← Back to list
Colossus (Invest Like the Best / Business Breakdowns)Podcast2 Jul 2024Source: joincolossus.comHost: Patrick O'Shaughnessy

Modest Proposal - AI Commoditization and Capital Dynamics - [Invest Like the Best, EP.380]

In plain words

This piece features investor Modest Proposal, who warns that the current AI infrastructure boom (chips, data centers) mirrors the 2000s commodity cycle: a shortage leads to overinvestment, then a glut. He's cautious on NVIDIA long-term, noting its high profits will attract rivals. He also highlights GLP-1 drugs (like Novo Nordisk's Ozempic) as economically huge but limited by patient adherence. He's bullish on Japanese stocks due to corporate reforms pushing buybacks.

AI SummaryAI-generated · may contain errors · verify against the original

At a Glance Modest Proposal discussed AI commoditization and capital dynamics on the Invest Like the Best podcast. The core argument: the current AI investment boom resembles the mid-2000s commodity market, carrying risks of overbuilding. NVIDIA benefits from surging AI computing demand, but open-so

~13 min full read · 9 sections
Deep Analysis

This Issue at a Glance

Modest Proposal (anonymous public and private market investor) discusses with Patrick O'Shaughnessy the AI investment frenzy, post-pandemic capital cycles, and the economic impact of GLP-1 drugs. Core judgment: The current AI infrastructure investment boom is replaying the classic capital cycle of the mid-2000s commodity market—"shortage inevitably leads to surplus"—yet the market continues to assign permanent valuation multiples to "temporary surpluses."


1. AI Investment Frenzy: A Replay of the Mid-2000s Commodity Capital Cycle

Modest Proposal argues that the current surge in AI computing demand closely mirrors the commodity market dynamics following China's WTO accession in the mid-2000s.

Historical Context: In the mid-2000s, China rapidly transitioned from a marginal player in global trade to the largest consumer of commodities, triggering sharp price spikes in copper, iron ore, and oil. Capital flooded into emerging markets, assigning high multiples to "temporary surpluses" at mining companies. Over the subsequent 3-5 years, a wave of new mines came online, and by 2014, oversupply led to a price collapse—"oil fell from $140 to $80, and the commodity index remains below its peak to this day."

Current Analogy: The AI sector is experiencing the classic scenario of "surge in demand vs. rigid supply":

  • Chips: TSMC's capacity is sold as soon as it is produced.
  • Electricity: U.S. electricity consumption was flat for the past 20 years, but the report forecasts annual growth of approximately 2.5% over the next 7-8 years, cumulatively exceeding 30%.

Key Mechanism: "Putting a multiple on a temporary surplus is essentially capitalizing a long-term cash flow." However, capitalism self-corrects—high prices attract supply, eventually leading to a glut. Modest adds, "I joke that we're smart enough to build AI and AGI, yet can't figure out how to supply power efficiently—the irony is quite something."

Specific Analysis of NVIDIA:

Dimension Short-Term (12 Months) Long-Term (Perpetual Cash Flow)
What Needs to Be Believed Frontier labs believe that "a 100x/1,000x/10,000x increase in compute intensity can achieve AGI," thus sustaining large-scale GPU purchases Can one company become a permanent "bottleneck" node for the five largest companies in history?
Risk Labs' willingness to spend may exceed expectations There are smart people in every garage thinking about how to run GPU-equivalent tasks more efficiently; some teams have already found ways to run certain functions on CPUs
Data Expected to achieve approximately $200 billion in revenue with 80% gross margins Such a massive profit pool will inevitably attract a flood of competitors

> Reader's Note: Modest is a public market investor, and their analysis is framed by a preference for identifying "capital cycle inflection points," not a denial of NVIDIA's business itself.


II. AI Commercialization: Two Paths — Open-Source Models vs. Frontier Models

Modest Proposal argues that a consensus is forming in the investment community — that LLM models are becoming commoditized — but this judgment may be falsified.

Current consensus: Open-source models (e.g., Llama, Mistral) will be "good enough," while frontier models may only serve truly leading applications and fail to transition to AGI. If so, the commoditized model layer will cede enormous value to the application layer.

