"The institutionalization of volatility as a proxy for risk has, unfortunately, incentivized many managers to design strategies aimed less at true investment success and more at achieving results that allocators value. While this alignment isn’t inherently harmful, it can become rigid and counterproductive—particularly during periods of crisis, when conventional metrics often fail to capture the bigger picture."
The key: Most investors discount future value too heavily because of hard-wired psychology. This creates an opportunity for those who can think in decades.
The math of compounding favors high returns on capital over high growth rates. 27. How do you think about valuation? Valuation is important, but it’s not everything. The goal isn’t to buy the cheapest businesses—it’s to buy quality businesses at prices that allow for satisfactory long-term returns. My framework has three levels:
The magician creates a circumstance in which you trick yourself. 69. How do you stay within your circle of competence? The circle of competence is about knowing what you know—and more importantly, knowing what you don’t know. The boundaries: I understand:
The process itself learns and improves, independent of who’s using it. This is why we can be confident that 10 years from now, even with different people working here, the same quality of analysis and decision-making will persist. The knowledge is in the system, not just in people’s heads.
The key: Selling should be based on one of four things: broken thesis, better opportunity, extreme valuation, or capital needs. Nothing else is a good reason.
When the portfolio is down 25% and friends are making money in index funds, will I stay invested?
The investment implications: For tech businesses, I focus on network effects and switching costs because these are the most durable. Technology advantages alone are too fragile. For traditional businesses, I focus on scale and brand because these take decades to build and decades to erode.
In the right circumstances and environment, hard work is hard, but also somehow fun. Stress isn’t stress anymore. The world changes. My job is creating the environment where people with tigers in their tanks can do their best work, where being uncomfortable is just information. Where the satisfaction comes from understanding, not from quarterly results. Where we can push the same rock up the same hill for years because we want to see what’s at the top.
The key: Conviction comes from understanding the business deeply enough that you can distinguish between thesis breaks (sell immediately) and noise (ignore or add). 107. What psychological traits are necessary for this strategy? Several psychological traits are essential: High conviction tolerance:
The goal is a team where adding one more person would make us worse, not better. That’s how you know you’re at the right size.
If the portfolio is down 25% while the market is up 15%, will I have the conviction to stay invested—or even add capital?
start small or invest elsewhere. If the honest answer is
is lower than you think. The bar for relentless is higher than any firm I've heard of. I love this business precisely because it's hard to do properly. It requires some kind of desire to know. A competitiveness whose source is unknown. I assure you, if you don't have that, a competitor of ours somewhere will. The reality is that the source varies from person to person. But without
this might be right for you. If the honest answer is
Where are the Walmart billionaires? With deference to the Walton family, we specifically mean, where are the non-family Walmart billionaires?
The reading process: First pass - Strategic narrative (2-3 hours): • Read the shareholder letter: How does management think? • Read business description: How do they describe the moat? • Read risk factors: What are they worried about?
The architecture that worked across scarcity breaks when the environment provides surplus. But the system as a whole remains functional because its underlying logic is so stable. It errs, but it errs consistently. And consistency, even flawed consistency, allows adaptation. In machine learning, we sometimes forget that robustness cannot be bolted on as an afterthought. It must be woven into the structure of the system. A model with billions of parameters may perform
A chip that can adapt locally in real time, without retraining the entire behemoth in the cloud, is not just a cool demo. It is a different cost curve for intelligence. When that happens — not if — the world will discover that
fiber had been lit. Internet traffic had grown from 100 petabytes per month in 1996 to over 150,000 petabytes per month in 2020—a 1,500x increase over 24 years. The compound annual growth rate of 33% annually was less than the
The key biases I fight: Confirmation bias: Once I like an idea, I want to confirm it. I force myself to actively seek disconfirming evidence and talk to skeptics.
we will build the biggest models and rent them to everyone
Worldly wisdom teaches that it is better for reputation to fail conventionally than to succeed unconventionally.
We are building the most advanced network on Earth,
The hardest part of this strategy is knowing that “temporarily” might mean 1-3 years of underperformance. It might mean watching opportunities you passed on deliver great returns. It might mean explaining to partners why you’re not changing the approach even though it’s not working right now. This difficulty is precisely what creates the opportunity. If it were easy, everyone would do it, and there would be no edge.
