#170 – Textile Mills & GPUs

AI Value Capture

Hi 👋 - AI is generating tons of value – but value creation isn’t the same as value capture. Warren Buffett’s 1985 textile post-mortem offers a timely lesson for today’s AI capex supercycle. As always, thanks for reading.

---

Buffett’s Miserable Choice

In commodity businesses, productivity gains often show up as lower prices for consumers, not higher margins for owners.

In 1965, Warren Buffett took control of Berkshire Hathaway, a Massachusetts-based textile mill making commodity cotton textiles and later rayon linings. The business soaked up capital and never earned adequate returns. In 1985, he shut the textile operation.

Buffett wrote that the “promised benefits” of productivity-enhancing textile investments were illusory – once rivals matched them, lower costs became lower prices industrywide. Projects that screened well individually became collectively self-defeating¹:

The promised benefits from these textile investments were illusory. Many of our competitors, both domestic and foreign, were stepping up to the same kind of expenditures and, once enough companies did so, their reduced costs became the baseline for reduced prices industrywide. Viewed individually, each company’s capital investment decision appeared cost-effective and rational; viewed collectively, the decisions neutralized each other and were irrational (just as happens when each person watching a parade decides he can see a little better if he stands on tiptoes)...Thus, we faced a miserable choice: huge capital investment would have helped to keep our textile business alive, but would have left us with terrible returns on ever-growing amounts of capital.

Buffett’s mills kept getting faster and cheaper, but his competitors’ mills did too. Productivity gains showed up as lower prices for customers, not profits for owners. This was great for suit buyers, but grim for suit-fabric makers.

Value Creation ≠ Value Capture

That dynamic isn’t unique to rayon linings. A key lesson from Berkshire’s textile business is that value creation doesn’t guarantee value capture.

This applies to technologies both new and old. In adoption waves, the spoils often accrue to the businesses that embed the technology in their operations, not to the average producer of the technology itself.

History rhymes. Investor-essayist Jerry Neumann sees containerization as a parallel to AI today. The technology transformed global trade, yet most ship owners earned meager returns because competition and capex suppressed profitability. Like with Berkshire’s textile mills, the value flowed downstream, in this case to operators who exploited cheaper, faster logistics – businesses like Costco, IKEA, and Walmart.

Applied to AI, Neumann argues that the largest gains are likely to accrue to firms that use AI to drive lower unit costs, faster cycle times, or better client outcomes in knowledge-heavy fields²:

All of which means that investors shouldn’t swim upstream, but fish downstream: companies whose products rely on achieving high-quality results from somewhat ambiguous information will see increased productivity and higher profits…The way to invest in AI is to think through the implications of knowledge workers becoming more efficient, to imagine what markets this efficiency unlocks, and to invest in those. For decades, the way to make money was to bet on what the new thing was. Now, you have to bet on the opportunities it opens up.

Consumer Surplus ≠ Producer Surplus

Another framework is who gets the better end of the bargain: buyers or sellers? Generating consumer surplus doesn’t guarantee producer surplus.

Consumer surplus is the difference between what you’d be willing to pay and what you do pay. Similarly, producer surplus is the difference between what you’re paid and what you’d accept. If you’d pay $7 for a latte that costs $5, that’s $2 of consumer surplus. If the cafe would have sold it for $3, that’s $2 of producer surplus.

Source: Below the Line and Gemini.

For consumers, there’s no doubt that products like Claude, ChatGPT, and Gemini generate oodles of value. ChatGPT easily saved me at least five hours of research with planning a trip to London. If I conservatively valued an hour of my time at New York City’s minimum wage of $16.50, that’s over $80 in value for $20.

Source: Below the Line and Gemini.

What’s less obvious is if my trip planning created any producer surplus for OpenAI.

Markets with elastic supply, intense competition, and low barriers to entry can generate significant consumer surplus with little producer surplus. Two of the three conditions – elastic supply (free or almost all-you-can-eat for $20 a month) and intense competition – hold for AI today. However, capital requirements and the limited expertise for building and training LLMs do create significant barriers to entry.

Some estimates peg the consumer surplus from generative AI at $97 billion in 2024, compared to roughly $7 billion of revenue³.

Ultimately, producer surplus and industry profitability will hinge on variables like improved monetization, competitive intensity, and if LLMs are end up as commodity or differentiated products.

Over the past two decades, tech companies have developed playbooks for capturing attention first and monetizing later. Given AI’s infancy, there’s every reason to believe that monetization will improve, whether through advertising, success-based fees (like Sierra), price increases, or otherwise. We could also be in AI’s Millennial Lifestyle Subsidy phase. Just as the Uber ride from Prospect Heights to Greenpoint in the twenty-tens that used to cost $12 is now reliably $25-$35, we may look back wistfully at when ChatGPT was just $20 per month.

AI & Capital Intensity

All that surplus is financed by a staggering capex bill. Alphabet, Amazon, Meta, and Microsoft are guiding nearly $400 billion on data centers and AI infrastructure this year, with more likely ahead.

Source: The Economist, What if the AI stockmarket blows up?, September 7, 2025.

When thinking about this massive scope of AI investment, I keep coming back to a comment Buffett made at the 2025 Annual Meeting about Berkshire’s gargantuan cash balance⁴:

The problem with the investment business is that things don’t come along in an orderly fashion, and they never will…We’re running a business which is very opportunistic…We would rather have conditions develop where we would have like $50 billion in cash rather than $335 billion in treasuries. But that’s just not the way the business works…If you told me that I had to invest $50 billion every year until we got down to $50 billion in cash, that would be the dumbest thing in the world. Things get extraordinarily attractive very occasionally. The long-term trend is up. But nobody knows – Greg doesn’t know, Ajit doesn’t know, nobody knows what the market is going to do tomorrow, next week, or next month.

The bull case is that AI represents one of those rare moments when it makes sense to deploy massive amounts of capital. But the arithmetic gets trickier as numbers balloon. Generating a 20% return on $400 billion requires an $80 billion profit. What’s more, AI chips have useful lives of three to five years⁵, so there’s a ticking clock element as well.

To paraphrase Buffett, size is the enemy of performance. As the capital base explodes, value capture must too. Otherwise we’ll get great products, grateful customers, and underwater investors.

---

Join 2,500+ readers at companies like Amazon, Google, and Morgan Stanley.

If this helped you think about value capture in AI, send it to one operator or investor who obsesses over margins.

---

More Good Reads and Listens

Jerry Neumann on why AI will not make you rich. The Wall Street Journal on if AI spending will ever pay off. John Huber on how big tech is becoming more capital intensive. Byrne Hobart on vendor financing, like Nvidia’s recent deal with OpenAI. Tyler Cowen on technological diffusion and AI bottlenecks.

Owns shares in Alphabet, Berkshire Hathaway, and Meta.

Photo by Venti Views on Unsplash

1. Berkshire Hathaway, 1985 Shareholder Letter, February 28, 1986.

2. Colossus, AI Will Not Make You Rich by Jerry Neumann, September 2025.

3. The Wall Street Journal, AI’s Overlooked $97 Billion Contribution to the Economy, August 3, 2025.

4. Steady Compounding, Transcript: Berkshire’s 2025 Annual Shareholder Meeting, May 4, 2024.

5. The Wall Street Journal, Spending on AI Is at Epic Levels. Will It Ever Pay Off?September 25, 2025.