Counterargument: In the past 10 days, two individuals from OpenAI (one former, one current) have respectively stated that "AGI will arrive within 3-5 years." If frontier models can establish a significant performance advantage, then:

  • Winners: Google, Microsoft/OpenAI
  • Losers: AWS, a large number of software companies

The economic consequences of the two paths are vastly different:

Scenario Value Flow Key Variables
Open-source models "good enough" Application layer captures most value Speed of model commoditization, data gravity (data stays with AWS if it is on AWS)
Frontier models significantly ahead Infrastructure layer (Google, Microsoft) captures most value Whether general-purpose models can be "good enough to replace everything"

Current commercialization status: On the revenue side (Microsoft Azure's AI incremental revenue, various software applications), the total amounts to roughly "mid-single-digit billions of dollars"; on the cost-saving side (e.g., the Klarna story), it is also limited. Combined, these "cannot justify the $200 billion in capital expenditure" — but Modest believes this is not unusual during an infrastructure platform transition, drawing an analogy to the "$120 billion fiber optic investment" in the late 1990s.


3. Post-Pandemic Capital Cycle: Shortage Inevitably Leads to Glut

Modest Proposal argues that the demand distortions caused by COVID are triggering a classic capital cycle across multiple industries—overinvestment followed by overcapacity.

Typical Cases:

1. Transportation & Freight Market: The surge in home consumption during 2020-2021 led to a massive influx of trucking capacity. In May 2022, Target and Walmart announced "inventories are full, stop restocking." Two years later, freight volumes have yet to recover, and capacity oversupply has resulted in "terrible pricing."

2. Beverage Can Industry: Global annual growth had long been 2-3%. During COVID, home consumption drove demand to 10% per year. Capital expenditures at Ball and Crown surged from $600 million and $500 million annually to $1.5 billion each. After the pandemic ended, demand fell, leading to severe overcapacity—"a historically very stable, oligopolistic, and capital-rational market structure was broken."

3. Amazon: "Rebuilt UPS in two years"—its delivery network expanded to the scale of building another UPS, followed by a slowdown in demand that triggered a digestion period.

Investment Implications: "If you can catch the bottom of these capital cycles—after excessive spending, capital expenditure cuts, capital expenditure hitting a trough, and then a new upcycle begins—that is a very interesting entry point." However, Modest cautions: "Freight hasn't reached that point yet; people have been trying to bottom-fish for a year."


4. GLP-1 Drugs: Enormous Economic Potential vs. Behavioral Barriers

Modest Proposal argues that the economic impact of GLP-1 drugs could rival that of AI, but patient adherence remains a key bottleneck.

Economic Potential: "Lifting a large portion of the population out of obesity, diabetes, and long-term care costs—this will free up productive resources in an unprecedented way." However, Modest cites data from Stanford's Burstyn research report, noting:

  • Long-term adherence rates for cancer drugs are approximately 60-70%
  • Adherence rates for other chronic disease drugs are even lower
  • If the GLP-1 adherence rate is only 35%, the number of new patients required annually to reach 50 million Americans on the drug would be extremely large

Key Issue: "Human behavior is very difficult to change, especially when even life-saving drugs have an adherence rate of only three-quarters." Long-term safety data is still being accumulated, but "safety looks tenable for now."


5. Public vs. Private Markets: Distorted Capital Allocation

Modest Proposal argues that the excessive flow of institutional capital into private markets has become one of the most distorted phenomena in capital markets.

Data Comparison:

Metric Public Markets (Nasdaq 100) Private Markets
10-Year Rolling Return Approx. 5x MOIC Even top-tier venture capital funds struggle to consistently achieve 5x MOIC
Private Equity IRR Claimed mid-teens IRR, corresponding to approx. 1.6-1.7x MOIC
Liquidity Fully liquid Locked up for years
Leverage None 4-6x

Key Data: "Over the past three years, capital returned to private markets has been approximately $300 billion less than capital deployed." Modest believes the rationale for institutional allocation to private assets has shifted from "optimizing the efficient frontier" to "smoothing returns and avoiding volatility"—"If you actually sit down and do the math, I don't think it's optimal for these pools to allocate 30-45% of their assets to illiquid private markets."

Recommendation for Large Allocators: "For the first time in a long while, bonds have become a genuine option today—the 10-year U.S. Treasury offers a 1-2% real return." The allocation should include: U.S. equities, international developed market equities, some bonds, and a small amount of private assets—but "one must be very clear about why they are investing in private markets."


6. Japan Equity Market: Potential Capital Management Reform Opportunities

Modest Proposal believes that the capital management reforms promoted by the Tokyo Stock Exchange may create structural opportunities.

