저자: RayDalio · 외국인
요약 보기
AI 기술 자체는 실재하고 경제적 영향도 크지만, 현재 밸류에이션이 수익화 타임라인을 앞질러 AI 관련 대형주 조정이 있을 것이라 주장. S&P500 상위 7개 기술주 비중 약 40%(1972년 이후 최고)를 근거로 제시. 대안으로 실버, 미국 에너지 인프라, 단기채·TIPS 등을 언급.
다듬은 글
I want to tell you about a specific moment in the spring of 2000. I was sitting in my office at Bridgewater and I had just finished reading through the latest earnings reports from a group of technology companies that were the most talked about names in the market at that time. Companies with extraordinary valuations.
Companies whose stocks had multiplied 10, 20, 30 times in the previous 18 months. Companies that were on the cover of every financial magazine and in every dinner party conversation across America. And the thing that struck me most was not the valuations, which were obviously extreme. It was the certainty, the absolute unquestioned certainty that this time was different, that the old rules of valuation no longer applied, that the technology was so transformative that any price was justified.
I had seen that certainty before in 1987, in the Japanese real estate bubble of the late 1980s, in the Latin American debt bubble of the early 1980s, and in 1982 when I lost everything because I was so certain I was right about something that the market had not yet agreed with. The certainty itself, the absolute consensus that the direction was obvious and the old rules did not apply, was the most reliable indicator I had ever found that the risk reward had shifted in ways that most participants could not yet see.
I reduced my technology exposure in early 2000. The NASDAQ fell 80% over the following 2 years. I am not telling you this story to prove I was smart. I am telling you this story because I am sitting in my office right now reading the latest reports on AI companies and I recognize the certainty. I recognized the dinner party conversations.
I recognized the magazine covers. And I want to explain exactly what I see, exactly what I did about it, and exactly what I bought instead. I am Ray Dalio. I built Bridgewater Associates into the largest hedge fund in the world, managing over 124 billion dollars. I predicted the 2008 financial crisis.
I navigated the dot-com collapse. I have studied every major asset bubble in the last 500 years of economic history. In 1982, I made the most catastrophically wrong public prediction of my career, lost everything, had to borrow $4,000 from my father, and rebuilt from nothing. That failure, more than any of my successes, shaped how I think about markets.
The lesson I took from 1982 was not to be more careful about my predictions. The lesson was that certainty is the most expensive emotion in investing. The moment a thesis feels obvious, and the crowd has assembled around it, the risk-reward has already shifted, regardless of whether the underlying thesis is correct.
Let me start with what is true about artificial intelligence, because I want to be precise. The technology is real. The economic impact will be transformative. I have studied every major technological transition in the last 200 years, and AI will rank among the most significant. The productivity gains will be substantial.
The economic value created will be enormous. I am not questioning the technology. I am questioning the price at which that technology is currently being valued by financial markets, and the assumptions baked into those valuations, that I believe are wrong in ways that will matter significantly to investors over the next two to three years.
Here are the specific numbers. The seven largest technology companies in the S&P 500, the ones most directly associated with the AI narrative, represent approximately 40% of the entire index by market capitalization. 40% of 500 companies. That is the highest concentration in the index since 1972. The current price-to-earnings ratios of these companies, even using forward earnings that assume AI delivers transformative productivity gains on an accelerated timeline, are at levels that imply a very specific future that has never actually arrived on schedule for any major technology in the last 100 years.
Not the automobile, not not the internet, not smartphones. Every transformative technology in history has followed the same pattern. The initial deployment period is followed by evaluation correction as the timeline for economy-wide adoption proves longer than the euphoric projections assumed. The technology then delivers on its promise.
But the investors who bought at peak valuation did not benefit from that delivery because they paid for it in advance and then waited years for it to arrive. The investors who compounded the most wealth from the internet revolution were not the ones who bought in 1999 at peak Nasdaq. They were the ones who bought the survivors in 2002 and 2003 at the bottom of the correction that was inevitable once the gap between price and reality became impossible to sustain.
