Skeptical About AI Rally: GQG Partners' Kersmanc

Key Concepts

• CapEx (Capital Expenditure): Investments in fixed assets like property, plant, and equipment.
• AI Mania: Over-enthusiasm and excessive investment in Artificial Intelligence.
• Conversion Rate: The percentage of users who transition from free to paid services.
• Unit Economics: The profitability of a single unit (e.g., a user) of a product or service.
• Hyperscalers: Companies providing large-scale cloud computing services (e.g., AWS, Azure, GCP).
• Cloud Commoditization: The process of cloud services becoming standardized and less differentiated, leading to price competition.
• Circularity Argument: The idea that investments in AI infrastructure are justified because they generate revenue that is reinvested in further infrastructure development.
• Special Purpose Vehicles (SPVs) & Joint Ventures (JVs): Legal entities created for specific purposes, often used to manage risk or finance projects.
• Large Language Models (LLMs): AI models trained on vast amounts of text data to generate human-like text.
• Post Training: Techniques applied to LLMs after initial training to improve performance and capabilities.

GS Commodities' Underweight Tech Stance

GS Commodities has become significantly underweight tech due to concerns about deteriorating fundamentals. A key concern is the reliance on cloud infrastructure, where CapEx is not directly reflected on companies' balance sheets.

CapEx Spending and Revenue Generation

• There has been substantial CapEx spending in the tech sector, totaling $600 billion.
• However, excluding infrastructure spending, only about $30 billion in revenue has been generated from this CapEx.
• This indicates a lack of headroom for returns on investment in many tech businesses.
• Microsoft is acknowledged to have a strong software business and steady earnings, but skepticism is growing about the returns from AI-related investments.

AI Mania and Alarming Fundamentals

• An alarm bell was rung on September 11th, suggesting the sector is at a significant inflection point, with excessive focus on "AI mania."
• The alarming fundamental is the current lack of substantial revenue generation from AI investments.
• The argument that OpenAI will generate $300 billion in revenue by 2030 to justify current spending is questioned.

OpenAI's Monetization Challenges

• OpenAI has a low conversion rate, with only about 2% of users paying for the service.
• OpenAI has 700 million users globally, with about half coming from emerging markets.
• AI is not like SaaS, where adding users directly increases profit. AI has a high cost to compute.
• The unit economics of AI are challenging, especially in emerging markets like India, where affordable mobile plans may make a $20/month AI subscription unappealing.
• OpenAI is valued at $500 billion in the latest secondary round.

Enterprise AI Adoption and Effectiveness

• The potential for enterprise AI adoption is being explored, but there are concerns about its effectiveness.
• An MIT study and feedback from tech consultants indicate a lack of effectiveness in many AI projects.
• One of the big three consulting firms reported that 85% of the 400 AI projects they worked on were "absolutely useless" in generating benefits.
• This echoes concerns that only about 15% of AI projects are vaguely working well.
• The argument that implementation issues are the primary cause of failure is challenged, with a need to demonstrate the math and monetization pathway.

Limitations of Large Language Models

• Large language models (LLMs) are facing limitations in their capabilities.
• The transition from GPT-4 to GPT-5 is not simply a matter of adding more compute.
• Models are peaking in terms of effectiveness, and further progress requires more post-training techniques.
• LLMs are extrapolations based on training data and cannot truly "think" or make decisions independently.

Hyperscalers and Cloud Pricing Pressure

• Much of the AI spending is concentrated on hyperscalers providing cloud services.
• Pricing dynamics in the cloud market are under pressure due to increased competition.
• Oracle is undercutting prices by 40-70% on enterprise deals, dragging down pricing across cloud players like AWS.
• This commoditization of cloud services is making it a less profitable venture.
• The situation is compared to the fiber build-out during the dot-com boom and bust cycle.

Circularity and Debt Concerns

• The circularity argument is questioned: if AI investments are so fantastic, why are participants funding their customers?
• Concerns are raised about obscure arrangements, special purpose vehicles (SPVs), and joint ventures (JVs) used to depreciate assets and manage debt off-balance sheet.
• This aggressive accounting and lack of transparency are typical of later stages in a cycle and raise concerns about the true economics of AI investments.

Conclusion

The main takeaways are that the AI sector is facing significant challenges related to monetization, effectiveness, and competitive pressures. The high levels of CapEx spending are not translating into sufficient revenue, and the limitations of current AI models are becoming apparent. The commoditization of cloud services and the use of complex financial structures raise further concerns about the long-term sustainability of the AI boom.