Mechanism: The TSE requires listed companies to submit capital management plans and regularly publishes lists of "submitted/not submitted" companies. Companies that fail to submit face the pressure of "public shaming." Common issues among Japanese companies include:

  • Overcapitalization
  • Extensive cross-shareholdings
  • Extremely low dividend payout ratios and share buyback ratios
  • "Strategic holdings established 60 years ago due to the absence of venture capital, unrelated to core operations"

Risk Note: "Every international investor has been rug-pulled at least twice in their career by Japan's 'this time is different' narrative." However, Modest believes that the reform starts with "the first step" and may bring several years of opportunities for Japanese equities.


Mentioned Positions

Position Guest Stance Key Data
NVIDIA Short-term bullish (lab spending willingness), long-term cautious (competitive risk) Estimated ~$200 billion revenue, 80% gross margin; current P/E ratio around 30x
Apple Neutral (hard to disrupt) P/E ratio around 30x; "No standalone device has ever beaten the smartphone"
Microsoft Positive (AI beneficiary) P/E ratio around 30x; Azure AI incremental revenue quantifiable
Google/Alphabet Positive (AI beneficiary) P/E ratio around 20x
Meta Neutral (proponent of open-source strategy) Plans to spend ~$40 billion to "quickly follow" AI
Amazon Neutral (post-pandemic capital cycle case) "Rebuilt UPS within two years"
ASML Positive (choke point) Cited as a case of "a moat that is hard to disrupt"
Ball / Crown Watch (beverage can capital cycle) Capital expenditure rising from $600M/$500M to $1.5 billion
Novo Nordisk (Ozempic) Positive (economic potential), cautious (compliance) Compliance rate may be only 35%

Judgments Worth Remembering

1. "The market persistently assigns permanent valuation multiples to temporary earnings—this is a confusion I have never been able to understand." —Modest Proposal

  • Support: Commodities in the mid-2000s, digital demand during COVID, current AI computing power—each time it's "surge in demand vs. rigid supply" → capital floods in → supply glut.

2. "NVIDIA's long-term problem is: Can one company become the permanent bottleneck for the five largest companies in history? Someone in every garage is thinking about how to bypass it." —Modest Proposal

  • Support: Some have already found ways to run part of GPU functions on CPUs; a $200 billion profit pool with 80% gross margins is the most tempting target in capitalism.

3. "Open-source models being 'good enough' vs. frontier models being significantly ahead—the economic consequences of these two paths are vastly different, but the investment community has almost written off the first path." —Modest Proposal

  • Support: If frontier models win, value flows to Google/Microsoft; if open-source wins, value flows to the application layer. OpenAI insiders say AGI will arrive in 3-5 years, contrary to market consensus.

4. "The capital cycle distortions caused by COVID are far broader than I imagined—beverage cans, freight, Amazon's delivery network—every industry is experiencing 'shortage followed by glut.'" —Modest Proposal

  • Support: The beverage can industry jumped from 2-3% annual growth to 10%, capital expenditure doubled, and capacity became excessive after demand fell; the freight market is still digesting two years later.

5. "The economic impact of GLP-1 could rival AI, but human behavior is the biggest obstacle—even cancer drugs have only a 60-70% adherence rate." —Modest Proposal

  • Support: If the adherence rate is only 35%, the number of new patients needed annually to reach 50 million Americans on the drug is extremely large.

6. "The rationale for institutional capital flowing to private markets has shifted from 'optimizing the efficient frontier' to 'smoothing returns and avoiding volatility'—this may be the most distorted phenomenon in capital markets." —Modest Proposal

  • Support: Over the past three years, capital returned from private markets has been about $300 billion less than capital deployed; private equity with 4-6x leverage has generated only 1.6-1.7x MOIC, while the Nasdaq 100 achieved 5x MOIC without leverage.

7. "The Japanese stock market may be a structural opportunity—the TSE is 'shaming' companies into submitting capital management plans, and Japanese companies are generally overcapitalized, with zero buybacks and low dividends." —Modest Proposal

  • Support: Every international investor has been rug-pulled by Japan twice, but this reform starts from "step one" and could last for years.

8. "The market structure has changed: the definition of 'small-cap' has shifted from a $1 billion market cap to $15 billion—companies without a story are essentially abandoned below that threshold." —Modest Proposal

  • Support: The trend is more persistent and more severe; attention is concentrated between "current hot spots" and mega-cap stocks, with the middle ground being ignored.