UNREAL

@NighthawkTradez @KingJames Stock would move up more from LeBron gojng to Warriors than any amount of GW that Dan could announce 😂

Added more $IREN today. Please @KingJames announce the deal.

@SemiAnalysis_ This is definitely packaging issues: TSMC over promised on CoWoS-L package size and couldn’t cash the check. This is why they are aggressively pursuing CoPoS

INTERESTING: Only 3 months after Rubin Ultra was announced at GTC 2026, the original 4-die Rubin Ultra has been cancelled due to manufacturing execution concerns. The new “Rubin Ultra” is half the size/~ half the real-world performance of the original Rubin Ultra. 1/4🧵

@wliang Appreciate that you’re long most of the names instead of fighting about which one is best. I’m long about 10 names with different approaches at different stages but all will be winners. I simply weight the position for my conviction so some positions are 10-15x others.

Btw, neocloud isn't just a 2025-2026 trade. We're still early… All the names we’ve covered will benefit… $NBIS, $CRWV, $CIFR, $IREN, $WYFI, and more. (I’ll share additional, new neocloud names this week as well 🫡) Quick recap: > AI inference cost has dropped ~90% annually for the past two years. Jevons Paradox is playing out in real time. (Cheaper compute = new use cases = increases total consumption) > The shift to agentic AI is multiplying compute demand by 100-1000x per task. > Token consumption volumes are growing exponentially. > Enterprise AI adoption is still under ~30%. Ppl have been saying "AI is here" for two years, but most enterprises are still in pilot phase. > Covered this many times, but supply literally can’t catch up. $TSM is booked through 2027, HBM sold out, power infrastructure takes 4-7 years to build, and many other constraints. IMO we are far from the peak of an AI bubble.

@jukan05 The MOU was signed months ago reporting on all of this. Anything new? m.samsungsem.com/global/newsroo…

Korean media reported that Samsung Electro-Mechanics is expected to sign a formal agreement this week with Sumitomo Chemical to establish a glass substrate JV. The two companies will invest KRW 500 billion, or about USD 320 million, to set up the entity. Samsung Electro-Mechanics is expected to own more than half of the JV and contribute around KRW 300 billion. The plan is reportedly to establish the entity at the Pyeongtaek site of Dongwoo Fine-Chem, Sumitomo Chemical’s Korean subsidiary, and begin operating production equipment in early 2028.

Very few people know about $DMGI in the AI infrastructure space, so here’s a break down of the last week: $HUT $IREN $WULF $SLNH $RIOT 📖 - 50MW LOI signed for their Christina Lake Facility. Large company, NDA, company will backstop funding. Have applied for an additional 150MW from their utility for that site. That site have mirrored substations setup, 75MW each. The company owns the infrastructure, 33 acres, and the building. They already have 12MW of cooling on site. They also have Nat Gas transmission lines running on the property, 50MW for potential back up. First GPUs active this year. - The have a utility MOU and a 30MW MOU with the @malahatnation. CEO said updates on that partnership can be expected in the coming quarter. - Canada just announced their AI strategy. DMG has met with defense procurement already and CEO said the Feds called the company this Monday. - They own SCIF rated modular data centers rated for defense sector grade AI. 2MW already on site, looking at buying 8MW more. CEO said after they signed their LOI for the 50MW, they have received increased interest for the SCIF units. - 2nd Canadian site and Oregon site, CEO said updates on those sites could be expected in the coming quarter. - Also, wholly owned digital asset trust, updates expected. - The company has expressed numerous times that they will not dilute to pay for buildout, as the deal is backstopped. CEO said financing entities were already contacting the company after the LOI. - The company trades at UNDER ASSET VALUE. Others have run the numbers for fair market value of their MW deal, comes to $5-$7 for just the 50MW. - 400 BTC HODL, still mining. - 206M shares outstanding.

@jiahanjimliu Great details Jim. Question, what do you think of $NUAI?

