Latest / Deep Values: Value Investing Deep Dives / 【13F】Point72 Steven Cohen's Strategy: The Industrialization of Alpha and AI Infrastructure
Transcript
- 0:00Welcome back to the Deep Dive. Today, we are strapping
- 0:04ourselves into the cockpit of one of the world's most
- 0:06influential and frankly, most secretive hedge funds.
- 0:08Stephen Cohens .72, Asset Management.
- 0:11Exactly. A firm managing A colossal $59.8
- 0:16billion. And get this They delivered an
- 0:20eye watering 17 1/2 percent return in 2025.
- 0:24Incredible performance. So we are performing a forensic
- 0:27examination of their Q 32025 portfolio, which is all laid out
- 0:31in their latest 13F filing. Our mission, as always, is to
- 0:35extract the highest conviction Nuggets of knowledge.
- 0:38And if you look past the sheer volume of trades, because this
- 0:41is, I mean, it's a massive operation, a very clear strategy
- 0:44emerges. The whole portfolio is what we
- 0:46call structurally bifurcated or you could call it the barbell
- 0:49approach A. Barbell, OK, break that down for
- 0:51us. On one end of this barbell, they
- 0:53are aggressively almost, you know, maniacally long the
- 0:56technological singularity, specifically AI infrastructure
- 0:59and the other end, but on the other end, they are massively
- 1:02hedged. They're betting heavily on
- 1:07macroeconomic deterioration, some kind of systemic tail risk.
- 1:11It's the philosophy of the paranoid bull.
- 1:13Exactly. They have total conviction in
- 1:15the long term future, but they clearly hold zero trust in the
- 1:18stability of the present system. And for all of us tracking this
- 1:22market following these institutional moves, it's just
- 1:25the fastest shortcut to understanding the current market
- 1:28temperature. And where the high conviction
- 1:30money is placing its bets. OK, let's unpack this.
- 1:32To really understand the machine today, we kind of need to know
- 1:35where it came from. Right.
- 1:37The roots of .72 lie in its predecessor, SAC Capital.
- 1:41It started way back in 1992 with just $25 million under
- 1:45management. A tiny sum compared to today.
- 1:48And their competitive advantage back then was what we call the
- 1:50informational edge. It was characterized by ultra
- 1:53high turnover, tape reading and just aggressive risk taking.
- 1:56But what was that original competitive edge, really?
- 1:59The market moves so fast now that kind of information
- 2:02advantage seems almost impossible to replicate legally.
- 2:05Well, that's because their advantage was largely
- 2:09qualitative, you know, based on human interpretation.
- 2:12They focused on tape reading, which is essentially just
- 2:15interpreting price action and volume flow as primary signals.
- 2:19Trying to see where the big money was moving.
- 2:21In real time, it's a very high touch, human centric approach
- 2:25which as we know eventually led to some regulatory issues.
- 2:29And that's the pivotal shift Post 2014, the firm transitions
- 2:32to .72 and consciously moves away from what was called the
- 2:36black edge, that sort of aggressive pursuit of unverified
- 2:41or let's say suspect information to the white edge.
- 2:44Which is just superior data processing is a complete costly
- 2:47institutional metamorphosis. A total reinvention.
- 2:50Yes, the firm realized the new alpha wouldn't come from human
- 2:54networks. It would come from superior
- 2:56analytical machinery, and this is where their proprietary
- 2:59market intelligence unit comes in.
- 3:01What does that unit do? It fuses traditional fundamental
- 3:04analysis, the kind of stuff Warren Buffett does, with really
- 3:07advanced data science. And when you say advanced data
- 3:10science, you're talking about generating what you called
- 3:12industrialized alpha. That sounds like they're
- 3:14creating their own version of reality.
- 3:17That's precisely right. They are not waiting for
- 3:19quarterly reports. They're ingesting every piece of
- 3:22alternative data they can possibly find.
- 3:24Like what? Give us some examples.
- 3:26Being credit card transaction logs to gauge consumer spending,
- 3:30satellite imagery of parking lots to track retail traffic,
- 3:33geolocation data from phones, webs, great pricing data.
- 3:38The whole goal is to create these real time proxies for
- 3:42corporate performance to predict earnings beats and misses before
- 3:46they even appear in public reports.
- 3:48So they're manufacturing intelligence, but they also have
- 3:51to manufacture the talent to handle this level of complexity.
