Consider a long-term bearish view on cloud providers like Amazon (AMZN), Microsoft (MSFT), and Google (GOOGL), as their business models face risk from customer concentration in the AI sector. While AI drives short-term growth, these large AI customers may eventually build their own data centers, eroding the hyperscalers' future revenue base. The software sector is also facing headwinds, so investors should be highly selective, favoring companies with defensible "systems of record" that are deeply integrated into customer workflows. Avoid broad software investments, as AI is creating legitimate shorting opportunities against companies that fail to adapt. Lastly, the single biggest risk to the global economy is a potential conflict over Taiwan, which controls the critical supply of advanced semiconductors.

Artificial Intelligence (AI) & Large Language Models (LLMs)

Dan Sundheim's firm, D1 Capital, has major private stakes in leading AI companies like OpenAI and Anthropic. He believes AI is creating the greatest synergy he's ever seen between private and public market investing.

• Business Model Debate: The key question for LLMs has shifted from if they can make money to what the return on capital will be.

  • These are the most capital-intensive businesses in history, requiring massive spending to train new models.
  • The biggest risk is that enterprise adoption of AI tools happens slower than expected, which is dangerous for businesses that require so much upfront capital.

• Analogies for the Business Model: Sundheim views the LLM business model as a combination of Netflix (NFLX) and Spotify (SPOT).

  • Netflix-like: They spend a huge amount of money upfront to build a fixed asset (the AI model), which they then sell at very high incremental margins. This creates a flywheel effect where revenue from one model funds the creation of the next, better model.
  • Spotify-like: The underlying models from different companies (e.g., OpenAI vs. Anthropic vs. Google) may become similar over time, like the music catalog on different streaming services. The key differentiator and "stickiness" will come from personalization and the data history a user builds with a specific service.

• Company Strategies:

  • OpenAI: Is pursuing a broad strategy, trying to apply its models to everything at once (consumer, enterprise, robotics, hardware). This is ambitious and hard to execute, but they have the talent to potentially pull it off.
  • Anthropic: Took a more focused approach on the enterprise market, where they have gained a market-leading position, especially in coding tools. This focus has led to a lot of success.

Takeaways

• AI is a long-term theme: The innovation happening in private AI companies provides a crucial lens for evaluating almost every public company, as AI will eventually impact them all.
• Focus on Capital Returns: When evaluating AI companies, the key question is not just about technological progress but whether that progress will generate a sufficient return on the enormous capital being invested.
• Personalization is the Moat: In the long run, the winning AI platforms may not be the ones with a temporary technological edge, but the ones that use personalization to create a sticky user experience, similar to Spotify's recommendation engine.

Hyperscalers / Cloud Providers (AWS, Azure, GCP)

Sundheim presented a bearish long-term view on the business models of the major cloud providers like Amazon's AWS (AMZN), Microsoft's Azure (MSFT), and Google's Cloud Platform (GOOGL).

• Short-Term Growth, Long-Term Pain:

  • In the short term, these businesses will likely see accelerating growth because their biggest customers are the rapidly expanding LLM companies.
  • However, Sundheim believes this is a long-term negative. The hyperscalers' customer base is shifting from thousands of different corporations to being highly concentrated in a few massive AI companies.

• Risk of In-Sourcing:

  • Once the major LLM companies become highly profitable (in the next 5-10 years), it will make economic sense for them to build their own data centers and insource their computing needs.
  • This would remove the hyperscalers' biggest and fastest-growing customers. Meta (META) is an example of a large tech company that already insourced its own compute.

• Increased Competition and Capital Intensity:

  • The business is becoming more capital-intensive because AI workloads are more expensive than traditional ones.
  • New, smaller competitors ("NeoClouds") are emerging that are better at running the specific GPU clusters needed for AI. Chipmakers like NVIDIA (NVDA) have an incentive to support these smaller players to ensure their customer base remains diversified.

