A Dual-Pivot Strategy for the AI Era: Experiment Fast, Evolve the Portfolio Slowly
AI has reduced not only the cost of intelligence but also the “cost of being wrong.” Yet as experimentation becomes easier, usage and total demand can surge, making it essential to design for both rapid exploration and patient investing.
In the AI era, we should rapidly repeat experiments with low failure costs.
Proven, high-quality companies should be held for the long term with low portfolio turnover.
Trying more while trading less is the dual-turnover strategy for the AI era.
How AI Changes the Cost of Optimism and Pessimism
In the past, it was natural to gain sufficient confidence before taking action. Turning an idea into a product and bringing it to market required substantial money and time.
AI is changing this calculation.
The cost of misplaced optimism is falling.
When you act because you believe something will work but it fails, the cost of that failure is lower. Using AI to build prototypes, write code, and create content reduces the time and labor required to test a hypothesis.
The cost of misplaced pessimism is rising.
If you decide something will not work and therefore do not try it, the loss can be significant when a competitor discovers the opportunity first. If others are testing dozens of hypotheses at low cost while you wait until you feel certain, you will inevitably fall behind in the speed of exploration.
Does that mean we should always act even without sufficient confidence? Not necessarily, but there is less reason to keep hypotheses that can be tested on a small scale trapped in the conference room for too long.
AI has made intelligence cheaper and, more importantly, made being wrong cheaper.
Why the Bill Can Grow Even as Unit Costs Fall
Improvements in AI efficiency reduce the computing power and cost required for each task. However, as prices fall, even tasks that were previously uneconomical become viable, increasing overall usage.
The Jevons paradox is the phenomenon in which greater efficiency lowers the unit cost of using a resource, but total resource consumption actually rises because usage grows even faster. In 1865, William Stanley Jevons observed that improvements in steam-engine efficiency could expand the scope of coal use rather than reduce coal demand. Applying this concept to AI computing is an interpretation that extends this historical logic to modern technology. Yale University Energy History
The following diagram shows how AI costs and usage translate into total infrastructure demand.
flowchart LR
A["① Improved AI efficiency"] -->|Lowers cost per task| B["② Lower unit costs"]
B -->|Unlocks latent demand| C["③ More tasks and greater usage"]
C -->|Consumption growth outpaces efficiency gains| D["④ Higher total compute demand"]
D -->|Requires more facilities and services| E["⑤ Greater demand for data centers, semiconductors, and electricity"]
E -->|Passes supply costs on to customers| F["⑥ Potential increase in total costs"]
In the first stage, the same task is completed at a lower cost. The lower unit cost then attracts new users and use cases. When the increase in the number of tasks exceeds the efficiency gains, total compute demand and demand for AI infrastructure expand.
This is similar to driving a more fuel-efficient car but greatly increasing the number and distance of trips. The cost of driving one kilometer falls, but the monthly fuel bill may still rise.
Why does the bill grow when prices are falling? Because consumption is increasing faster than unit costs are declining.
An Era That Demands Higher Experiment Turnover
In an environment where the cost of being wrong has fallen, it makes sense to test more hypotheses over shorter cycles. AI should be used not merely as a way to reduce labor costs, but as a tool for expanding the scope of exploration and the number of experiments.
| Category | Traditional Approach | Approach in the AI Era |
|---|---|---|
| Sequence of execution | Acted only after gaining sufficient confidence. | Run a small experiment first, then build evidence. |
| Handling failure | Focused on avoiding failure itself. | Manage both the cost of failure and the speed of learning. |
| Role of AI | Used to reduce the cost of existing work. | Used to test more hypotheses and options. |
| Key metrics | Evaluated primarily by cost per task. | Evaluate both learning speed and real value creation. |
However, conducting more experiments does not necessarily mean creating more value. The efficiency gains from increased usage must exceed the additional AI bills and the added costs of design, manufacturing, and software.
Personally, I suspect the success or failure of AI adoption will depend not on how extensively AI is used, but on how quickly an organization abandons flawed hypotheses and scales valid ones. I could, of course, be wrong, but organizations that measure validated results rather than usage are more likely to maintain a long-term lead.
Why Investment Portfolio Turnover Should Be Low
In business operations, experiment turnover should be high, but investing requires the opposite principle.
Portfolio turnover is a metric that indicates how frequently the securities in a portfolio are replaced over a given period. Under the official calculation used for funds, it is generally calculated by dividing the lower of annual purchases or annual sales by the portfolio’s average monthly value during that period. U.S. SEC Form N-1A
Therefore, turnover declines when trading volume decreases or the value of assets under management increases. However, reducing unnecessary trades is something investors can directly control, unlike simply relying on asset appreciation to increase the denominator.
Quality investors pursue low portfolio turnover, holding outstanding companies with competitive advantages, financial strength, and growth potential for the long term rather than trading frequently. This does not mean blindly holding just any company. It means allowing time to work for companies whose investment thesis remains intact.
The Long-Term Advantages of Low Turnover
Low portfolio turnover reduces transaction costs, curbs emotional decision-making, and gives long-term compounding time to work.
| Long-Term Advantage | How It Works |
|---|---|
| Reduces transaction costs. | Fewer purchases and sales lower commissions and other trading expenses. |
| Manages the tax burden. | In taxable accounts, it reduces taxable gains that frequent trading may realize. |
| Prevents emotional investing. | It reduces impulsive performance chasing and panic selling in response to market volatility. |
| Focuses on intrinsic value. | It emphasizes growth in corporate earnings and cash flow rather than short-term stock prices. |
| Strengthens long-term compounding. | It provides the holding period needed for returns to generate further returns. |
The SEC’s investor guidance explains that high portfolio turnover can lead to greater transaction costs and tax burdens. It also notes that even small fees can accumulate over time and make a significant difference in portfolio value. Investor.gov Guidance on Portfolio Turnover, Investor.gov Guidance on Investment Fees
Of course, low turnover alone does not guarantee high returns. If a company’s competitive position has deteriorated or the original investment thesis was wrong, holding it for a long time may compound losses rather than returns.
Ultimately, the essence of low turnover is not simply avoiding trades. It lies in carefully choosing what to hold for the long term and distinguishing market noise from fundamental changes in a business.
Try More While Trading Less
The key in the AI era is not to raise or lower every type of turnover indiscriminately. Exploration and experiments with low failure costs should be repeated quickly, while carefully selected, high-quality companies should be held for the long term with low portfolio turnover.
In the same vein, AI can increase experiment turnover in the process of discovering and analyzing companies. But if investment decisions are reversed every day, the enhanced exploratory power provided by AI can turn into excessive trading.
The fact that AI has made being wrong cheaper does not mean every decision should be made lightly. Easily reversible experiments should move quickly, while capital allocation that depends on compounding should move slowly.
One-line comment. In the AI era, it may be wise to test-drive widely—but once you have chosen a good car, avoid switching vehicles repeatedly before reaching your destination.