Telegram Key Posts

1/ The reason I could not put every position into Korea at KOSPI 3000 was that SK Hynix was still waiting for HBM-led re-rating while Samsung Electronics looked terrible. Anyone who held KOSPI positions before June 2025 deserves to be called a value investor: someone who built a long position on belief before knowing exactly when earnings would turn. You deserve it.

2/ For crypto traders, trend followers and earnings-play investors, the real entry window for Korea came after SanDisk's NAND shortage call in late September 2025. Before then, KOSPI 3000 and ETH 3000 felt aligned, and few crypto investors cared about SNDK. Crypto still looked as if it had a future as bright as AI, and DAT plus ETF demand seemed endless. People seeking AI exposure may have owned M7, NVIDIA and power equipment, but likely lacked memory exposure.

3/ Crypto's different logic was pure flow. When price rose, the asset's value simply required someone to run the crime. It was the most innovative casino in human history: a market showing how powerful wash trading and deregulation can be in concentrating wealth quickly in an asset without value. The 10.10 event may have broken that structure.

4/ I cannot say crypto will never revive. I think it will, because it is a casino that can change whenever human desire changes. But now we need to ask whether crypto can be held long term as value investors. Just as someone value-invested in Samsung below KRW 60,000 through DCA, crypto investors must become that kind of value investor if they want to find EV.

5/ Taking leveraged long futures in crypto now feels suicidal for investors who should choose value investing. Value investing is 0 or 100, but crypto forces a game where you can hit zero before value appears. The board is becoming one where the actor running the crime takes the money, not one where people share lottery tickets by rock-paper-scissors. Hunting crime coins looks lower probability than buying a lottery ticket.

6/ The leaders of the industry that people want to value-invest in are trying to pivot into AI stocks. Investors may decide to value-invest, but they are not shareholders and remain the weakest party. I find it hard to agree that crypto is the only path to life-changing wealth. It is no different from thousands of people pooling into a lottery to produce one winner, and the odds may be worse.

7/ This is not a command to buy stocks. New entries here may not look attractive after the gains value investors have already enjoyed. But there is no need to chase crime coins for life-changing money. Find assets that can truly be value-invested. The crypto scene now seems obsessed only with finding the next crime coin.

8/ Some people cannot accept the historically asymmetric AI market, and many styles do not fit that market. Still, I think crypto may shine again only after today's crime coins disappear. A longer time horizon and slower breathing are needed. If not, crypto is not yet too big to fail and can simply disappear.

9/ After watching Build and Build yesterday, I still thought: this really is not it yet.

1/ After the 2000s smartphone revolution, Apple chose to reduce Capex and returned it to shareholders 2/ A paradigm-breaking shift happened in the 2020s — following Apple's lead, major Big Tech companies (Microsoft, Google, Meta) also expanded shareholder returns instead of Capex 3/ We experienced an ambiguous period where tech stocks straddled the line between growth and value during the 2022 Russia-Ukraine war aftermath (Big Tech's shareholder returns during the high-rate era were incredibly attractive) 4/ Big Tech didn't have anything worth investing in anyway — except Meta's all-in bet on the Metaverse 5/ Now our M7 has changed. They're showing a clear investment purpose for the first time. It's AI 6/ Big Tech companies are the most profitable in the world. The market may always question current Capex levels, but up to $700B is coverable by FCF. Will this be sustainable next year? Looking at OCF growth rates, Capex above current levels appears feasible 7/ So what future do they see? They want full automation. Amazon in particular has spent massive Capex on process automation for a decade, and through strategic investment in Anthropic, has captured both Robotics and AI simultaneously 8/ If this investment bears fruit? How will M7's cashflow change when they no longer need to spend massive labor costs on planning and paying for simple tasks? How much value should we assign to the creative destruction they're creating? 9/ If you think about companies that can hire thousands of AI agents more capable than humans — maybe we're looking at the market too naively right now

Sorry for moving long posts to the web. In short:

1/ This is a different pattern from Russia-Ukraine.

2/ I think stagflation is unlikely, with weaker impact than the Russia-Ukraine shock.

3/ Memory and hardware pullbacks look like good entries.

4/ In about four weeks, I think the risk-off phase will already be over.

https://newmoneymoves.io/reports/report-haelangdal-market-view-2026-q1

The largest growth industry in the US is AI, and the most important task is building data centers for AI. The needs are largely twofold: semiconductors and power. Most stocks that have risen steadily since GPT and 2024 are tied to those two areas. NVIDIA GPUs for training came first, memory for inference now reflects the current era, and power has moved from gas turbines and transformers into fuel-cell themes. Beyond that are liquid cooling and many other themes for thermal control. It is time to organize the map and check which narratives are being missed. Semiconductors: NVIDIA and Broadcom as fabless names, TSMC and Samsung as foundries, and Samsung, SK Hynix, Micron, SanDisk and Kioxia for DRAM and NAND. Power: GE Vernova, Eaton and Quanta for turbines and infrastructure; Hyosung Heavy Industries, HD Hyundai Electric and LS Electric for transformers and distribution. For big tech, GPUs, TPUs and memory remain core. The changing bottlenecks are packaging process shifts, copper-wire power and heat issues, and the search for energy sources that solve land and environmental constraints.