- 3:55Absolutely. The .72 Academy, which they
- 3:58started in 2015, is critical to that.
- 4:00Right, the training program. It's a 15 month paid training
- 4:04program. It rigorously trains recent
- 4:06college graduates in their specific homogeneous culture of
- 4:09risk management and analytical rigor.
- 4:11So they're building analysts in their own image.
- 4:14Exactly. By growing their own talent from
- 4:16scratch, they ensure consistency across their 185 independent
- 4:21investment teams or pods. It ensures the machine just
- 4:25keeps running efficiently. OK.
- 4:27So that's the machine discipline, data-driven, highly
- 4:30centralized. Now let's look at what that
- 4:32machine built in Q32025. You called it the barbell.
- 4:37Extreme bullishness on AI, extreme caution on the macro.
- 4:41And that caution is visible right at the top of the filing.
- 4:44They have executed a massive $4.5 billion notional hedge
- 4:49using index options. OK, for those of us not familiar
- 4:52with the terminology, what does notional hedge mean?
- 4:54And why is 4 1/2 billion dollars just such a staggering amount?
- 4:57Right. So a notional hedge refers to
- 4:59the total value of the assets that the options contracts
- 5:02cover. It's not the small premium they
- 5:04paid for the contracts themselves.
- 5:05I see. And that 4.5 billion signals
- 5:08they have bought insurance protecting nearly 8% of their
- 5:10entire visible portfolio against a generalized market downturn,
- 5:14Specifically 3.5 billion in SP Y puts and about 970 million in
- 5:19QQQ puts. 7 1/2% of the entire portfolio just dedicated
- 5:24explicitly to downside protection.
- 5:26That is loud. That level of insurance tells
- 5:30you exactly what Cohen and his team think of systemic risk.
- 5:33It lines up perfectly with his public comments, right?
- 5:36He mentioned a toppy market and a. 45% probability of a
- 5:39recession. Precisely that protection is the
- 5:43precondition for their aggressiveness everywhere.
- 5:46Else it's what lets them sleep at night.
- 5:47It lets them hold these extremely high beta volatile AI
- 5:51stocks, which we'll get to while protecting the whole fund
- 5:55against the systematic collapse. Without that hedge, their long
- 5:59book would look much, much different.
- 6:01And this commitment to safety leads us to their risk
- 6:04framework, which is famously disciplined, the 521 rule.
- 6:07A 5% maximum single position size, 20% maximum sector
- 6:10exposure. But what really stands out is
- 6:13they're almost maniacal focus on liquidity.
- 6:16Right. They avoid things they can't
- 6:18sell quickly so. They can de gross rapidly during
- 6:21market shocks. For those unfamiliar with that
- 6:23term, what is de grossing in simple terms?
- 6:26It just means rapidly selling off assets and reducing leverage
- 6:29to dramatically cut your market exposure.
- 6:32You move into cash when things get turbulent.
- 6:34The ability to shrink the portfolio instantly.
- 6:36It preserves capital when everyone else is trapped.
- 6:39And that focus on discipline is so extreme, they actually made
- 6:42the counterintuitive decision to return 3 to $5 billion in
- 6:46capital to investors in 2025. I mean, returning capital is
- 6:50almost unheard of for a hungry hedge fund trying to maximize
- 6:54fees, So why? Do it.
- 6:55It's based on a key insight. They believe excessive size
- 6:59forces funds into illiquid positions or it makes them
- 7:03become the market, which erodes their alpha.
- 7:06They think they're getting too big to trade effectively.
- 7:08Exactly. It's a defense mechanism.
- 7:10So, discipline. Oversize.
- 7:11A rare and powerful statement. OK, let's pivot from risk
- 7:16management to the core growth engine.
- 7:18AIQ 3 saw a decisive capital flight out of some big names and
- 7:21into others. What was the central thesis
- 7:24here? The thesis is crystal clear.
- 7:26They favored the builders of AI infrastructure over the users.
- 7:29Builders over users. They moved capital out of the
- 7:32consumer facing hyperscalers companies that use cloud AI, and
- 7:36into the pureplay silicon and hardware manufacturers, the ones
- 7:40who make the tools. Let's look at the sell side of
- 7:42that. They sold almost 35% of their
- 7:45Amazon position and cut their snowflake position by more than
- 7:48half, 54%. Why?
