Goldman Sachs' "AI Narrative Framework": Five Key Controversies About AI

Author: Dong Jing

Source: Hard AI

After the Communacopia+ Technology Conference held by Goldman Sachs in September, along with a series of corporate and industry announcements regarding "long-term AI capacity," the debate in the market about whether "we have entered an AI bubble" has become increasingly vocal.

On October 8th, according to Hard AI, Goldman Sachs' TMT research team recently released a new research report that conducts an in-depth analysis of five key controversies in the current AI field, attempting to provide the market with a clear narrative framework for AI. These five core controversies cover the current state of consumer and enterprise AI adoption, expectations for AI spending, demand for power infrastructure, and the highly concerning issue of bubble risk.

The research report states that, according to Goldman Sachs' analysis, the speed of consumer AI adoption has exceeded expectations, with ChatGPT reaching a record of 700 million active users per week in July, but its monetization capability still lags behind infrastructure investment. Goldman Sachs noted that the return on investment for AI at the enterprise level remains limited; despite the continued expansion of internal deployments, MIT research shows that only 5% of companies see measurable impacts on their profit and loss statements from AI.

Meanwhile, Goldman Sachs indicated that AI infrastructure investment has reached historic levels this year, with the capital expenditures of the world's five largest cloud service providers (Amazon, Microsoft, Google, Meta, and Oracle) expected to reach $381 billion, a year-on-year increase of 68%. The report pointed out that the rapid expansion of AI workloads will significantly increase demand for data centers, with global power demand expected to grow by more than 165% by 2030.

Regarding the highly concerning issue of AI bubble risk, Goldman Sachs believes that although the current market environment bears similarities to the late 1990s, the price-to-earnings ratio of the Nasdaq 100 index is 46% lower than the peak of the internet bubble, and IPO activity is also far below the levels at that time, indicating that conditions for a significant correction are not present.

Consumer AI Adoption: Rapid Growth but Lagging Monetization

The growth rate of consumer AI usage has outpaced that of enterprise applications and continues to grow rapidly, but AI companies' ability to monetize consumer services and generate revenue lags behind their spending on AI infrastructure to meet demand.

According to Goldman Sachs' analysis, the speed of consumer adoption is rapidly increasing, with OpenAI reporting that ChatGPT reached a record of 700 million active users per week in July, far exceeding the speed of monetization growth.

Based on data analysis from SensorTower:

• ChatGPT dominates both the global and U.S. markets, consistently leading in monthly active users and daily active users.

• Google Gemini has become the second-largest platform due to its distribution channel advantages, achieving significant user base growth, but still significantly lags behind ChatGPT.

• Claude and Perplexity maintain relatively small user bases.

It is noteworthy that platform differentiation is becoming apparent. According to usage reports released by various platforms:

• OpenAI disclosed that ChatGPT is increasingly used for non-work purposes (with a ratio of approximately 70/30 for non-work to work-related tasks);
• Anthropic's report indicates that Claude is most commonly used for programming/computational tasks (36% usage), suggesting that users are beginning to choose based on the specific strengths of each model.

More notably, employees from 90% of companies report regularly using personal AI tools at work, but only 40% of companies have purchased official LLM subscription services.

The research report points out that the recent wave of product releases highlights the ongoing maturation of AI capabilities into tangible, monetizable applications. The instant checkout feature of ChatGPT enabling in-chat purchases with partners like Etsy, as well as the strategic collaboration between Google and PayPal in agency commerce, underscore the potential growth of AI-driven transactions.

Goldman Sachs expects 2026 to be a key year for AI monetization, with the cross-application of advertising and commerce driving scalable monetization.

Corporate AI Deployment: Internal Application Expansion but ROI Still Needs Improvement

Despite companies continuing to deploy generative AI internally and externally, the visibility of ROI (return on investment) remains low.

In terms of internal applications, companies are deploying AI to drive incremental efficiency improvements and support margin enhancement. Goldman Sachs has observed tangible efficiency gains in complex tasks such as content and software development, advertising creative generation, inventory management, and dynamic pricing. Efficiency gains are reflected in increased productivity and a slowdown in hiring growth.

However, external corporate applications that can generate revenue growth or market share shifts are progressing slowly. The MIT Business AI Status Report shows that only 5% of companies have seen measurable impacts on their profit and loss statements. Goldman Sachs' analysis suggests that this is primarily due to inconsistencies in AI deployment methods and oversight systems.

In the consumer internet space, Goldman Sachs identifies that AI has the potential to disrupt billion-dollar profit pools. Traditional advertising agencies face automation threats from AI-driven platforms like Google Performance Max and Meta Advantage+, with a global advertising agency ecosystem representing approximately $161 billion in profit pools.