6/27 Portfolio Update 93.85% $IREN I acknowledge the disappointment for the cost of the Golden State Sponsorship and the real disappointment of GPU deliveries and ramp. The disappointment can be attributed to three sources: 1. Prince George revenue miss for Q1 earnings. @_Sgr_A_Star went through the account receivables and most of the GPUs are billing for Q2 (1). PG is a retrofit but UPS weren't previously needed for BTC mining and British Columbia has slow permitting process so a majority of the PG GPUs became billing at the very end of Q1. I expect 90-110m AI revenue from PG for Q2 earnings around August ~6. 2. Horizon 1-4 deliveries for Microsoft. From satellite pictures, @FransBakker9812 subgroup expects late July / early August delivery. With the H1 issues flushed out and delivered, H2-4 should be much faster and on track for end of this year, worse case Jan 2027. 3. Comparisons to $NBIS. $NBIS stock price has rocketed as they delivered 399m revenue from self-owned Finland and other colocation sites. However, Vineland just swapped out power sourced from unauthorized Bergen Engines to $BE Fuel Cells (2) and it's clear they will be using other locations to meet obligations for their $MSFT contract. When IREN announces H1 delivery in late July/August, $IREN will be ahead in delivering SOTA training GB300 clusters for Microsoft as $NBIS Vineland sorts out it's power workaround. I acknowledge that in order to meet Nvidia roadmap, $IREN has made a recent pivot to software on approximately half their capacity. Although IREN is behind in software, software is zero margin cost for distribution and not the bottleneck for AI. For the most part, it's been physical constraints like memory fab and power like $BE that has been handsomely rewarded, not SaaS. $IREN has been fine selling to leading AI platforms like FireworksAI, TogetherAI, Hume AI and Baseten via Fluidstack but needs to have higher priority GPU deliveries and system bring up. Once H1 de-risk's $IREN's bringup speed, we will see revenue really start to ramp. For software, they are bringing Mirantis for Orchestration for Bare Metal+. $IREN remains my #1 position for the following reason. In 2 years, we saw $MU's net margin go from 4.9% to 68.1% and gross margins hit 84.6% as HBM became the bottleneck. Goldman Sach reports for datacenter capacity: "Year-over-year capacity additions are scheduled to reach 13.6 GW in 2026 and 36.3 GW in 2027" (3). 36.3GW is from 66GW of planned buildouts (3) as the remaining datacenter buildouts will be delayed or cancelled. At 36.3 GW DC capacity, HBM is the bottleneck, but if we have more misses and only 30GW DC capacity is added in 2027, the bottleneck shifts from HBM to power. Only AWS and GCP in-house buildouts are in the clear to have most their planned capacity. Overall, 13.6 GW added in 2026 to 36.3GW and put massive strain on the BTM supply chain. Hypertec 5C, a very large private Neocloud, is already reaching pass $BE Fuel Cells for $FCEL to meet timelines (5). $CRWV has had missed in the pass and is likely to miss more in an tighter environment. $NBIS has announced GWs left and right but ironically, is still sorting out power for their Vineland site which had power secured in March 2025! BTM is hard to do as evidence by delays from $ORCL, Crusoe getting cancelled on by Google for mising checkpoints (4), Hypertec 5C reaching out to the frontier with $FCEL which is not unverified commercialized at scale (5). Many GWs of "secured power" look just as secured as Vineland but not all can be saved last minute by $BE fuel cells. 2.40% Cash 2.03% $OUST $OUST is trading at a premium but you also get a premium team in robotics. Writeup: x.com/jiahanjimliu/s… 1.72% $TSLA I've usually don't all in my portfoilo and my only other all in was previous $TSLA. These are my leftovers. Writeup: x.com/jiahanjimliu/s…

Enterprise can’t pay capex for AI cloud, not yet / not at scale Only Hyperscalers (HS) can afford it So HS gobble up power and build up their in house chips (TPU, Trainium, Maia, etc). Where does this leave $NVDA? Losing ground to custom ASICs Not because they are better than NVDA GPUs Just because they are cheaper. No one wants to pay the rich $NVDA margin. Remember. “Your margin is my opportunity” So what does $NVDA do? Partners with Neoclouds. $CRWV $NBIS $IREN support the NVDA hegemony against the HS homegrown chips The backbone of this strategy? $IREN 5.8 GW and “multiples” of that in the pipeline. Millions of GPUs worth hundreds of billions. Locking enterprise customers into DSX ecosystem and CUDA native SW. Keeping HS incursion and the precious revenues and data analytics at bay. While $IREN secures and develops power and leaves HS scrambling for colocation deals and expensive BTM schemes. Meanwhile $IREN launches cloud services via Mirantis to secure astronomically better cost economics compared to multi-year, bulk HS deals So investors are complaining about $IREN not selling out their whole pipe in mediocre bulk HS deals… “I want deal now” But $NVDA and $IREN have strong synergistic plan that simultaneously enforces NVDA ecosystem, curtails HS, gives $IREN top margins, and provides easy outlet for the inevitable rise of well-funded sovereign customers. $IREN and $NVDA have better vision than all the investors who have been selling.

