$META still has a lot of room to run. The most interesting part of Meta Connect, beyond the camera-less glasses and VR, was how Meta plans to monetize Muse. Zuck said: “We believe that Muse will make you money and we are standing behind this by making Muse free for a huge number of tokens with the expectation that over time we will profit by taking a small fee from transactions.” We’ll have our agents buy things for us. We’ll have our agents sell things for us. And $META will take a small cut of those transactions. Just to understand the potential scale, assume: • 500M Muse users • $2,000 transacted per user annually • 1% average transaction fee That’s $1T in annual transaction volume. At a 1% cut, that’s $10B in annual revenue for $META. And this doesn’t even include potential $20 or $100/month Muse subscriptions. Truist estimates Muse could generate $28.5B in incremental revenue by 2030. And this goes far beyond $META. This could be a preview of how the top AI labs eventually monetize their agents: Make the agent cheap or free, let it create economic value for users, and take a tiny cut of the transactions it enables. At scale, a tiny cut becomes a massive business.

$IREN has been under pressure over the last two days after SemiAnalysis ranked it “Not Recommended” for GPU cloud, describing its Canadian sites as among the worst in the industry. But here’s the funny part. Akamai was ranked in the EXACT same category as $IREN. That same day, Akamai announced a deal potentially worth ~$20B to provide cloud infrastructure to Anthropic. And Anthropic isn’t just a customer. Akamai issued Anthropic warrants to acquire up to ~5% of the company. Yes, the deal is primarily focused on CPU workloads, while SemiAnalysis is evaluating GPU cloud infrastructure. But Anthropic is committing potentially ~$20B AND taking equity exposure to a company SemiAnalysis literally ranks “Not Recommended.” These rankings are being given way too much weight. A benchmark can tell you how a specific GPU infrastructure setup performed at a specific point in time. It cannot tell you who will win in AI infrastructure. The market is confusing the two.

$NFLX is one of the most overlooked opportunities in the market right now. Down ~50% from its highs, Netflix trades at just ~20x forward earnings. Meanwhile, management has laid out some massive 2030 ambitions: • $1T market cap • $80B revenue • $30B operating income • $9B advertising revenue With a current market cap of ~$295B, reaching $1T by 2030 would imply a ~35% CAGR from here. And Netflix doesn’t need to reinvent itself to get there. Scale the core. Turn advertising into a monster. Keep taking live sports away from traditional TV. At 20x forward earnings, the market is pricing Netflix like its best days are behind it. I think they’re still ahead.

My 5 favorite buys in the market right now, in order: 1. $GRAB 2. $LMND 3. $SOFI 4. $ORCL 5. $NVO

$CRWV has a $130B backlog. $NBIS has “only” ~$40B. Most investors see that and assume $CRWV has the advantage. They’re looking at it backwards. For $CRWV, that massive backlog isn’t just an advantage. It’s a necessity. These contracts are overwhelmingly with hyperscalers and big AI labs. As $NBIS management has explained, these customers don’t need the full software stack. They mostly need bare-metal compute. That means lower margins and less revenue per MW. So why lock up so much capacity with them? Because massive contracts with highly creditworthy hyperscalers allow neoclouds to raise debt at much better terms. And nobody needs that more than $CRWV. $NBIS has other levers. It can potentially monetize non-core assets such as Toloka, Avride and its ClickHouse stake. $CRWV is far more dependent on external financing to fund its buildout. That completely changes how you should look at its $130B backlog. The hyperscaler contracts aren’t valuable just because of the revenue they lock in. They’re valuable because of the billions in cheaper capital they can unlock. For $CRWV, backlog is a financing weapon.

$ORCL is one of the easiest buys in the entire market right now. While $NBIS is up ~10% today and $CRWV ~4%, $ORCL is down another 4%. Oracle is now ~60% below its all-time high and trades at just ~21x earnings. It has also fallen below its 200-week moving average. Over the last 17 years, $ORCL has traded below its 200 WMA in only 4 distinct periods. Every single time, it reclaimed it within weeks. And now Oracle has $664B in RPO. On top of that, the $20B ATM was already completed last quarter, removing that dilution overhang in the short term. A 60% drawdown. 21x earnings. $664B in contracted backlog. And one of the rarest technical setups $ORCL has seen in the last two decades. I’m buying $ORCL here.

$AMD has gone 8x over the last 17 months. $NVDA? “Only” 2.4x. Yet over that same period, Nvidia’s revenue grew 118%, while AMD’s grew 55%. So why did $AMD crush $NVDA? Because the move started long before April 2025. From the beginning of 2023 to April 2025: $NVDA: +540% $AMD: +25% Nvidia had already captured almost the entire AI trade. AMD had been left behind despite being exposed to the same massive AI boom. Then the gap closed. When a sector explodes, stop chasing the leader. Buy the laggard whose fundamentals are catching up. I see the same setup today: • $SOFI vs $HOOD • $NVO vs $LLY • $IREN vs $NBIS The biggest opportunity is often not the stock everyone is chasing. It’s the one everyone forgot while the business kept getting better.

2.5 years ago, I put $3,000 into an investing experiment. On Monday, it crossed $35,000. I did it without “buying and holding forever.” I concentrated the money into just 1–2 high-conviction stocks at a time: • Buy when I believed the stock was significantly undervalued • Let it run • Sell when price ran too far ahead of fundamentals • Rotate into a better risk/reward opportunity $3,000 → $35,000. More than 11x in 2.5 years. The biggest lesson? Price matters. A great company can be a great investment at $100 and a terrible investment at $300. I’ve become increasingly convinced that “buy and hold forever” is rarely the optimal strategy. Capital should go where the best risk/reward is. It’s also one of the reasons I completely exited $AMD in my main portfolio today. I still believe in the company. I just believe my money has better places to go.