- 7:51The rationale likely hinges on relative value in competition.
- 7:55Amazon's fortunes are tied to consumer spending, which might
- 7:58be softening. And Snowflake.
- 8:00Snowflake, while a great company, is sensitive to
- 8:03enterprise budget cuts and it's facing rising competition from
- 8:07open source alternatives, .72 is hunting for the purest, highest
- 8:11leverage way to play this theme. And yet they also trimmed Arista
- 8:15networks by about 20%, which is a pristine hardware play.
- 8:19Was that a negative signal? No, I don't think so.
- 8:21I view that trim as evidence of their ironclad discipline, the
- 8:24521 rule in action. It just got too big.
- 8:27Arista had likely run up so much that its position size exceeded
- 8:30their internal risk limits. It was discipline, profit
- 8:32taking, not a signal of any deterioration in the business.
- 8:36OK. Now for the conviction moves,
- 8:38where did the capital flow tell us about the buy side?
- 8:41The highest conviction moves were squarely in core hardware.
- 8:45They increased their NVIDIA state by almost 13%, up to 1.35
- 8:51billion, just doubling down doubling.
- 8:53Down on the Blackwell and Ruben ship cycle.
- 8:55But the most staggering move by far was Teradyne TER.
- 9:00Staggering is the only word. They increase their teradyne
- 9:03position by 1370%. 1370 that's. Not a trade.
- 9:08That's a structural conviction shift.
- 9:10What is the deep thesis behind such an obscure testing
- 9:14equipment maker? This is truly playing chess
- 9:16while others are playing checkers.
- 9:18Teradyne makes automated test equipment for chips.
- 9:20Right, .72 is making a brilliant second derivative bet as AI
- 9:25chips get exponentially more complex.
- 9:27We're talking advanced packaging, 3D stacking.
- 9:29The intensity of the testing required per chip goes up
- 9:32massively. I see they're.
- 9:34Betting that the complexity of manufacturing scales faster than
- 9:38the volume of chips they are along the difficulty of making
- 9:41AI hardware. That is a fascinating angle
- 9:44focusing on a bottleneck the market often ignores.
- 9:47They also boosted Broadcom and Meta.
- 9:49Broadcom up 187%. That's a great play on custom
- 9:53silicon or ASICS, which comes like Google and Meta design
- 9:57themselves. And Meadow is U almost 300.
- 9:59Meta is fascinating because, you know, it looks like a consumer
- 10:02company. But Cohen treats it as an AI
- 10:04leader because of its llama models and its massive internal
- 10:07infrastructure build out. It's a way to get AI exposure at
- 10:11a arguably cheaper valuation. And their conviction only got
- 10:15stronger after the emergence of that Chinese DeepSeek AI model,
- 10:19right? Which seems counterintuitive.
- 10:21A powerful competitor should create fear.
- 10:23This is one of the most important takeaways from the
- 10:25entire filing. Cohen publicly stated that
- 10:28DeepSeek was bullish for this sector.
- 10:30Bullish. Why?
- 10:31His theory is that models like DeepSeek, which lower the cost
- 10:35of AI inference, the cost of actually running the model, will
- 10:38drive a huge proliferation of new AI applications.
- 10:41More applications mean more users, which means a far greater
- 10:45unexpected total demand for the foundational hardware.
- 10:47So competitor efficiency actually validates their bet.
- 10:50It expands the pie. Exponentially.
- 10:52That brings us to our final section, .72 might be a high
- 10:55turnover fund, but the concentration in these top
- 10:58names, it demands A deeper look. Let's put on our long term
- 11:01lenses and do a Buffett style analysis on these top holdings
- 11:05starting with their number one long position, NVIDIA.
- 11:08NVIDIA passes the mote test with flying colors.
- 11:11It's Mote isn't the hardware, it's the software.
- 11:14It's the CUD walled garden. The software ecosystem.
- 11:18For 15 years, developers have built their entire libraries and
- 11:21expertise around CED. That software lock in creates
- 11:26incredibly high switching costs. It's a network effect that is
- 11:29virtually unassailable right now.
- 11:31And the financials are just insane.