@MilkRoadAI Zuck’s future Vision has lost sync with reality. The company whiffed on Metaverse and is losing on AI. They are losing relevance. Not the first take from @MilkRoadAI that’s out of touch.

I will die on this hill but Meta is the most undervalued mega cap stock in the market right now (Save this). The market is treating Meta's AI Capex as a liability but it's actually a weapon. Meta trades at just 16.4x forward earnings cheaper than Walmart, Costco, and dramatically cheaper than Alphabet at 30x despite Meta growing faster. Its own 3 year average PE is 20x and if the multiple simply reverts to that historical average with no earnings growth at all, the stock is worth 30% more than today. As Bill Ackman said it best, If a company is increasing growth capex while earning above average returns, you should applaud." That is exactly what is happening here. Meta's ad business is projected to hit $240 billion in revenue in 2026 alone, up 22% year over year. This is not a company fighting for market share because Meta has 3.5 billion people using its apps every single day. AI is not threatening that moat but rather turbocharging it. Meta's Q4 2025 AI model rollout drove a 24% increase in incremental conversions for advertisers, brands using Advantage+ AI campaigns are seeing 41% higher return on ad spend and 17% lower customer acquisition costs versus manual campaigns. When your ad product gets measurably better for every buyer, every buyer spends more and the market is also completely ignoring what is coming on top of the core ad business. Wolfe Research projects Meta will generate over $26 billion in entirely new incremental revenue by 2027 from AI products, WhatsApp monetization, business messaging, and agentic commerce, none of which exists in current earnings estimates. WhatsApp alone has over 2 billion users and is barely monetized in most of the world. In India, Indonesia, Brazil, and Mexico, some of the fastest growing middle class markets on earth, WhatsApp is the primary communication platform and Meta is just now turning on business AI tools there. Here is the math that should make your jaw drop. At today's $1.4 trillion market cap, Meta trades at 15.2x 2027 earnings and 13.2x 2028 earnings. A company with 3.5 billion daily users, 22% revenue growth, 40%+ operating margins, and AI just beginning to compound, trading at 13x earnings two years out. Apple, which grows at 6%, trades at 28x. The market is punishing Meta for investing aggressively in AI capex but Meta generates so much free cash flow that it is funding $125–145 billion in annual AI capex almost entirely from its own ad business. Bullish on Meta here!

@franklee6924T Hell of a post mate. I do believe you are right, and this is the crafted Vision that IREN executes to and many can see. And yet many more are focused on short term returns.