Takeaways

• Re-evaluate Long-Term Moats: The traditional view of hyperscalers as unassailable, high-margin businesses may be challenged over the next decade.
• Concentration Risk: The increasing reliance on a few large AI customers creates significant concentration risk and could erode the hyperscalers' pricing power over time.
• Potential Bearish Thesis: While growth may look strong now, the long-term thesis is that the fundamental business model for hyperscalers is getting worse, not better, due to the structural shifts caused by AI.

Software Sector

The discussion addressed the recent market sell-off in software stocks, driven by fears that AI will commoditize software development.

• A Worsening Business Model: Sundheim believes the software sector will likely be a worse business model going forward than it has been in the past.

  • He compares the situation to Walmart (WMT) having to adapt to the rise of e-commerce. It was a painful, expensive transition that required massive investment and compressed margins, but the company ultimately survived.

• "Systems of Record" Are More Defensible:

  • Not all software is created equal. Companies that provide "systems of record"—like complex ERP or CRM systems that an entire business runs on—will be much harder to displace.
  • He notes that even the AI companies themselves are buying traditional ERP systems, not trying to code their own from scratch.

• Adapt or Die:

  • Software companies cannot afford to be complacent. They must actively integrate AI into their products and business models to stay relevant, much like how traditional retailers had to build an online presence.

Takeaways

• Be Selective with Software Investments: The broad-based tailwind that lifted all software companies is likely over. Investors should be more selective, focusing on companies with deep customer integration and strong distribution.
• Look for AI Integration: When analyzing a software company, a key question should be: "What is their AI strategy?" Companies that successfully leverage AI to enhance their existing products are more likely to thrive.
• Shorting Opportunities: For the first time, AI is creating legitimate short-selling opportunities in the software sector against companies that fail to adapt.

SpaceX (Private)

SpaceX is a major private holding for D1 Capital. The investment thesis has evolved and strengthened significantly over time.

• A Durable Low-Cost Producer: The core of the investment thesis is that SpaceX has built an impenetrable moat as the world's low-cost provider of launch services.

  • This is considered one of the "most beautiful" types of businesses: one where a low-cost advantage creates a positive feedback loop of more volume, which drives costs even lower.

• Starship is a Game Changer: The success of the Starship rocket is expected to reduce the cost of launching payloads into space by as much as 97%.

• Expanding Market: The initial thesis was just about the launch business. Now, with the success of its satellite internet constellation, the company's total addressable market (TAM) is the entire global telecommunications market.

  • SpaceX is on a path to be dramatically cheaper than any other method of delivering broadband internet.

Takeaways

• Power of Compounding Decisions: The success of SpaceX illustrates how great founders constantly make decisions that compound over time, leading to outcomes far greater than originally anticipated.
• Low-Cost Moats are Powerful: Businesses that achieve a sustainable and significant cost advantage are extremely powerful and difficult to compete with. This is a key attribute to look for in any long-term investment.

Geopolitics & Semiconductors

Sundheim identified the geopolitical situation around semiconductors as the single biggest tail risk facing the global economy.

• Collision Course with China: He believes the U.S. and China are on a collision course over Taiwan, which produces over 90% of the world's most advanced semiconductors.

• Fragile Supply Chain: This supply chain is critical to the global economy, powering everything from phones to cars to AI data centers. It is extremely fragile and would take 10-20 years to replicate elsewhere.

• Risk of a "Great Depression" Scenario: If the semiconductor supply chain were to be disrupted by a conflict, the economic fallout would be catastrophic, potentially on the scale of the Great Depression.

Takeaways

• Monitor Geopolitical Risk: This is a major macro risk that investors should be aware of, as it could have devastating effects on global markets and nearly every industry.
• Onshoring is a Multi-Decade Process: While there are efforts to build semiconductor fabs in the U.S. and Europe, this is a very long and expensive process. The world will remain heavily dependent on Taiwan for the foreseeable future.
• AI Raises the Stakes: The increasing importance of advanced chips for AI makes control of the semiconductor supply chain an even more critical geopolitical issue.