- 11:33In 2025, NVIDIA generated 60.8 billion in free cash flow, a
- 11:38125% increase year over year. And the gross margins are
- 11:42consistently around 72 to 75%. I mean 72% margins, that is
- 11:47monopoly like pricing power. But the number that truly
- 11:50defines its quality is the return on invested capital or
- 11:54ROIC. It's a staggering 97.8%. 97.8%
- 11:59That's almost double what other big tech players achieve.
- 12:02What does that kind of roic even tell us?
- 12:05It means NVIDIA is a classic compounder.
- 12:08It requires very little tangible capital relative to its profits
- 12:12to grow. It's just a cash generating
- 12:14machine, the hallmark of a high mode business.
- 12:16OK, Next up, Microsoft, the perennial safe haven.
- 12:20Microsoft's mode is Cure Enterprise Utility.
- 12:23It's the seamless, inescapable integration of Azure, Office 365
- 12:28and Windows. Leaving one product diminishes
- 12:30the value of the others. And they have an unmatched
- 12:32distribution advantage with Copilot, right?
- 12:34Pushing AI features to hundreds of millions of enterprise seats
- 12:37overnight? Unmatched.
- 12:39But there's a massive caveat here, something that would make
- 12:41a traditional value investor a little nervous Capital
- 12:44expenditure. Absolutely.
- 12:45While free cash flow is robust at 71.6 billion, that FCF is
- 12:49being depressed by massive planned spending.
- 12:51How much? Microsoft plans to spend $80
- 12:54billion in 2025 alone on AI data centers. 8 billion.
- 13:00That spending is necessary to defend the Moat in the cloud
- 13:03wars, but it raises a huge question about the long term
- 13:07capital intensity required just to stay ahead in this race.
- 13:10Finally, let's revisit Arista Networks, the position they
- 13:13trimmed despite its quality. Arista's competitive advantage
- 13:16is its software, the EO's Extensible Operating System,
- 13:20which handles high speed data center switching.
- 13:22It's why hyperscalers demand it and the.
- 13:24Balance sheet. It's a fortress. 10.1 billion in
- 13:27cash and 0 debt. Net margins are near 40%.
- 13:31So again, why would Cohen let a quantitative rule override such
- 13:34a pristine debt free business? That decision just reinforces
- 13:38the dominance of the firm's centralized risk framework.
- 13:41Arista was creating it roughly 40 times earnings too.
- 13:44Expensive for their model. No matter how great the
- 13:46business, a trade that violates their internal valuation
- 13:49discipline has to be cut. They sold based on price and
- 13:52discipline, not on the deterioration of the business.
- 13:55That's the difference between a hedge fund and a value investor.
- 13:57Right there it is. OK, that brings us full circle.
- 14:00The strategic synthesis here is just brilliant.
- 14:04They are long the singularity fully invested in the AI
- 14:08revolution and at the same time they are long disaster with
- 14:11those massive index puts. The final key take away is that
- 14:14relentless focus on infrastructure builders over
- 14:16users and the institutional discipline, the 521 rule.
- 14:21The firm remains the perfect paranoid bull, fully committed
- 14:24to the future but heavily insured against the present.
- 14:27And that paranoia is where you should focus your attention
- 14:30moving forward. .72 strategy relies heavily on those massive
- 14:34index put positions. The single most telling signal
- 14:37for their confidence will not be their next stock picks.
- 14:39No, it'll be their put call ratio.
- 14:41Exactly. A sustained reduction in that
- 14:43SPY put position would signal Cohen is shifting from cautious
- 14:47bullishness to all clear aggression.
- 14:49So what stands out to you about the level of fear still embedded
- 14:52in this $60 billion portfolio? And understanding that embedded
- 14:56market temperature is really essential for informed
- 14:59investing. And if you weren't looking for a
- 15:01quick and easy way to track total market temperature in real
- 15:04time, remember you can search and download the iOS app Buffett
- 15:08Indicator from the Apple App Store.
- 15:10As always, remember that all analysis we provide is based on
- 15:13publicly available sources and is subject to revision.
- 15:16The content is for educational purposes only and does not
- 15:19constitute investment advice. Always conduct your undue
- 15:21diligence. If you want to listen to our
- 15:24full deep dive on this topic or access more advanced content,
- 15:27you're welcome to join the Deep Value Club membership or
- 15:29subscribe to us on Apple and Spotify podcast channel Deep
- 15:33Value Investing. We'll get you on the next deep
- 15:35dive.