Michael Burry, the investor who predicted the 2008 housing collapse, posted publicly on Substack this month that the AI market feels like the last months of the 1999 to 2000 bubble. He has taken put options on the semiconductor ETF betting on a significant decline within 8 months. He is not alone. Jamie Dimon, the CEO of JP Morgan, said publicly this spring that there is too much exuberance in AI valuations.
Ray Dalio, which is to say myself, said publicly that we are definitely in a bubble, not possibly. Definitely. These are not fringe contrarian views. These are the assessments of three of the most credible macro investors alive, and we are all saying the same thing at the same time. Now, I want to walk you through the specific mechanism that I believe will produce the AI valuation correction that I am positioning for.
Because understanding the mechanism is more important than knowing the conclusion. If you understand the mechanism, you can update your view as new information arrives. If you only know the conclusion, you are just following someone's opinion. The mechanism starts with the AI infrastructure buildout.
The $3 trillion investment in data centers, semiconductors, cloud computing, and AI infrastructure that is currently underway is being funded primarily through corporate debt issued at interest rates that assume the revenue from AI applications will arrive on a specific timeline. That assumption is the key.
The revenue is coming. The question is the timeline. And the timeline matters enormously because the companies that borrowed at current interest rates to build this infrastructure need the revenue to service that debt. If the revenue timeline extends by 18 months or 2 years relative to current projections, which is the historical norm for major technology transitions, the debt service math becomes strained.
The companies that are most leveraged to the AI buildout, the semiconductor manufacturers, the cloud computing providers, the data center operators, face the most significant risk from a timeline extension because their capital structure was built on the assumption of rapid adoption. The second part of the mechanism is what happens to passive investors when the seven companies representing 40% of the index begin to correct.
Most retail investors believe they own a diversified portfolio through their S&P 500 index funds. They do not. They own a highly concentrated position in seven technology companies. When those seven companies reprice, the entire index reprices with them. The passive money machine that has provided mechanical buying support for these stocks, tens of millions of Americans contributing to their 401K every paycheck, does not stop when the stocks fall, but it slows.
And the momentum that drove the valuations to current levels reverses at the same speed in the opposite direction when the institutional sellers who are positioned in these names need to raise cash or take profits. The 2000 to 2002 Nasdaq correction saw stocks that everyone believed in and every analyst had buy ratings on fall 80 to 90% not because the technology failed, because the price had been pushed to a level that the actual timeline of adoption could not justify.
The third part of the mechanism is the macro backdrop. In 2000, the Federal Reserve had room to cut interest rates aggressively when the tech bubble burst. Rates went from 6 and 1/2% to 1% over 3 years, providing a significant monetary stimulus that cushioned the broader economic impact. In 2026, the Federal Reserve is operating in a stagflationary environment, where cutting rates would accelerate inflation and holding rates risks recession.
The backstop that cushioned the 2000 correction is less available this time, which means that if the AI valuation correction begins, the policy response will be more constrained and the impact on the broader market will be less buffered. Now, let me tell you what I bought instead, and I want to be very specific because vague advice to diversify is not useful.
I am going to tell you exactly what I see as the assets that carry lower crowd risk, better risk reward at current prices, and structural exposure to the same macro forces that made AI stocks attractive without the valuation extreme that now makes them risky. The first thing I bought is physical silver.
I want to explain why silver specifically rather than gold or other precious metals. Gold has already made a 24 times return from its 2001 lows. It is no longer the contrarian trade. It is the consensus trade. Retail investors are buying it. Central banks are buying it. Your neighbor is asking about it.
When your neighbor asks about an asset, a significant portion of the move is already happened. Silver is different. The gold to silver ratio is currently at one of its most historically extreme levels. Silver is historically cheap relative to gold, and silver carries an industrial demand component from solar, electric vehicles, and AI infrastructure itself that gold does not have.
