Don't just focus on chips, Goldman Sachs: From data management to employee training, AI Agents require "large-scale non-hardware investments".

Don't just focus on chips, Goldman Sachs: From data management to employee training, AI Agents require "large-scale non-hardware investments".

```

Author: Long Yue

Source: Hard AI

Market attention on AI investment is overly concentrated on hardware, while “non-hardware investments” such as data infrastructure, organizational redesign, and workforce restructuring are rapidly rising; their scale and impact can no longer be ignored.

Joseph Briggs and other analysts from the Goldman Sachs Global Economic Research team recently released a report stating that U.S. AI-related hardware capital expenditures have reached $360 billion (1.1% of GDP), global hyperscale cloud vendor capital expenditures will hit $400 billion in 2025, and are expected to surpass $700 billion in 2026. But this is only half the story.

Analysts estimate that U.S. enterprises' current AI-related labor costs have reached $153 billion annually, and executives’ time investment in AI organizational transformation translates into $40 billion per year. Historical patterns show that for every $1 increase in hardware investment, $2 in intangible capital is driven—based on this, global non-hardware AI investments in the coming years may break through $1 trillion.

How large is non-hardware investment? Estimation by four dimensions

Analysts estimated the scale of current non-hardware AI investments from four angles.

First, IT labor costs. In the past two to three years, the proportion of AI-related IT positions has risen to about 25%, which matches the data from enterprise surveys that “20% to 40% of IT budgets are allocated to AI projects.” Based on this, analysts estimate that annualized investment by non-tech companies through internal IT teams for AI tool development is about $153 billion.

Second, executive time costs. Surveys show about one-third of large enterprise executives’ performance reviews and compensation are tied to AI strategy; AI has consistently been the top priority for U.S. executives. Referring to Corrado Hulten Sichel’s (2005) methodology and assuming 20% of executive time goes to organizational innovation, of which 35% is focused on AI, analysts estimate annualized AI-related organizational capital investment exceeds $40 billion (based on the U.S. executive payroll of about $600 billion).

Third, workforce restructuring costs. Analysts compiled AI-driven layoffs from companies such as Block, Atlassian, HP, Oracle, Accenture, Salesforce, Chegg, and C3.AI. The average restructuring cost per affected employee is about $84,000.

Currently, AI’s impact on the job market is still limited—Goldman Sachs estimates AI-related hiring resistance is lowering employment growth by about 10,000 jobs per month, corresponding to current workforce restructuring investment of around $10 billion/year. However, extrapolating from forecasts that AI will replace 6% to 7% of the workforce, total U.S. workforce restructuring costs over the AI adoption cycle may reach $800 billion to $900 billion; amortized over 10 years, this is roughly $90 billion/year.

Fourth, historical statistical patterns. Analysts used the EU KLEMS database for regression analysis and found a significant statistical relationship between ICT hardware investment and intangible capital investment: historically, every $1 increase in hardware investment drives $2 in intangible capital—including $1.3 in data/software, $0.5 in organizational capital, and $0.2 in other intangible assets.

Applying this multiplier to current AI hardware investment scales, it corresponds to about $700 billion in the U.S. and $1 trillion globally in supporting intangible capital investment. Analysts also referenced Atlanta Fed survey data as evidence—covering software, subscription services, hardware, employee training, and IT support—to estimate enterprise AI-related spending in 2026 at $280 billion.

The revenues of data management companies already tell the story

The acceleration of non-hardware investment can already be traced in market data.

Leading data management and infrastructure companies such as Snowflake, Databricks, and Palantir have seen their revenues triple or more since the emergence of ChatGPT in 2022. The combined enterprise value of these three companies soared from less than $100 billion in 2022 to over $650 billion by the end of 2025.

Cloud revenues also confirm this trend.

The cloud revenues of Amazon AWS, Microsoft, Google, and Oracle have grown from roughly $200 billion in 2022 to over $500 billion now, with market expectations that it will exceed $1 trillion by the end of this decade. For 2026, consensus forecasts of revenue have already been raised by more than $150 billion compared to the end of 2022.

The essence of non-hardware investment: intangible capital

Analysts classify the above non-hardware investments as “intangible capital investment,” including patents, trademarks, brands, software, R&D, employee training, and organizational management capabilities.

According to the EU KLEMS database, intangible capital investment now accounts for over 50% of all investment in the U.S. and U.K., and about 48% for the G10 as a whole.

Over the past 20 years, the rise in proportion was mainly driven by organizational capital and database/software investments—precisely the largest directions of investment for AI and AI Agent deployment.

Productivity “J-curve”: today’s underestimation is tomorrow’s surprise

Large-scale intangible capital investment will depress GDP and productivity statistics in the short term—this is a documented economic “J-curve effect.”

This is because when companies shift resources to internal AI tool development, process redesign, and employee retraining, these expenditures are recorded as costs rather than investments in national accounts, and the corresponding asset value is not counted.

Analysts estimate that just AI-related organizational investment (executive time $40 billion + workforce restructuring $10 billion, totaling about $50 billion/year) has already caused U.S. GDP to be underestimated by at least 0.2%. If the statistical relationship holds, the underestimation could be as high as 2% of GDP.

In other words, recent improvements in U.S. productivity data still have room.

Who will become the next generation of “superstars”?

This is the report’s most direct message for investors.

Goldman Sachs referenced the “Four S” framework proposed by economists Haskel and Westlake in “Capitalism Without Capital” (2017), pointing out the fundamental differences between intangible and traditional tangible capital:

  • Scalability (marginal cost approaches zero)
  • Sunk cost (cannot be resold, higher risk)
  • Spillover (knowledge and practices are easily copied by competitors)
  • Synergy (different types of intangible assets reinforce each other and amplify value)

This cost structure—high fixed cost, low marginal cost—naturally favors first movers. Historical data has already confirmed this: in the past 40 years, the revenue share of “superstar” companies and their intangible capital investment scale have almost risen in tandem.

The report’s regression analysis further shows that every percentage point increase in the intangible capital share leads to a 0.2-0.3 percentage point decrease in the labor cost share of added value within 2–4 years, with the effects most notable for brands, software/database, and organizational capital.

The conclusion is: Today, companies that invest more effectively in data, workforce, and organizational infrastructure to deploy AI Agents are very likely to become the next generation of superstar firms with excessive valuations.

Goldman Sachs also gives two important caveats.

First, the above analysis does not involve the issue of profit distribution within the AI technology stack. Chip manufacturers or foundation model providers may also become true AI superstars—this depends on where market power ultimately accumulates, which is still highly uncertain.

Second, automation and business process standardization brought by AI will lift productivity for most firms, not just a few winners. Analysts maintain their previous judgment: AI will bring broad, economy-wide productivity improvement; after full AI adoption in developed economies, labor productivity and GDP are expected to rise by about 15%.

 

This article is from WeChat official account “Hard AI”. For more cutting-edge AI news, please visit here

Risk Warning and DisclaimerThe market has risks, investments need to be cautious. This article does not constitute personal investment advice and does not take into account the special investment objectives, financial situation, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article suit their specific situation. Investing accordingly is at your own risk. ```