favoring natural gas over renewables. This strategy aligns with major tech companies' commitments to green energy and the requirement for dependable power sources for AI data centers operating around the clock.
• Due to the rapid growth of data centers and the REPowerEU Plan, which aims to accelerate electrification, European power demand is expected to increase by 40 – 50% within the next decade. Europe plans to add nearly 800 GW of wind and solar capacity in the next 10 – 15 years.
• The expansion of AI data centers in Europe is expected to trigger an 80 – 100% increase in investments in power grids, marking a secular capex supercycle in the power infrastructure sector. 
• “Electrification Compounders” are likely to benefit significantly from the trend toward increased power demand, as they are strategically positioned to leverage the growth in power grids and renewables.

AI Infrastructure Investments Propel Market Dynamics and Technological Advancements

• Tech giants' substantial investments in AI infrastructure in 2024 will position them strategically to leverage massive distribution networks and customer bases, potentially yielding higher returns than previous technology cycles.
• AI infrastructure investments by hyperscalers in 2024 are projected to exceed $60 – 80 billion, aimed at the development of “killer applications” that could dominate this tech cycle.
• During the 1Q24 earnings season, significant capital expenditures in AI infrastructure by mega-cap technology companies resulted in a 22% year-to-date return for Phase 2 stocks. These include firms involved in semiconductor production, cloud computing infrastructure, data centers and other essential components supporting AI technology.
• Capital expenditures representing more than 30% of cloud revenues in 2024 are strategically targeted at AI initiatives, with expected substantial economic gains within the next 6 to 18 months.
• Projected AI investment exceeding $1 trillion in 2024 indicates a fertile ground for niche market innovations and the development of specialized applications addressing specific industry needs.

AI's Role in Utilities and Energy Sectors Opens New Avenues for Strategic Investments

• The projected 2.4% CAGR in U.S. power demand from 2022 to 2030 necessitates significant investments by utilities to meet this increasing demand.
• An estimated $50 billion is required in investments through 2030 to support new power generation for AI-driven data centers.
• The utilities sector demonstrated a robust performance with a 16% return from March to May 2024, positioning it as the best-performing sector in the S&P 500 during this period and indicating strong market confidence and growth potential. 
• Utility capital expenditures are expected to increase nearly 40% from 2024 to 2027, totaling about $140 billion annually, primarily to support AI integration and power demand.

Innovation Amid Constraints: AI Chip Supply Challenges

The semiconductor industry is grappling with significant supply constraints, particularly in High-Bandwidth Memory (HBM) technology and Chip-on-Wafer-on-Substrate (CoWoS) packaging, which are critical for AI applications. The HBM market is projected to grow at a 100% CAGR from $2.3 billion in 2023 to $30.2 billion in 2026. Despite this rapid growth, HBM demand is expected to outstrip supply, with a forecasted undersupply of 3% in 2024, 2% in 2025 and 1% in 2026. This shortfall is driven by the increasing HBM content requirements and the supply discipline of major suppliers like Samsung, SK Hynix and Micron. Additionally, the complexity of the HBM stacking process results in lower manufacturing yield rates compared to traditional DRAM, further constraining suppliers' ability to increase capacity.

Similarly, CoWoS packaging, a 2.5-dimensional wafer-level multi-chip packaging technology, is another bottleneck. TSMC's CoWoS capacity is expected to more than double in 2024 and nearly double again in 2025, but until then, the shortage will likely constrain AI chip shipments. This advanced packaging technology is essential for high-performance computing applications, and its limited supply has been a gating factor since the emergence of ChatGPT in late 2022. The tightness in CoWoS capacity has been highlighted by major players like Nvidia and AMD, who continue to face challenges in meeting the robust demand for AI chips. These supply constraints are expected to drive innovation at the chip layer as companies invest in developing more efficient manufacturing processes and alternative technologies to address these critical shortages.

Stay Ahead with Brightwave

Don’t miss out on the critical insights that could define your investment strategy. With Brightwave, you gain the power to produce granular analysis of vast document collections, seamlessly connecting the dots across diverse data sources. Request a demo today and experience firsthand how our AI-powered platform can help you stay ahead of the market, uncover hidden opportunities and make confident, informed investment decisions.