Is A.I. Spending Sustainable? | Barron's Streetwise

Key Concepts

• AI Infrastructure Spending: Capital expenditures by companies on hardware, software, and services to support artificial intelligence development and deployment.
• Debasement Trade: An investment strategy driven by concerns about currency devaluation or inflation, leading investors to move cash into assets perceived as hedges, such as Bitcoin and gold.
• Valuation: The assessment of an asset's current worth, often measured by metrics like price-to-earnings (P/E) ratios.
• Polycule: A term describing a relationship network involving multiple partners, used metaphorically to illustrate the complex interdependencies in the AI industry.
• Remaining Performance Obligations (RPOs): A financial metric representing future contracted revenue, indicating a company's future revenue stream from existing contracts.
• Meme Stock: A stock that experiences a significant price surge driven by social media hype and retail investor coordination, often detached from fundamental value.
• Convertible Securities: Hybrid financial instruments that combine features of both bonds and stocks, allowing holders to convert them into a predetermined number of shares.
• Capital Expenditures (Capex): Funds used by a company to acquire, upgrade, and maintain physical assets such as property, buildings, and equipment.
• Dark Fiber Moment: A situation where significant infrastructure is built in anticipation of future demand that does not materialize as quickly as expected, leading to underutilization.
• Power Wall: A constraint on data center growth due to insufficient power supply or the high cost of electricity.
• Inference vs. Pre-training: In AI, pre-training is the initial, computationally intensive phase of model development, while inference is the phase where the trained model is used to make predictions or generate outputs, typically requiring less compute.
• Hedging Strategy: An investment technique used to offset potential losses or risks in a portfolio.
• Equity Euphoria Index: A proprietary index that measures the level of euphoric behavior in equity markets by analyzing data from the derivatives market.
• Selling Out-of-the-Money Call Options: A strategy where an investor sells call options on a stock they own, betting that the stock price will not rise above a certain level, generating premium income.
• Buying Puts: A strategy where an investor buys put options, which give them the right to sell a stock at a specified price, acting as insurance against a price decline.

AI Infrastructure Spending: Not Going to Zero

The discussion begins with the assertion that AI infrastructure spending is highly unlikely to decrease to zero. While acknowledging that spending could potentially decline, the speaker, Venu Krishna (Head of US Equity Strategy at Barclays), suggests it's not necessarily expected to fall from current levels. This spending is a significant driver of current market performance, with nearly 70% of first-half GDP growth attributed to AI-related capex spending by some estimates.

Warning Signs for Investors

Jack How, the podcast host, outlines three "glaring warning signs" for investors, emphasizing that these are presented for discussion and not necessarily as reasons to sell.

1. AI Open Relationships (The AI Polycule)

This section uses the analogy of a "polycule" (a relationship network with multiple partners) to describe the complex and interconnected financial relationships within the AI industry. A Morgan Stanley report titled "AI Mapping Circularity" is cited, illustrating these intricate connections between major tech companies like Nvidia, AMD, Oracle, Microsoft, Coreweave, and OpenAI.

• Key Points:
  • Interdependence: Companies are heavily reliant on each other, with complex financing and partnership structures.
  • Coreweave Example: This company, formerly in crypto mining, now deploys Nvidia AI chips in data centers. Nvidia is also an investor in Coreweave, and vendor financing is involved.
  • OpenAI's Role: OpenAI accounts for a significant portion of Remaining Performance Obligations (RPOs) for companies like Oracle (two-thirds) and Coreweave (40%), highlighting their dependence on OpenAI's success.
  • Challenges in Evaluation: New financing structures and off-balance sheet partnerships make it difficult for investors to evaluate risks.
  • Market Impact: AI companies have driven almost all of the S&P 500's rise since ChatGPT's public launch, meaning even index fund investors have substantial exposure to this "AI polycule."

2. Re-fried Memes

This section revisits the phenomenon of "meme stocks," exemplified by GameStop, where heavily shorted stocks experienced dramatic price surges driven by social media coordination.

• Key Points:
  • Relaunch of Meme Stock ETF: Roundhill has relaunched its Meme Stock ETF, which previously closed after less than two years. The initial launch coincided with a speculative peak.
  • Shift in Methodology: The new ETF uses an active stock picker rather than short-selling activity and online chatter as gauges of meme status.
  • Holdings: The new ETF's top holdings include quantum computing, fuel cells, crypto, AI, and rare earth metals. Many are high-flying stocks with distant profit prospects.
  • Reggetti Computing Example: This quantum computing company is the top holding, with a stock up over 6,000% in a year, despite minimal revenue and no expected profit this decade.
  • Definition of a Meme Stock: The speaker argues that a true meme stock requires a "punchline" or an element of humor in its speculative rise, distinguishing it from sincere investments in promising but unproven technologies. The relaunch of a meme fund without traditional meme stocks is presented as a potential warning sign.

3. Convertible Securities Outperforming

The third warning sign is the strong performance of convertible securities, which are outperforming the stock market.