At the same time, AI makes digital advertising more effective and targeted, with the potential to accelerate the shift of advertising budgets from traditional channels to digital channels, bringing about $170 billion in incremental opportunities for digital channels between 2025 and 2028.

In the software industry, Goldman Sachs' 2025 industry dialogue indicates that the AI application ecosystem is maturing. SaaS leaders like Microsoft, Salesforce, and ServiceNow have begun to disclose AI-specific contributions, the key is when these contributions can have an additive effect on growth algorithms.

Nevertheless, the ROI in the enterprise market is still pending, as substantial investments in building, training, and utilizing foundational model AI have yet to penetrate enterprises in a materially economic return manner.

AI Spending Forecast: Unprecedented Scale of Infrastructure Investment

Goldman Sachs anticipates an increase in AI infrastructure spending levels from both sides of the ecosystem by 2025. Since the release of the report analyzing AI's impact on industry profit pools in June, investor attention to the mid-term and long-term returns on super-sized AI spending levels has significantly increased According to Goldman Sachs estimates, the capital expenditures of the five major hyperscale cloud service providers (Amazon, Microsoft, Google, Meta, and Oracle) have further increased this year, and are expected to reach a total of $381 billion by 2025, a year-on-year increase of 68%.

Most AI investments will be announced in the second half of 2025, with hundreds of billions of dollars in partnerships announced in September, such as Oracle's $300 billion deal with OpenAI and Nvidia's $100 billion investment in OpenAI.

Goldman Sachs stated that the further increase in these expenditures, combined with observed consumer demand continuing to grow and potentially exceeding future computing supply (despite current investments), adds tension to the existing debate about whether hyperscale capital expenditures will yield long-term returns.

Goldman Sachs expects that the five major hyperscale cloud service providers will continue to increase AI-related capital expenditures to keep up with the growing consumer demand, estimating that expenditures will grow to approximately $1.4 trillion from 2025 to 2027.

However, the research report pointed out that a recurring theme is the significant gap between the current demand for AI services (from users and large platform/product development needs) and the currently available capacity. This disconnect is most evident in the growing backlog of cloud computing companies, which, if realized, should sustainably support revenue growth in the next 2-3 years.

Power Infrastructure Demand: 165% Growth Brings Construction Challenges

According to Goldman Sachs' global data center and utilities team, the impact of AI and non-AI workloads on data center power demand is significant.

The rapid expansion of AI workloads will significantly increase data center demand, with global power demand expected to grow by more than 165% by 2030. Previous analyses by Goldman Sachs' data center team showed that global data center demand was approximately 62GW in the second quarter of 2025, with AI accounting for 13%; it is expected that total demand will reach approximately 92GW by 2027, with the share of AI workloads rising to 28%.

Meeting this demand requires large-scale power generation capacity construction. Goldman Sachs expects that by 2030, data center power demand will grow by approximately 165% compared to 2023. In the United States, 60% of future demand will require new power generation facilities, with an expected incremental capacity of 72GW, primarily sourced from natural gas (60%), solar (25-30%), and wind (10-15%).

Grid investment expectations have been raised from $720 billion in July to $780 billion (by 2030), an increase of $60 billion, with a focus on distribution infrastructure, while transmission capital expenditures are growing faster to maintain data center supply growth.

Bubble Risk Assessment: Similar but Not Completely Aligned with the 1990s

Goldman Sachs believes that the current market shares certain similarities with the late 1990s, but has not yet reached a level that would lead to a significant repricing of publicly traded stocks.

First, current valuation levels are far below those of the late 1990s: The Nasdaq 100 index is currently trading at a discount of about 46% compared to the internet bubble period, with the LTM price-to-earnings ratio at the end of 1999 being 68.4 times, while as of October 3, 2025, it is approximately 37.0 times Secondly, the comparison of IPO activities also shows differences: The number of IPOs in the United States from 1998 to 2000 was 892, while it is significantly lower from 2023 to 2025, although the average transaction size increased from USD 176 million in 1998-2000 to USD 254 million in 2023-2025. This indicates that the current market is more cautious, with only larger, more mature companies choosing to go public.

Finally, macro conditions are also stricter: From March 1999 to March 2000, the average yield on 10-year U.S. Treasury bonds was about 6.0%, while from September 2024 to September 2025, it is about 4.3%. Goldman Sachs' macro team expects further interest rate cuts of 25 basis points in October and December, which may improve capital inflows in the near term