In this domain, on this hardware, with this kind of data, a different style of learning is simply better.
is exposed to a change in physics. The ones that treat LLMs as one powerful modality among several, and are willing to adopt more efficient or more grounded learning methods where they make sense, will quietly gain structural advantage as the noise dies down. The implication for investors is that you should start thinking about AI exposures the way you think about factor exposures. Are you unintentionally long
That kind of absolutism is for panel discussions, not capital allocation. The implication is that any firm whose story is entirely wrapped around
industries where great businesses exist. Overvaluing management quality: I place enormous weight on management teams. Sometimes a great business can succeed despite mediocre management, and I might miss those opportunities. Perfectionism in research: I want to understand everything before investing. This thoroughness is usually a strength, but it sometimes causes me to miss opportunities where
In this way, we are beginning to wonder whether it is worthwhile and perhaps very profitable to extend the holding period of our investments, specifically for those companies that have very high terminal values, and that even when discounted using a higher risk-free rate, have higher target prices than our current underwriting.
Diversification is protection against ignorance. It makes little sense if you know what you are doing.
The research on this is quite compelling—as new independent risks are added to the bundle, the large majority of diversification benefits are obtained without sacrificing the quality of the portfolio. I have insight into just a few things. Expanding beyond this does not decrease risk, it increases it.
In machine learning, value functions are conceptually recognized but practically marginalized. They are treated as components of algorithms rather than essential aspects of agency. But evolution's lesson is that intelligence cannot function without an internal system for evaluating outcomes. A mind that cannot prefer one state of the world over another is not a mind that can act effectively. Humans are able to behave, decide, and adapt because we have deeply ingrained,
The compounding effect: High returns on capital + Long runway for reinvestment = Compounding machine A business earning 30% returns on capital that can reinvest most of its earnings at those same returns will compound intrinsic value at close to 30% per year. Do this for 10-15 years and you create extraordinary wealth.
A company growing 50% that will hit a wall in three years is less interesting than a company growing 15% that can sustain it for 15 years. Compounding is about duration, not just rate. The type of growth also matters enormously:
The worst permanent losses come from leverage forcing sales at the bottom. Misunderstanding the business: You thought you understood it, but you didn’t. This is the most common source of permanent losses.
Magic is just someone spending more time on something than anyone else might reasonably expect.
The nuance: valuation is about forward returns, not backward-looking metrics. A “cheap” stock can get cheaper if the business is deteriorating. An “expensive” stock can deliver great returns if the business quality exceeds what the market understands. The practical application: I’ll pay up for quality, but there’s a limit. At some price, even the best business becomes a poor investment. The art is knowing where that line is for each specific business.
The financial manifestation of brand value: • Higher prices than competitors for similar products • Lower customer acquisition costs (people seek out the brand) • Higher customer lifetime value (they buy repeatedly) • More resilience during recessions (they cut other things first)
A new competitor couldn’t just hire sales reps and replicate these relationships. Trust in medical products takes years to build.
understanding would suffice. The key is knowing your blinders exist and trying to counteract them. I force myself to look at
I remember when the Great One in Omaha went long and the Great Market Animal known as Druck went short. Times have changed. But below the surface, they've been exploring architectures that don't fit neatly into the "bigger transformer on more GPUs" story at all. Neuromorphic experiments (yes, that is a real term), in-memory compute, event-driven chips, hybrid classical-quantum pipelines:
The tectonic plates: Different parts of the market move at different speeds, creating stress. Growth stocks might soar while value lags. Tech might boom while industrials stagnate. These divergences create pressure.
The process itself learns and improves, independent of who’s using it. This is why we can be confident that 10 years from now, even with different people working here, the same quality of analysis and decision-making will persist. The knowledge is in the system, not just in people’s heads.
What would it take to build an apprentice that can learn anything important?
we can refine those priors with reinforcement learning. In other words, LLMs get us close enough often enough that we can bolt on experience afterwards. The problem is deeper. To serve as a prior, there has to be an underlying reality you are approximating. You need something that counts as actual knowledge. In the large-language-model framework, there is no definition of actual knowledge. There is no criterion for a
By 2001, 360networks will generate $5 billion in annual revenue with 70% EBITDA margins. This is a once-in-a-generation opportunity.
National boundaries are becoming irrelevant. Data wants to be global, instantaneous, and unlimited. We will be the first true global communications company.
Why are we spending billions on this style of compute if there is a cheaper way to train and adapt in our domain?
In markets, however, the pace of time can accelerate beyond the measured progress of normal industrial logic. The change may not be false; it might be very real. But once there’s no practical barrier to
The layers above it, the ones closest to the user, generate much smaller gains yet often do so more reliably. This discrepancy will not last forever. Eventually, the economics will migrate down the stack as competition increases and the cost of intelligence declines. But for now, the financial architecture of AI reflects the classic pattern of early-stage infrastructure revolutions. The first to profit are those who supply the indispensable tools, not those who wield them.
134 quotations from David Levin Steinberg's investment philosophy
