Anthropic’s forthcoming stock market listing has brought an old question back into view. If a company is genuinely exceptional, does that make its shares a good investment? We think the two are related, but they are not the same thing and confusing them is one of the more expensive mistakes a long-term investor can make.

A durable moat is a claim about the business, not a promise about the share price. Warren Buffett’s “economic moat” idea captures this. A great company is an economic castle whose returns are protected from competitors by a wide, lasting advantage such as its brands, distribution, cost position, network effects, or switching costs. He argued that capitalism guarantees attacks on any business earning high returns, so the question is not merely whether a company is growing, but whether its advantage can endure.

Microsoft’s moat, for example, rests on switching costs, distribution, developer ecosystems, and an installed base that makes its software the default infrastructure of much of the corporate world. Those advantages do not disappear because a quarter disappoints or because investors rotate away from large-cap technology. Operating profit of more than $128 billion in fiscal 2025 is not something a competitor can conjure quickly.

Unilever’s moat is different but real. Brands embedded in daily habits, shelf space built over generations, distribution across emerging and developed markets, and the ability to price modestly ahead of cost inflation. That is not the explosive economics of software, but it is a durable system, where consumers reach for familiar products, retailers allocate space to reliable sellers, and the company reinvests cash into brands that remain difficult to displace.

The enthusiasm around AI listings is why this matters now. Investors who had expected Anthropic to come to market at around $2 trillion are reported to be discussing something nearer $1.5 trillion, and the final price is not yet set. Whether Anthropic will have a sustainable moat over decades is yet to be proven. In our opinion, even the lower figure presumes it will, and that leaves us nervous as investors, whatever we think of the technology.

The mistake is to translate “durable competitive advantage” into “the stock rises every year.” A moat determines whether a business can earn attractive returns over long stretches; the share price also reflects starting valuation, interest rates, sentiment, index flows, currency, and the market’s shifting appetite for a company’s story.

Microsoft can compound earnings impressively and still see its multiple compress. Over the five years to June 2025 its shares returned well ahead of the S&P 500, but the path included a fall in the second year. If you owned that business, you still endured a flat-to-negative stretch. The moat did not fail, expectations and valuation simply reset.

Unilever illustrates the same principle from the opposite direction. Its underlying sales grew in 2025, yet reported turnover declined 3.8% because currency and disposals overwhelmed operational progress. The brands remained valuable whilst the reported numbers and the share price could still be hostage to foreign exchange, portfolio changes, and a market that prefers faster growth.

Over a year, price can be mostly noise. Over a decade, it becomes harder to ignore whether the business is actually defending and extending its advantage. Microsoft’s advantage shows up in recurring enterprise relationships, high-margin software economics, and the ability to fund new platforms from existing profits. Unilever’s shows up in brand pricing power, cash generation, and the persistence of products bought weekly in millions of households.

In the portfolios we manage, we therefore keep two questions separate. First: is the moat intact? For Microsoft, we ask whether customers remain locked into its ecosystem and whether new products reinforce it. For Unilever, we ask whether brands retain consumer loyalty and whether margins reflect genuine pricing power rather than temporary cost timing. Second: is the price sensible? A wonderful company bought at an excessive valuation can deliver poor returns, and a solid company bought cheaply can reward patience even without dramatic growth.

AI businesses may possess genuine moats, proprietary models, scarce compute, proprietary data, distribution through existing platforms, and high switching costs once customers embed AI deeply into their workflows. But a moat does not automatically make the share price rise from here. Many AI-linked companies already trade as though every ambitious forecast must be met. Enormous capital spending, rapid revenue growth, expanding margins, and continued dominance are all priced in before they are proven. If adoption is slower, competition intensifies, or returns on AI investment disappoint, the business may remain strong while the stock falls simply because reality fails to match expectations.

Buffett’s lesson applies with particular force at present. Identify the castle, inspect the moat, and then ask what future perfection the current price already assumes. For our portfolios, that means we are content to hold excellent businesses through dull or difficult years, provided the moat holds and we did not overpay to begin with. It also means we would rather watch a new AI company prove its moat in public, and wait for the price to meet the evidence, than pay today for decades of perfection.