The Most Important Thing in the Neo-Cloud Industry: Building a Self-Reinforcing Capital Flywheel What is the best way to evaluate the long-term potential of the Neo- cloud industry? I believe there is only one answer: identify which company can quickly build a self-reinforcing flywheel between capital investment and capital returns. In this model, a company continuously invests in high-quality AI computing infrastructure while generating growing free cash flow. At the same time, its reliance on external financing gradually declines. It establishes a clear path of capital investment → cash generation → larger investment → greater cash generation, ultimately reaching a point where its own cash flow is sufficient to sustain the entire cycle. Only company that achieve this have a real chance of winning in the future. Stock prices can be highly misleading. What the market focuses on in the short term is not necessarily what will determine long-term success. Instead, we should evaluate how each participant performs in building this flywheel. To build such a flywheel, six conditions must be met. First, low-cost energy assets. Whoever can secure abundant, stable, low-cost, and green energy over the long term will enjoy a generational advantage in cost structure. Energy is the first principle of this industry. It is the foundation upon which the flywheel begins. Second, stable cash flow from high-quality tenants. This is the cash-generating engine of the flywheel. Third, package stable cash flow together with computing assets into Asset-Backed Securities (ABS/ABF), allowing future cash flows to be monetized in advance through the capital markets. This is the accelerator and financial leverage of the flywheel. This is also the step that turns something ordinary into something extraordinary. By packaging long-term customer contracts together with AI computing assets into financial products similar to REITs or Asset-Backed Securities (ABS/ABF), companies can monetize future cash flows today through the public capital markets, enabling capital recycling and recovery. Fourth, expand capacity to improve energy utilization. This is where scale economies come into play. Larger scale further reduces marginal costs, improves Power Usage Effectiveness (PUE), and attracts more high-quality customers who value stability and cost efficiency. The larger the platform becomes, the better it can optimize scheduling, spread costs, and improve utilization, creating a positive feedback loop that allows the flywheel to spin faster and faster. Fifth, computing orchestration and full-stack service capabilities. These maximize the efficiency of the flywheel. Only when a company possesses integrated hardware-software scheduling systems, large-model fine-tuning capabilities, and highly efficient operations with minimal losses can energy truly be transformed into marketable AI computing products. Sixth, alignment with policy incentives and industry standards. This includes carbon footprint compliance, green finance, carbon credit mechanisms, and similar frameworks. This ensures the flywheel can continue operating smoothly. Green AI infrastructure gains access to lower-cost financing and stronger policy support, further improving capital efficiency and reducing expansion costs. These six conditions are strict benchmarks. Any company can be measured against them, and we can quickly see who is actually on the right path. Take Oracle as an example. Oracle has an outstanding legacy business, a stable customer base, and excellent cash flow. After recognizing the enormous opportunity in AI, it dramatically accelerated investment. But has it successfully built this new industrial flywheel? The answer is no. Rather than comparing every point one by one, I'll simply mention the result. All three major credit rating agencies currently place Oracle's credit rating close to the edge of non-investment grade. One more downgrade would move it into junk bond territory. Google, Microsoft, Meta, and Amazon all possess exceptional cash flow businesses and strong balance sheets. They are diversified giants. However, their size also creates limitations. They cannot devote all of their attention to developing the new cloud industry. Although they continue investing heavily, they failed to secure first-mover advantages in premium energy resources, especially grid-connected gigawatt-scale sites. The power projects they have announced are impressive, but most will not come online for another three to four years. Many rely on geothermal, nuclear, or other emerging energy sources whose technological maturity, cost control, and commercial viability are still being tested. CoreWeave and Nebius both attempted to build neo- cloud businesses using an asset-light model, hoping to avoid heavy infrastructure investment. Reality has largely disproven that approach. Both companies now recognize the problem and are attempting to pivot. Unfortunately, the optimal window for making that transition has already closed. Economically attractive sites have largely been taken. If they move into behind-the-meter power generation, they must still confront questions surrounding reliability, technological maturity, and cost control. Measured against these six requirements, CoreWeave and Nebius remain far behind. CoreWeave also carries a credit rating approaching dangerous territory. The AI industry has barely begun, yet its financial position has already become highly leveraged. That reflects very poor execution by management. Nebius has delivered the strongest stock performance this year. But what foundations has it actually built for long-term development? Its front-end execution has indeed been excellent. It has signed numerous customer contracts, filled