The solar industry consumes roughly 20% of annual global silver supply. AI My centers require silver in quantities that were not modeled 5 years ago. The global silver market has been in structural deficit for six consecutive years. When these three forces converge, the monetary premium, the industrial demand, and the supply deficit, the move that follows is typically more powerful than gold on a percentage basis because the starting point is more undervalued.
The second thing I bought is domestic American energy infrastructure. The same geopolitical disruption that is putting pressure on the monetary system is also creating an extraordinary opportunity for domestic American energy producers. United States shale companies, LNG exporters, and domestic energy infrastructure operators are the direct beneficiaries of Middle Eastern supply disruption.
Their revenue is tied to global energy prices that are rising. Their costs are fixed in dollars. The margin expansion is automatic, and they are completely insulated from the geopolitical risk that is creating the supply disruption. Because every barrel they produce and every unit of gas they export comes from American soil.
In the 2022 energy shock, when the S&P 500 fell 20%, the energy sector gained 60%. The same mechanism is operating now with a more sustained supply disruption than 2022. The third thing I adjusted in my positioning is my duration exposure in fixed income. Long duration bonds are directly in the path of the two risks I have described.
If AI valuations correct, and the Fed is forced to choose between fighting inflation and supporting markets, long duration bonds face either the inflation risk or for credit risk of a government that is already paying $1 trillion per year in interest on $40 trillion of debt. Short duration instruments that roll over quickly and adjust to new rate environments carry far less of this risk.
TIPS that adjust their principal automatically as inflation rises carry less of this risk. These are not exciting trades. They do not have a narrative that makes them compelling at dinner parties. And precisely because of that, they are available at prices that reflect their actual risk, rather than prices inflated by crowd enthusiasm.
I want to address something that I know is forming in your mind right now. If you are saying the AI bubble will correct, are you not just being the person who said the internet would fail? Are you not making the same mistake that the people who dismissed the internet in 1998 made? It is the right question, and I want to answer it precisely.
The people who said the internet would fail in 1998 were wrong about the technology. I am not saying AI will fail. I am saying the valuations have run ahead of the timeline on which the revenue will arrive. The people who said the internet would succeed in 1998 and bought Nasdaq stocks at peak valuation in 1999 were right about the technology and lost 80% of their money.
The people who said the internet would succeed and bought the survivors in 2002 at the bottom of the correction made extraordinary returns. Being right about the technology is not sufficient. The price at which you own it determines whether you benefit from being right. Let me close with the framework that I apply consistently and that I believe is more useful than any specific trade recommendation.
The most reliable indicator I have found over 50 years of investing that a trade has become crowded and the risk-reward has shifted is not valuation alone. Valuation can stay extreme for longer than anyone expects. The most reliable indicator is the combination of three things simultaneously. Consensus agreement that the thesis is correct, extreme positioning by financial market participants who are long because the asset is going up rather than because of a fundamental thesis, and a price level that is pricing in outcomes that require things to go right on an accelerated timeline that has never actually been achieved in comparable historical cases.
Right now, AI stocks check all three boxes and that is why I reduced my AI exposure and rotated into assets that carry similar macro exposure without the three indicators that consistently precede major corrections. Comment below right now and tell me this honestly. Do you currently own AI stocks or an S&P 500 index fund and have you ever calculated what percentage of your portfolio is effectively a concentrated bet on seven technology companies delivering on the AI promise within the next 2 years.
Write the percentage, not a story, the percentage. Because the act of calculating it honestly is the first step toward understanding whether your current positioning is a considered bet or an accidental concentration that happened through passive investing. Subscribe, share this with one person who thinks their index fund is diversified and does not know that 40% of it is in seven AI adjacent stocks and go make a decision.
The technology is real, the price has moved ahead of the timeline, the crowd is assembled. What you do with that combination of facts is entirely up to you.
Comments 0