My 3-year price targets: • $GRAB: $15 • $META: $1,500 • $SOFI: $70 • $IREN: $300 • $LMND: $300

I fully exited my $AMD position at $617/share. I’m still bullish on the company. I’m confident AMD will become much bigger than it is today. But the stock is simply no longer as compelling as it was in 2025, when it traded at ~20x forward earnings. $AMD has gone 8x in just 17 months. And parabolic moves like the one we saw earlier this week usually mark the end of a move, not the beginning. The thesis hasn’t changed. The risk/reward has. There are far better opportunities out there that aren’t nearly as extended as the market is right now.

Nothing is more bullish than a CEO putting serious money behind his own company. On April 6, $OSCR CEO Mark Bertolini bought $12M worth of shares at an average price of $11.92. Today, $OSCR trades around $31. Nearly a 3x in just over 5 months. Now it’s $GRAB’s turn. On September 21, CEO Anthony Tan bought more than 10M shares on the open market at an average price of $2.89. That’s ~$30M of his own money. And he’s buying right as $GRAB is entering a completely different phase as a business. 2025 was its first full year of net income profitability. Now profits are accelerating, margins are expanding, and the business is scaling across mobility, deliveries, and financial services. The market is still pricing $GRAB like the old $GRAB. Tan is buying the new one. $OSCR was the setup five months ago. $GRAB is the setup today. And Tan wasn’t alone. On the same day, $GRAB’s COO also bought ~$866K worth of shares. Insiders are buying. So am I.

$HIMS continues to be one of the best buys in the entire market. Extremely asymmetric. The market sees the shift from compounded GLP-1s to branded treatments like $NVO’s as bearish because it will pressure margins in the short term. I see the exact opposite. It’s a massive validation of the $HIMS thesis. Hims is sacrificing short-term margins to bring blockbuster weight-loss treatments onto its platform and attract a massive number of new customers. Subscribers increased by 307K QoQ in Q2, although part of that growth came from the Eucalyptus acquisition. But weight loss is just the entry point. Customers will come to Hims for GLP-1s. Then Hims can cross-sell them higher-margin treatments across the rest of its platform. And every new subscriber generates more health data. More customers → more data → better personalization → better outcomes → more customers. That flywheel will keep compounding. $HIMS isn’t building a weight-loss company. Weight loss is the customer acquisition engine for something much bigger.

People still don’t understand the scale $NBIS is targeting. Arkady said Nebius will build 1 GW of compute capacity per year starting in 2027, and likely ramp that to 1.5–2 GW per year over time. The entire AI industry will build roughly 10 GW in 2026 and 15–20 GW in 2027. That means $NBIS alone could represent roughly 5–10% of the entire industry’s annual compute buildout. And this is happening in a market that will become a trillion-dollar revenue business. $NBIS is building at hyperscaler scale.

$AMD reached a $1T market cap today. Since Lisa Su became CEO in 2014, $AMD has gone up 500x. And she’s not done yet. Winners tend to repeat themselves.

The market is sleeping on these names: • $AVGO at $356 • $NFLX at $71 • $SOFI at $16 • $LMND at $47 • $GRAB at $2.79 • $UBER at $70 • $CRWV at $81 • $META at $665 • $HIMS at $27 • $NVO at $43 • $ORCL at $147 • $ROOT at $49 • $IREN at $46 • $NKE at $35 There are always opportunities in the market. Always.

If you’re buying one MAG 7 stock, it has to be $META. It trades at just 20x forward earnings while growing revenue faster than every MAG 7 company except $NVDA. Q2 2026 revenue: +28% YoY. But the market is still underestimating $META’s real moat: distribution. 3.6B people use at least one Meta app every single day. Facebook, Instagram, and WhatsApp are the three most used social platforms in the world. That means every new product $META builds can be put in the hands of billions of people almost overnight. No other company can distribute a new product to billions of people this fast. And with the progress Meta is making with Muse, that advantage is about to matter even more. $META doesn’t need to find billions of users for its AI. It already has them.

The explosion of open-weight AI models will be an inflection point for $NBIS and $CRWV. Companies will fine-tune open models on proprietary data, creating smaller, specialized models for specific use cases. These models will be dramatically cheaper to run, lowering token costs and expanding margins. And this is where the real opportunity begins. Cheaper AI won’t reduce compute demand. It will make it explode. Lower costs will unlock thousands of new AI use cases, driving exponentially more inference and compute consumption. OpenAI has already shown this in practice. Luna costs 80% less to use, and demand increased more than 10x. Cheaper intelligence creates more demand for intelligence. That’s exactly what $NBIS and $CRWV are building for.

Some predictions: • $NBIS becomes a hyperscaler • $LMND becomes the largest insurance company in the world • $AMZN becomes a $10T company by 2030 • $HIMS builds the most valuable subscription in the world • $IREN monetizes its entire 5.8 GW power portfolio and becomes a $100B+ company

The market is sleeping on $AVGO. The stock is down ~30% from its highs. It makes no sense. Here’s why $AVGO is now my favorite semiconductor stock: • Hock Tan expects Broadcom to exceed $30 in EPS in FY2028. Put a 30x multiple on $30+ of EPS and you get $900+ per share. That’s roughly +150% upside in just over two years. • AI semiconductor revenue is expected to reach $115B in FY2027 and $230B in FY2028. That means doubling two years in a row. • Anthropic is set to become Broadcom’s largest custom-chip customer in 2027, overtaking $GOOGL. • Yet $AVGO is trading at roughly 20x forward earnings. $AVGO is down 30% while its AI business is preparing to double two years in a row. This makes no sense.