most of its capacity, acquired several important software companies, and established the vision of building an AI Token Factory. Its cash balance appears healthy. Its AI business growth appears impressive. However, it has largely treated the unavoidable issue of heavy infrastructure investment as a black box. Its leasing strategy temporarily obscures the problem, but ultimately this challenge cannot be avoided. Especially now that Nebius is increasingly emphasizing vertical integration, it will eventually have to undertake substantial capital expenditures. When that happens, today's seemingly abundant cash balance could disappear very quickly. Because of space and time constraints, I cannot compare every AI cloud company individually. I have selected only several representative examples. The barriers to entry in this industry have become extremely high, especially now that development has entered the era of gigawatt-scale AI factories. Realistically, the number of companies capable of competing at that level can probably be counted on one hand. Among companies that satisfy all six conditions—and continue strengthening them—$IREN stands out the most. Rather than comparing every point again, I'll focus only on the third requirement: Packaging stable cash flows together with computing assets into Asset-Backed Securities (ABS/ABF), allowing future cash flows to be monetized in advance through the capital markets. This is the accelerator and financial leverage of the flywheel. On June 1, 2026, IREN announced the successful completion of its ABF financing. This $3.65 billion investment-grade GPU financing became a landmark event for the entire industry. Not only did it provide the core funding supporting IREN's AI cloud agreement with Microsoft, but it also received an A rating from Fitch and an A(low) rating from DBRS, making it the highest-rated publicly disclosed GPU financing transaction to date. It also became the first structure of its kind to enter the U.S. private placement market, marking the first time AI computing assets were accepted by capital markets as genuine investment-grade infrastructure. The financing structure itself was even more sophisticated than many expected. IREN combined $2.1 billion of U.S. private placement fixed-rate notes (SOFR + 2.13%) with $1.55 billion of delayed-draw term loans (SOFR + 2.25%), resulting in an overall blended borrowing cost of approximately 6.00%. This financing covered roughly 96% of the $5.81 billion GPU capital expenditure required under the Microsoft contract. Goldman Sachs and JPMorgan served as joint lead arrangers, further strengthening the transaction's credibility among institutional investors. Even more importantly, once Microsoft's $1.94 billion customer prepayment under its five-year, $9.7 billion contract is included in the overall capital structure, IREN's effective financing cost falls further to approximately 3.31%. In other words, IREN can fund nearly all of its GPU capital expenditures without issuing significant new equity, achieving an exceptional balance between infrastructure expansion and shareholder dilution. Looking at the outcome, this financing represents much more than simply raising capital. It serves as a model for optimizing capital structure. It enables IREN to execute its Microsoft contract without sacrificing shareholder interests, while bringing GPU assets into mainstream institutional portfolios as investment-grade credit. It marks the beginning of the financialization era for AI computing infrastructure. Remarkably, when this major announcement was released, it generated almost no reaction in the capital markets. Based on my understanding of what matters most in this industry, I believe this was the single most important piece of AI infrastructure news during the first half of the year. It demonstrates that the long-standing gap between capital investment and capital returns is finally being bridged through high-credit financial structures. Its significance is enormous. For $IREN investors, three additional implications deserve attention. First, the most important signal is the fundamental upgrade of the collateral structure. This represents a true milestone. The combination of GPUs + Microsoft's take-or-pay contract cash flows + data center ownership received an investment-grade A rating. For the first time, financial markets are pricing AI computing assets with the credit quality normally reserved for infrastructure bonds rather than assigning them the risk premium typically associated with high-growth technology companies. As capital costs decline and financing channels expand, the financialization of AI computing assets becomes genuinely repeatable. For the first time, financial markets recognize AI infrastructure as an asset class that can be pledged, priced, and rated. This externally validates the foundation of the flywheel. Second, the rating agencies placed tremendous emphasis on ownership of the data centers themselves. Daniel Roberts made this point very clearly. The foundation of investment-grade financing is not the GPUs alone. Rather, it is the combination of physical ownership of the data centers where those GPUs operate together with high-quality customer contracts. Together, these two elements create the confidence necessary for institutional capital to participate while demanding much lower financing costs. In other words, IREN's build-own-operate model is not merely an operational advantage—it is also the prerequisite for upgrading its financing structure. Companies that rely primarily on leased facilities—whether CoreWeave, Nebius, or even Oracle—cannot easily replicate this credit structure. Third, this transaction establishes an industry precedent. It is not