AI Meets Buffett

Dear Shareholders:

This year, I want to discuss something I would rather avoid: artificial intelligence.

Charlie Munger and I have lived by a simple, stubborn rule: we never invest in what we do not fully understand. To be honest, I still do not completely understand AI.

But I can no longer avoid it. AI is changing the very logic of "understanding" itself — reshaping the fundamental rules of information, cognition, advantage and risk. So I have chosen to work through this the way we have always worked through things, and share my judgment with you honestly and directly.

I am often asked: have you spent your life profiting from information gaps?

The question sounds fitting, but it completely misunderstands the essence of value investing.

An information gap is knowing something others do not. A cognitive gap is everyone looking at the same thing, and only you seeing a different future.

Let me offer an example that has stayed with me my entire life. In 1951, at age 21, I was flipping through the Moody's Manual in a library when I came across an auto insurer called GEICO. Its P/E ratio was extremely low, but more importantly, it sold car insurance directly to government employees with no middlemen, giving it a structural cost advantage that crushed competitors.

Anyone could walk into a library and leaf through that manual. The information was entirely public, absolutely equal. Most people glanced and flipped past. I stopped. What I saw was not a set of cold numbers, but a durable business model that could last decades and a lasting competitive advantage that was hard to replicate.

That is a cognitive gap.

Now AI is sweeping through, and many claim: the era of Buffett is over. AI can read every annual report, break down every 10-K filing, and screen every risk anomaly in a tenth of a second. Information gaps will be completely erased.

I must correct this misunderstanding: what AI eliminates is only information gaps, not cognitive gaps.

An information gap is about whether you have the data. A cognitive gap is about how deeply you see. AI puts every investor at the same information starting line, giving everyone equal, comprehensive, precise public information. But those standing on the same starting line still see vastly different landscapes, make different directional judgments, and summon vastly different courage to bet.

Tools can distribute information evenly. They can never distribute vision evenly.

My entire investment faith comes down to a single word: moat.

A great company always possesses wide, deep structural advantages. Brand power, network effects, economies of scale, switching costs — these layered defenses take competitors years and billions to breach. They are what allow a company to endure across cycles.

I hold this conviction: AI itself can never be a durable moat for any business.

Look across the history of industrial transformation — electricity, the computer, the internet. Every one was a universal tool that benefited everyone. They reshaped eras and reorganized industries, but none, by itself, created a permanently dominant company. Because tools are public goods. Anyone can plug in. Anyone can use them. Early adopters may seize temporary advantages, but those advantages cannot last — they dilute rapidly as the technology spreads. AI is no different in its fundamental nature.

But AI can deepen an existing moat and build a unique composite barrier.

The real core is not the AI algorithm. It is a proprietary data flywheel. Public corpora and web-scraped data are accessible to everyone and generate no barrier. But exclusive user behavior data, closed-loop supply chain data, proprietary scenario data, and real-time iterating field data — these are scarce resources that money alone cannot buy.

Tesla provides the clearest case. Every vehicle on the road continuously collects and iterates on proprietary real-world driving data. This massive, real-time, exclusive data system forms a barrier no competitor can replicate regardless of capital investment.

AI technology is equal for all, but the compound advantage of data plus context plus iteration is forever exclusive. AI does not build moats. It only amplifies those who already have them.

Of all my investment principles, the most important, the one that has kept us alive, is not stock-picking skill or valuation models. It is the margin of safety.

When I invest in a bridge, I do not only calculate its rated load. I leave ample buffer: even if actual load-bearing capacity is 30% below my estimate, that bridge still stands firm. This is not pessimism. It is clarity. I always acknowledge that humans misjudge, data lies, and the future defies control. The margin of safety is a lifeline left open for everything unknown and every possible mistake.

The biggest problem with AI is this: it has no margin of safety whatsoever.

Every dataset it outputs, every analysis, every conclusion arrives precise to two decimal places, accompanied by a crisp confidence parameter. It is always certain, always complete, always internally coherent. You never see it hesitate, doubt, or leave blank space.

But this perfect confidence is precisely the most dangerous trap in investing.

The core wisdom of investing is knowing what you do not know. Charlie and I have spent our lives repeating: we cannot predict the future. This is not humility. This is our deepest margin of safety. We acknowledge our limits, so we maintain reverence, leave room, cut losses, and wait.

AI cannot do this. It cannot admit ignorance. It cannot reserve redundancy. It cannot hold reverence for the unknown. Whether its conclusions rest on massive samples or thin marginal correlations, it outputs answers with identical absolute confidence.

If you use AI for investment decisions, one thing must be remembered: AI carries no inherent margin of safety. All risk protection must be added by your own hand.

Berkshire's most famous line: our favorite holding period is forever.

Many interpret this as patience or sentiment. It is neither. The underlying logic is compounding, tax efficiency, and long-term certainty spanning half a lifetime. The truth of compounding is never short-term windfall — it is steady long-term accumulation. A consistent 15% a year for fifty years far exceeds fleeting short-term spikes.

AI's natural talent is extreme short-term optimization.

It can solve for the optimal move in every second, every round of trading. High-frequency trading, short-term arbitrage, trend capture — these are its native domain. In quantitative games that have no "long-term," no "conviction," no "cycle," AI is irreplaceable.

But the core of value investing is a terminal judgment spanning fifty years.

Fifty years from now, will people still drink Coca-Cola? Fifty years from now, will society still need insurance systems? Fifty years from now, will railroads still be the most efficient form of logistics infrastructure? AI can disaggregate data, enumerate logic, and analyze probabilities for these questions, but it can never answer them with conviction.

Because the essence of long-term judgment is not calculation — it is a wager. It is betting your cognition, your reputation, your capital on a future that cannot be verified now, one that spans half a lifetime.

I have often said: be fearful when others are greedy, and greedy when others are fearful.

AI has no greed, and no fear. It cannot comprehend human mania and despair. It can only count the probability of rises and falls, trace the curve of sentiment. It can never stay clear-headed when the crowd is euphoric, and never dare to act when the entire market is in panic.

My lifelong partner, Charlie Munger, left us last year. If he could see today's AI, I know he would offer his clearest judgment:

"AI is a supremely useful tool. But letting it think for you is no different from letting a calculator live your life."

AI can read more financial reports in a tenth of a second than a person can in a lifetime. It can capture patterns invisible to the human eye. It can screen every hidden risk that human attention overlooks. It is supremely efficient, supremely precise, supremely comprehensive.

But there is one thing it can never do: at three in the morning, seized by a conviction that is certain, solitary, and sees across the cycle, get out of bed and write two words on a notepad — "Buy."

The end of investing is never calculation. It is judgment.

It is the resolve to hold firm against the roaring tide. It is the courage to say no when the crowd is blind. It is the conviction and faith that endure across half a lifetime.

AI can calculate everything for you, but it can never bear your decisions, hold your convictions, or stand against human nature on your behalf.