• Key Points:
  • Performance: The iShares Convertible Stock ETF has returned around 23% year-to-date.
  • Nature of Convertibles: These are hybrid instruments that combine features of bonds and stocks, offering companies a way to raise capital by agreeing to pay interest or give up ownership.
  • Holdings: The top holding in the iShares fund is Strategy, a company that hoards Bitcoin. Crypto, fuel cells, and AI "cash burners" are well-represented.
  • Interpretation: The strong performance of convertibles from these companies suggests that their underlying stocks are performing exceptionally well, rather than a neglected part of the fixed-income market gaining traction.

Conversation with Venu Krishna: AI Spending and Market Implications

The podcast then features an interview with Venu Krishna, who provides a more in-depth perspective on AI spending and its market implications.

AI Capex Spending: Demand vs. Supply

• Current State: AI capex is near a record high as a percentage of revenue. Many large companies have substantial cash flow to fund these investments, while others are borrowing.
• Demand Outstripping Supply: Currently, demand for compute power is outstripping supply. AI is already being deployed in core businesses, leading to improved profitability for major spenders (e.g., Microsoft's profitability improving despite layoffs due to software productivity gains).
• Consumer vs. Enterprise Utilization: While consumer-level utilization of AI (e.g., ChatGPT users) is high, enterprise-level utilization across larger datasets is still in its early stages.
• Dotcom Bubble Comparison: Krishna acknowledges concerns about a potential "dark fiber moment" (infrastructure overbuild) similar to the dotcom bubble. However, he argues that the current situation is fundamentally different because companies investing now are in a much stronger financial position than those during the dotcom era.
• Race for First Mover Advantage: The intense competition and the desire for first-mover advantage in AI are driving significant capital spending, even if there's a risk of mistiming investments. The scale of current spending, even inflation-adjusted, is far greater than during the dotcom period.

Circular Financial Relationships in AI

• Intertwined Deals: Krishna highlights the increasing prevalence of circular financial relationships, where companies are their own best customers, providing financing, investing in partners, and selling to them.
• AMD-OpenAI Example: The deal where AMD sells chips to OpenAI and also provides warrants (effectively an equity stake) is cited as surprising. This suggests OpenAI has leverage as a major customer, and AMD is keen to secure future demand.
• Conflicting Incentives: This intertwined relationship creates a trade-off: as a chip buyer, OpenAI wants the lowest price, but as an equity holder in AMD, it wants AMD to profit richly.
• Market Intelligence and Acquisition: Such stakes are also seen as a way for large companies to gain market intelligence and potential acquisition options for emerging innovations.
• Nvidia's Dominance: The current market structure, with Nvidia holding a significant market share (around 85%), allows it to command higher prices and margins due to overwhelming demand.

Potential Constraints on AI Buildout

Krishna identifies three to four broad factors that could slow down AI data center growth:

1. Power Constraint (The Power Wall):

   • Issue: Data center growth is outpacing power availability. This also impacts broader electricity prices for consumers.
   • Political Aspect: Local opposition to data centers due to electricity price concerns is growing.
   • Fermy Example: The IPO of Fermy, a single-asset REIT focused on supplying power for data centers, illustrates the long lead times (6 years for the first nuclear reactor) and regulatory hurdles in securing power.

2. Demand Wall:

   • Shift from Pre-training to Inference: Demand for compute decreases as AI models move from pre-training to inference.
   • Model Efficiencies: Improvements in AI models could lead to a better balance between demand and supply, reducing the need for extensive capital spending.

3. Funding Issues:

   • Rising Scale: As the scale of funding requirements increases, the ability to secure subsequent rounds of funding on sound financial footing will become crucial.

Hedging Strategies for Investors

Krishna offers insights into hedging strategies for investors concerned about potential market downturns or a slowdown in AI capex.

• Auxiliary Beneficiaries: Investors can look beyond the "hyperscalers" (major AI players) to sub-sectors within energy, industrials, power, and utilities that benefit indirectly from AI capex. However, exuberance is already showing in these areas, with multiples rising rapidly (e.g., hard drive companies trading at 20x earnings, up from single digits).
• Equity Euphoria Index: Barclays' derivatives team has launched an "Equity Euphoria Index," which currently shows near-peak levels of euphoric behavior among stocks, particularly smaller companies that are AI auxiliary beneficiaries.
• Cheap Options: A suggested strategy is to look for cheap options on these beneficiaries and buy protection.
• Selling Out-of-the-Money Calls and Buying Puts: For companies trading at high multiples due to AI hype, investors can sell out-of-the-money call options to generate premium income. This income can then be used to buy put options, effectively hedging against a potential price decline. This strategy limits upside potential but provides downside protection.
• DeepSea Example: The "DeepSea" event (concerns about Chinese companies' AI capabilities) is cited as an example where hyperscalers and auxiliary beneficiaries experienced sell-offs, leading to a rotation into more defensive and quality/value names. This demonstrates that even during a broad market downturn, 70% of S&P stocks can still be up, highlighting the importance of seeking protection.

Conclusion

The discussion emphasizes that while AI infrastructure spending is robust and unlikely to disappear, investors should be aware of potential risks and consider hedging strategies. The complex interdependencies within the AI ecosystem, the potential for overbuilding, and the constraints of power availability are key concerns. The conversation concludes by reiterating the prudence of protecting oneself against potential slowdowns in capex spending.