simply another debt financing. It is the first transaction of its kind in the U.S. private placement market. Fitch assigned an A rating. DBRS assigned an A(low). These ratings demonstrate that agencies recognize the structure as a relatively high-quality debt instrument, opening an entirely new financing channel for AI computing assets. Because the collateral structure has now been accepted by the market, investment-grade financing has become reality, and the capital recycling mechanism has taken shape, the third stage of the flywheel has officially begun. Why is this event so important? Because it fundamentally changes how IREN finances growth. The creation of the ABF represents a paradigm shift. Previously, expansion depended largely on shareholder dilution, market sentiment, and external equity financing. Now, the process begins with contracted rental cash flows, which become collateral for structured financing that funds new construction and attracts additional high-quality customers. IREN has evolved from "asking the market to believe in the company" to "letting the assets speak for themselves." The prerequisite for all of this is high-quality collateral. If Horizon 1 can already support A-rated debt financing today, then what could become possible once the flagship DSX-powered AI Factory is completed? Can everyone appreciate just how powerful the flywheel IREN has initiated could become? The market has long criticized IREN for shareholder dilution. But if this flywheel is successfully established, future dependence on ATM equity issuance could largely disappear. If that happens, today's decisions will prove worthwhile. Within today's aggressive and highly competitive AI industry, this type of conservative yet efficient financial leverage could prove genuinely transformative. Now that the ABF financing structure has been established, several additional amplifiers can further accelerate the flywheel. One of them is the monetization path enabled by Mirantis. Its importance extends far beyond completing the software layer. It transforms IREN's physical infrastructure from AI compute that can only be sold locally into a standardized, repeatable, exportable AI factory solution capable of commanding premium pricing. Everything—from compute orchestration and container management to large-model runtime environments—can eventually be standardized and delivered as a complete solution. This means IREN can continue expanding revenue through software and platform capabilities without proportionally increasing physical capital investment. Once its flagship AI factory is operational, this advantage should become even more pronounced. This changes growth from a linear model into a nonlinear one. The flywheel evolves from expanding through additional assets to expanding through replication of an integrated system. At this stage, the single most important milestone for IREN is Horizon's delivery schedule. Only if new capacity comes online on time and successfully attracts even higher-quality customers can capacity expansion and customer upgrades connect seamlessly. Whether the market speculates about Anthropic, sovereign AI, or enterprise sovereign AI, all of those developments rank behind Horizon in importance. Now compare this with CoreWeave and Nebius. When they sign a new deal, the value created is largely limited to the deal itself. The benefits mainly include: Receiving customer prepayments that ease funding pressure. Gaining additional credibility that helps them raise financing from banks and financial institutions. When IREN signed its Microsoft agreement, however, it achieved both of those benefits—and something far more important. It transformed those assets into A-rated ABF securities, unlocking the value of its asset base and creating a high-quality capital recycling mechanism. With 5.8 GW of secured power capacity, imagine how much capital this structure can ultimately unlock. This is precisely why IREN has not rushed to sign additional deals before completing these foundational milestones. The company first needs to provide the market with confidence that new computing capacity will be delivered on schedule. Horizon must demonstrate that capability. Once 5.8 GW of secured power translates into reliable AI factory delivery, every future customer contract can potentially become another ABF transaction. Combined with the growing cash flows from newly deployed capacity, IREN could ultimately solve the industry's greatest challenge: Building a heavy-asset business capable of funding its own future growth through internally generated cash flow. Everything required to achieve this actually began back in 2018. Secure energy assets through land acquisition, power agreements, and long-term contracts. Sign the Microsoft agreement and develop Horizon. Build the NVIDIA partnership and the DSX flagship AI Factory. Deploy Sweetwater. Complete ABF securitization with an A rating, packaging AI computing assets and contracted cash flows into investment-grade debt. Every step serves the same ultimate objective: Transform IREN from a company that depends on continuous external financing into one whose own assets generate the capital needed for future growth. Whoever achieves this first will possess the industry's hardest competitive advantage to replicate. Technology can be copied. Power shortages may eventually ease. Even data centers themselves can be replicated. But once a capital structure develops this kind of path dependency, it becomes extraordinarily difficult to duplicate. Finally, I would like to thank @FransBakker9812 for emphasizing the importance of the ABF structure and for highlighting the significance of Horizon's delivery timeline. He identified what truly matters.

