저자: 미상 · 외국인
요약 보기
드러켄밀러의 Duquesne 13F 분석: 아마존 주식 약 1,000% 증가, 알파벳 신규 매수, 브로드컴·인텔·마이크론 청산, TSMC 증액·AMD 신규. 하이퍼스케일러 백로그 1조달러, AWS 36.7%·구글클라우드 82% 성장을 근거로 클라우드 플랫폼 수혜를 논하나, 아마존 잉여현금흐름 음수·2026년 CapEx 2,200억달러 리스크도 병기. 중립적 리스크 서술 포함.
다듬은 글
Let's jump right into one of the most fascinating portfolio rotations we've seen this entire year. Stanley Druckenmiller, a legendary macro investor who literally hasn't had a losing year in three decades, is making a sweeping structural realignment in how his Duquesne Family Office plays the artificial intelligence supercycle.
Okay, let's dive into this because the sheer scale of this move is well, it's pretty staggering. Look at these numbers from the recent SEC filings. In just one single quarter, Duquesne's common share position in Amazon absolutely exploded. We are talking about going from a relatively modest 45,800 shares all the way up to over 541,600 shares.
That is an increase of over 1,000% in just 3 months. And that's just the common stock by the way. He also holds massive call options on the e-commerce and cloud giant. Plus, he didn't just back up the truck for Amazon. At the exact same time, he initiated a huge new position in Alphabet, scooping up over 336,000 shares.
So, he is aggressively loading up on these mega cap tech platforms. But here's the thing. To fund a buy that big, you have to sell something. And this is where the plot really thickens. To build these platform positions, Druckenmiller ruthlessly purged specific layers of his chip stack. He completely liquidated his holdings in Broadcom, Intel, and Micron.
But notice the nuance here. He didn't abandon hardware entirely. He actually increased his stake in Taiwan Semiconductor, the foundational foundry physically making the chips, and opened a brand new position in AMD. So, out with the intermediate hardware designers, but he's keeping the core foundries.
Which brings us to the central mystery of this explainer. Why? Why would one of Wall Street's absolute greatest macro investors suddenly dump these highly successful semiconductor designers and pivot so aggressively into the mega cap hyperscalers? I mean, these are the very companies spending billions to buy those chips.
What exactly is he seeing that the rest of the market might be missing? Section one, tracing the thesis. To crack the code of this massive rotation, we really have to trace his documented track record because his entire career is built on tracking where the capital is flowing next rather than where it's already been.
And this brilliantly illustrates his evolutionary approach to the AI trade over the years. Look at this timeline. He was incredibly early on Nvidia, buying massive amounts back in 2022 when its market cap was just a fraction of what it is today. By 2024, he fully exited Nvidia. A move he later admitted was a bit early, famously saying he screwed it up.
But he didn't stubbornly buy back in at the top. Instead, in 2025, he rotated into intermediate chips like Broadcom and Micron. And now, here in 2026, he's wiped those out to fund this massive platform expansion into Amazon and Alphabet. This strategy perfectly encapsulates his core philosophy. As he famously put it, "The wonderful thing about our business is that it's liquid and you can wipe the slate clean on any day." Druckenmiller doesn't fall in love with his stocks.
He doesn't remain loyal to yesterday's narrative. When the macro environment shifts, he just wipes the slate clean and moves his capital exactly to where the next phase of growth is quietly building. Which brings us to the hyperscale cloud shift. The hard economic data is telling us that the next big phase of growth is all about building massive data centers for generative AI.
These hyperscalers are suddenly seeing their economics change in a very dramatic way. Now, we obviously can't read his exact thesis straight from a 13F filing, but the financial clues from Q2 2026 are honestly staggering. Amazon Web Services generated over $42 billion in revenue, but the really wild part is the growth rate, 36.7%.
That's the fastest in 18 quarters. Google Cloud, they're growing at a blistering 82%. But here is the absolute most crucial metric, the backlog. Combined AWS and Google Cloud are sitting on over $1 trillion in pre-sold contracted future revenue. And this isn't just speculative spending, either. Amazon CEO Andy Jassy recently defended their huge infrastructure build-out by breaking down the unit economics, and it makes total sense.
It takes 2 years to build a data center. The servers break even in under 3 years, but they actually last for 5 to 6 years. Most importantly, all this AI capacity is being sold on contracts that last 5 years or more. So, they are effectively building highly profitable pre-sold digital toll roads. Which ties directly back to Druckenmiller's core philosophy of investing 18 to 24 months into the future.
He has this great rule. The biggest mistake investors make is they invest in the present, rather than forward-looking and looking at where the puck is going, instead of where the puck is. If these hyperscalers have locked in a trillion dollars in contracted backlog over the next 5 years, well, the puck isn't really with the chip sellers anymore, right?
The puck is sliding over to the cloud platforms that will actually monetize those chips for the next half decade. The trillion-dollar CapEx risk. Because, as with any legendary trade, there is a massive counter narrative here, and we need to look at the other side of the coin objectively. The biggest risk to all of this is the staggering capital cost of actually building this AI infrastructure.
Now, what's really interesting about this slide is how that heavy capital expenditure is taking an immediate toll on the financials. Looking at trailing 12-month free cash flow, both Amazon and Alphabet have been pulled into negative territory. Amazon sits at negative $7.6 billion, and Alphabet is at negative $5.9 billion, which, by the way, is their first negative free cash flow reading since going public way back in 2004.
And the spending, it's only accelerating. To handle the incredible demand and those sky-high memory costs, Amazon literally just raised its 2026 capital expenditure plan to a mind-boggling $220 billion. I mean, let that number sink in for a second. $220 billion in a single year, largely just to build out data centers, secure energy, and buy hardware.
So, the stark reality of this trade is that Amazon now spends more on infrastructure than the entire company generates in cash. They are literally issuing debt to fund this build out. We want to be clear that neutrally speaking, this is the intense risk of the trade. If that $1 trillion backlog somehow evaporates, or if AI adoption just stalls out, these hyperscalers are going to be left holding the bag on hundreds of billions of dollars in server costs.
Redefining the AI trade. Clearly, when Stanley Druckenmiller looks at that $220 billion spending bill, and then looks at that 36% growth rate, he decides the reward vastly outweighs the risk. He is actively redefining what the AI trade looks like for the second half of the 2020s. So, crucial point is this three-step strategic shift.
Step one, exit the intermediate hardware suppliers, who have already seen their massive multiple expansions. Step two, buy the infrastructure spenders, the Amazons and Alphabets, who are taking the CapEx hit right now, but are building those long-term toll roads. And step three, capitalize as those platforms convert their massive multi-year backlogs into high-margin recurring revenue.
Druckenmiller's ruthless adaptability really is his superpower. Like he says, "I've never hung on to a security if the reason I bought it has changed." The whole reason to own the chip designers was to capture that initial frantic build out phase. Now that the hardware is actually being slotted into server racks and sold on five-year enterprise contracts, the fundamental reason for the investment has shifted over to the cloud providers.
Do these massive 13F moves reveal where the true long-term value of the AI super cycle actually sits today? Are the mega-cap hyperscalers about to reap the harvest that the semiconductor companies planted? Stanley Druckenmiller is betting hundreds of millions of dollars that the answer is yes. Thank you so much for joining this explainer. Keep your eyes on those cloud margins, and as always, keep learning.
Comments 0