A Rare Window into How We Invest
INVESTOR LETTER · 3 AUGUST 2026
Something unusual happened in July. The strongest part of the global market became, for three weeks, the weakest. Semiconductor and AI-related shares, which had led equities for more than a year, fell hard and fast, and the selling reached from the United States to Korea, Japan and China within days.
When markets move like this, most commentary rushes to tell you what happened. We would rather use the moment differently. A sharp correction is one of the few occasions when an investment process becomes visible in real time: what an investment adviser does when prices fall reveals far more than what they say when prices rise. So this piece is less a market update than a window into how we think.
THE FACTS
The main semiconductor index fell more than 20% from its June peak. That sounds alarming until you remember where prices were coming from: the same index had roughly doubled over the previous twelve months, and some individual names had risen several hundred percent. After gains of that size, a sharp pullback is not a verdict on an industry. It is how markets digest very large advances. They always have.
But there is a stranger fact sitting underneath the headlines, and it shapes our entire reading of this correction: company profits went up while share prices went down. The selling was not triggered by bad news from the businesses themselves. During the very weeks their shares were falling, the world's leading chipmakers reported some of the strongest results in their history. One posted a quarterly profit among the largest any company has ever recorded. Another reported record revenue with its production fully booked for the year ahead, and its shares fell double digits in a single session anyway.
When excellent results meet falling prices, the business has not disappointed. Expectations had simply run ahead of even excellent results, helped along by a more hawkish central bank and a nervous mood. That is a valuation correction, not a fundamental one. The world's demand for computing did not fall in July. Only the prices did.
THE AMAZON STORY
Amazon's Tri-Cities expansion — building capacity ahead of demand.
Whenever spending on a new technology reaches historic scale, the word “bubble” appears. It deserves a serious answer, and the best answer we know is not an argument. It is a story most investors think they remember, but whose lesson is easy to forget.
Through the late 1990s and 2000s, Amazon poured every available dollar into fulfilment centres, logistics networks and, later, the data centres of its cloud business. Because it reinvested everything, it reported losses or razor-thin profits year after year. In 2000 alone it lost US$1.4 billion. Commentators called the spending reckless. Some called Amazon a bubble stock that would never earn real money. For nearly two decades, the sceptics appeared to be right.
Source: Amazon annual reports. Figures are approximate and shown for illustration.
Then the arithmetic turned. Once the network was built, each additional parcel and each additional cloud customer travelled over infrastructure that already existed. Spending fell relative to revenue, and profits inflected: from around US$600 million in 2015 to roughly US$59 billion by 2024, close to a hundredfold increase in under a decade. The share price followed the profits, as it eventually always does.
The investors who were rewarded were not the ones who avoided Amazon during its heavy-spending years. They were the ones who understood what the spending was building. The spending was never the problem. The spending was the moat: infrastructure so large and so efficient that no competitor could replicate it.
WHY WE BELIEVE TODAY RHYMES
Today, the largest technology companies are spending historic sums building AI computing capacity, and their share prices are being punished for it, just as Amazon's once was. The natural question is whether this time the money is chasing demand that will never arrive, as happened with telecom fibre around the year 2000.
Here is the difference, and in our view it is decisive: this time the demand is arriving first, and the capacity is chasing it. The clearest illustration came the same week the correction bottomed, when Alphabet reported that customers have already signed contracts worth more than half a trillion US dollars for cloud computing it has not yet been able to deliver. That figure is not a forecast, not a hope, and not an analyst's estimate. It is demand that exists on paper and simply waits for the capacity to be built. Across the industry, the same picture repeats: memory production sold out a year or more ahead, order books at record levels.
|
More than half a trillion dollars of signed cloud contracts — demand Alphabet cannot yet deliver, not a forecast of demand it hopes to find. |
|
Bubbles are built on hope. This buildout is built on order books. |
HOW WE INVEST THROUGH IT
None of this means prices cannot swing violently along the way. July proved they can, and there will be more months like it. What matters is whether a portfolio is built to withstand those swings without being forced into bad decisions. The portfolios we advise are, and in three deliberate ways.
We own both sides of the cheque. The technology giants write enormous cheques for AI infrastructure, and somebody receives them: the chipmakers, the memory producers, the equipment and supply chain behind them. We hold the cheque writers and the cheque receivers together, so we do not need to guess which single layer of the AI economy captures the most value. If value migrates from one layer to another, as it often does in technology, we participate either way.
We surround the growth engine with shock absorbers. The AI holdings sit inside a structure that also holds established compounders whose earnings do not depend on the AI trade at all, together with meaningful allocations to gold, bonds and cash. Each has a job: gold as insurance against inflation and geopolitical shocks, bonds for income and ballast, cash for the freedom to act when opportunity appears. A large part of the portfolio exists not to participate in the AI theme but to protect it. That is why a 20% correction is something we can study calmly rather than react to.
We decide in advance what would change our mind. Discipline is easy to claim and hard to verify, so we state our tests openly. We would reassess the AI allocation if cloud order books stopped growing, if the big spenders cut their plans because customer demand disappointed, if memory pricing rolled over because supply caught up, or if the returns on all this invested capital began to deteriorate rather than improve. In July, none of those things happened. Demand rose. Earnings rose. Order books rose. Prices fell. Price volatility alone is not a reason to abandon a thesis the underlying data keeps confirming.
CLOSING THOUGHTS
Corrections inside powerful long-term trends are uncomfortable, but they are the price of admission. Amazon's shareholders endured many of them on the way to extraordinary returns, and the ones who were rewarded were those who understood what the spending was building. We believe the same discipline applies today.
Long-term investing does not require complexity. It requires clarity about what you own, discipline about why you own it, and honesty about what would make you change your mind. We hope this letter has given you a genuine look at all three. If you would like to discuss how these ideas apply to your own plans, we encourage you to speak with your financial consultant.
|
AT A GLANCE ✦ July's correction was a valuation reset, not a fundamental one — chipmaker profits and order books rose even as prices fell. ✦ The Amazon precedent: heavy build-years are often mistaken for bubbles, but the investors rewarded were the ones who understood what the spending was building. ✦ We hold both sides of the AI value chain, surround it with gold, bonds, cash and non-AI compounders, and have stated in advance exactly what would change our mind. |
With warm regards,
Unicorn InvesCo
NOTE AND SOURCE
Figures cited in this article are drawn from public sources including company earnings releases and earnings calls, Bloomberg, Reuters, CNBC, the Wall Street Journal, the Financial Times and FactSet, and reflect information available as at late July 2026. Chart figures are approximate and shown for illustration. This article is for information only and does not constitute investment advice or a recommendation of any security. Past performance is not indicative of future results.