@bovtokno They are targeting enterprise too.

The goal for $KEEL is to secure three major contracts before the New Year for its key sites (Panther Creek, Sharon, and Moses Lake) Potential customers include tech giants like $MSFT $META and AWS, as well as fast-growing AI clouds and model developers such as $NBIS and $CRWV

Let's make a video with Intel CTO about what the right metric should be :)

@danielisdizzy This is absolutely not happening. $IREN is plugged deeply into the $NVDA ecosystem, meeting with Jensen and sponsoring their DSX factory blueprint. Aligning with $GOOGL would burn their bridge with Jensen and knock them to the back of the line for Vera Rubin shipments.

Mark this: A deal between $IREN and $GOOGL is coming. By 2030, Google is expected to have around 32 GW of data center capacity. Roughly a quarter of that is expected to come from leased infrastructure. Unlike $AMZN, which is expected to build most of its own capacity, Google will rely much more heavily on third-party providers. $IREN is one of the few companies building AI infrastructure at the scale hyperscalers need. No doubt a meaningful portion of Google’s leased capacity will come from $IREN. All $IREN needs to take off is one major hyperscaler agreement. That moment is coming. The period between now and the next few announcements is the time to load the boat.

$INTC. Packaging is the moat.

@Omercheema Sovereign semis should be seen less as a binary outcome and more about trade leverage

Semiconductor independence is a myth. It makes for good government PowerPoint slides, but delivers zero real-world value. For years we've heard the rhetoric about self-reliance, yet the global semiconductor supply chain remains deeply specialized and interdependent:Tl Taiwan continues to dominate advanced logic chip manufacturing. South Korea remains the undisputed leader in memory (DRAM and NAND). ASML (Netherlands) is the critical equipment bottleneck for leading-edge lithography, with no viable alternative in sight for years. EDA tools are overwhelmingly American. High-end GPUs and CPUs are designed and controlled by American companies. So what “self-reliance” are we actually talking about?

I’m glad you cover all perspectives and empathize with the pain investors have been expressing. The blunt bearish take is that this spending is wasteful in the face of shareholders dealing with dilution and weak SP performance. The counterpoint is seeing the confidence behind such a move. Confident or not, management has to deliver tangible SP increases soon or they will have a revolt on their hands.

$IREN's new Warriors Sponsorship Investors are currently losing their minds over $IREN's new multi-year jersey patch deal with the Golden State Warriors, one of the most prominent teams in the NBA. Most seem to view this as wasteful spending, since the NBA caters to consumers. That may be great for B2C advertising, but the argument goes that it isn't as relevant for a B2B company like $IREN. Call it cope, but I completely disagree with that take. For one, B2B-focused advertising in sports is nothing new. CrowdStrike, AWS, Oracle, Salesforce, and Dell are all prominent F1 sponsors. Oracle also owns the naming rights to the San Francisco Giants' stadium (MLB), now called Oracle Park. Contrary to what investors seem to think right now, I believe $IREN's ad spend is highly targeted rather than random. They consistently sponsor global AI tech events, which is a direct way to get their branding in front of the exact customer base they want to reach, and they're increasingly layering in more alternative advertising channels in key markets on top of that. San Francisco is the undisputed AI and tech capital of the world. The vast majority of tech executives live and work here, which makes it an ideal market for an emerging cloud provider to advertise in. That's exactly what $IREN has been doing lately. Sponsoring one of the most prominent teams in the NBA, based right in San Francisco, is a genuinely clever way to get IREN's brand in front of the tech execs and Silicon Valley founders who attend home games and sit courtside. Every time Steph Curry hits a 3-pointer, people will see the $IREN logo, whether on television, courtside, or in local coverage. In a crowded market like SF, where the highways are plastered with tech ads from every company under the sun, this is a smart way to compete for the headspace of the key decision-makers living there. When you put it into perspective, $IREN is likely paying somewhere around $40-$60m per year for this sponsorship, which is fairly modest against the >$4b in revenue and multi-billion operating cash flow they'll generate next year. In any case, I think the bigger reason people are frustrated is that the stock isn't doing so hot right now, and $IREN hasn't signed a multi-hundred-megawatt deal since last November. The optics couldn't be worse against that backdrop, since it gives investors the impression that management is focused on all the wrong things. But I don't think this kind of thing diverts any real attention away from top management. You can execute on multiple fronts simultaneously. I also think investors should judge news like this on its own merit rather than letting price action sway their judgment. If $IREN had recently signed a big deal and was trading north of $100, I bet people would be calling this exact sponsorship a genius marketing play targeting the Bay Area. As for the lack of new deals, I have some thoughts that I'll get into in an upcoming post here on X. Don't get me wrong, $IREN is testing my patience, but I remain very optimistic that they'll deliver in Q3. You'll hear from me soon. Cheers!

Alex this is not happening. Intel needs to spend capex on 14A and capacity ramp. They aren’t going to blow their capex budget trying to get into a new space thats already dominated. It would take years to buy tools and ramp a new process and by that time the other memory giants will have increased their capacity and improved their process. It makes zero sense to take a direct fight when China is also ramping up memory.

Every Intel shareholder on X wants Intel to get back into memory biz Elon seeing 90% gross margins is prob demanding Intel help him But so far there are 0 Intel job postings for it (some Quantum) If Intel posts scientist or process engineer jobs for NAND/DRAM. Then lmk $INTC

@Alex_Intel_ TSMC raising prices is massively bullish

I get pushback on Intel's PE ratio But wafer manufacturing is a fantastic business New process nodes experience a price jump of 20-30% Maintaining gross margins means a profit gusher TSMC has long term negative headwinds + event risk All TSMC's customers are US Co's like Intel

@NuttyCLD Many people don’t understand that it’s expensive to do all the temporary bonding and debonding steps associated with dry etching TSVs, filling with liner and Copper, polishing the wafer down to reveal them. The EMIB approach is infinitely more elegant.

I didn't expect this short post to land with so many people, so I wrote a full deep-dive on Intel's advanced packaging play against TSMC. Would love for you to read it and tell me what you think. nuttycld.substack.com/p/intels-ai-pa…

@MilkRoadAI The Chanos take is going to age so poorly, this is a pretty bad look for @MilkRoadAI

Jim Chanos just made one of the most contrarian calls in the AI infrastructure trade right now (Save this). His argument is simple, the bull case for alternative energy stocks is built on a constraint that is temporary and when it resolves, the valuations collapse. Chanos's core claim is that the United States does not actually have a power shortage but rather has a permitting bottleneck and a turbine delivery backlog. His view is that if AI demand is as large as everyone says it is, those barriers will be cleared within two to three years because the economic pressure to resolve them will be overwhelming. The valuation math is what makes this compelling. Power costs represent only 5 to 7% of a data center's total revenues. Companies trading at 50, 60, or 70 times earnings and 30 to 40 times EBITDA to potentially supply a fraction of that cost line are priced as if they have won a permanent, irreplaceable monopoly on AI power. That is the bet Chanos thinks is wrong and it is hard to argue with the logic and the data partially supports his concern. Close to half of all planned U.S. data center builds in 2026 are projected to be delayed or canceled because the grid cannot deliver power fast enough. Eleven gigawatts of announced data center capacity is sitting frozen in 2026 with no construction underway, purely due to grid access limitations. That sounds like it validates the bull case for energy stocks but Chanos's rebuttal is that this is a bottleneck story, not a shortage story. A structural shortage means elevated prices for a decade, but a bottleneck means prices spike temporarily, smart capital floods in to fix the choke point, and valuations built on that scarcity eventually mean-revert. FERC has already voted unanimously to order grid operators to accelerate data center connections and the regulatory machine is beginning to move in exactly the direction Chanos predicted. Here is where the bear case gets complicated. Bloomberg NEF forecasts more than 106 gigawatts of new data center demand by 2035 and natural gas is the only fuel that can deliver round the-clock baseload power at the speed AI infrastructure requires. Small modular reactor technology is still years from commercial scale, and geothermal remains a niche contributor limited to favorable geographies. The companies trading at 60–70x earnings are not providing guaranteed power at scale but rather are providing optionality on technologies that may or may not arrive when the market needs them. Chanos is likely right on the valuation argument for the most speculative names. Companies at 50x+ earnings supplying a single-digit cost line to a data center are priced for perfection in a market that rarely delivers it. Whether that trade works in 2026, 2027, or 2028 is the harder question and the bottleneck is proving more stubborn than even the bears expected. Milk Road is tracking the power story in full, the bottlenecks, the valuations, and the companies we think are genuinely positioned versus the ones just riding the narrative. Come join Milk Road Pro and get our analysis every day for just a dollar, link below.

@kevinbhobart @LandoInvests Indeed. The thesis has even evolved into something more bullish. Sovereign AI. $IREN is positioning for the best dollar capture in a groundbreaking market, and the expert “investors” on X are whining that the stock price isn’t going up fast enough.

@LandoInvests Same. The more fear, the more I am buying. The underlying thesis is stronger than ever.

The bears don’t know I’m buying more $IREN