The core is no longer a single chip. A system package determines performance, yield, cost and power efficiency. AI accelerators now compete through advanced packages integrating die/chiplets, HBM, interposers or bridges, fine RDL wiring and ABF substrates. Advanced 2.5D/3D packaging faces bottlenecks in lines, equipment, process talent and testing. Key players include TSMC, Samsung, Intel and OSATs such as ASE, Amkor and JCET. HBM packaging and stacking combine memory with package yield, heat, power and signal integrity. Equipment, bonding, assembly and inspection vendors matter as hybrid bonding and chiplets increase process difficulty. Do not look only at CoWoS. OSAT, packaging equipment, substrates, testing and burn-in need to be viewed as one basket.

The copper-wire heat issue combines two problems: power transmission/distribution and data movement. As rack power density rises, current surges, causing I2R losses, heat and cable or busbar growth. The near-term path is 48V racks; the longer-term path is 800V-class HVDC architecture in high-density zones. Power-side baskets include SiC/GaN power semis, power distribution and protection, UPS/PDU/busway, cables and conductors, connectors, transformers, switchgear and distribution networks. In many cases, transformers, switchgear and interconnection lead times become the first bottleneck before generation. For data movement, reducing copper means moving toward photonics. AI clusters spend enormous power on networking and interconnects, so CPO and silicon photonics matter. The full AI Factory map requires compute/memory, power and networking via switch ASICs plus optical.

Big tech is building several energy-source portfolios not only because demand is large, but because lead time, regulation, carbon, water and 24/7 availability all matter. Nuclear and SMR offer 24/7 large-scale carbon-free power but face permitting, time and policy risk. Geothermal offers 24/7 baseload but is location-constrained. Renewables plus storage provide scale and speed, but intermittency and grid linkage are key. Onsite gas turbines or gas engines avoid grid delays but face carbon, fuel-cost and regulation issues. Fuel cells are onsite, relatively clean and modular but depend on fuel economics and policy. Representative baskets include GE Vernova, Siemens Energy and Mitsubishi Power for turbines; Caterpillar, Cummins and Rolls-Royce mtu for gas engines and generators; Bloom Energy for fuel cells; and Quanta Services for grid EPC.

Packaging includes OSAT, equipment, substrates and testing. Add networking through switch ASICs and optics/CPO. Track the power architecture shift from 48V to 800V HVDC. Watch grid bottlenecks in transformers, switchgear and interconnection lead times. That is the map for now.

The memory thesis has been repeated since late October, so the core should be clear by now. - NVDA: sales 187B, revenue 99B, MC 4.5T, OPM 52% - TSMC: sales 122B, revenue 55B, MC 1.7T, OPM 45% - MU: sales 42B, revenue 12B, MC 0.5T, OPM 28% - SK Hynix: sales 70B, revenue 27B, MC 0.5T, OPM 38% - SanDisk: sales 13B, revenue 4B, MC 79B, OPM 37%, assuming 3Q FY25 revenue of 4.4-4.8B If SK Hynix 2026 revenue reaches KRW 130-140T, around $100B, with 60% GPM and similar OPM, a simple TSMC valuation comparison implies MC around 1.3T. It cannot be directly compared with foundry, but 0.5T still looks cheap.

Gold and silver look like a show of force around a new Fed chair nomination: the market may not tolerate excessive dollar weakness. Chinese investors in particular appear intensely focused on commodities. This could be systemic risk, but CME open interest of 152,000-170,000, which normally requires roughly 1-5% physical silver, plus global silver ETF assets near KRW 100T, does not look large enough to create a systemic crisis even under a silver shortage scenario.

Big tech capex and margin pressure have been repeated concerns. Meta's latest earnings, however, helped justify heavy capital investment. Meta showed that internally developed AI can be embedded into its own platform and open monetization pipelines. That reduced part of the fear that companies are simply pouring money into an illusion, which means the AI cycle can continue.

Trump's 10% credit-card interest-rate cap idea means traditional banks need compensation, and Wall Street's lobbying power is strong. That is why stablecoin interest income and regulation of DeFi operators are being discussed through the Clarity Act. Because the US has the strongest capital base, once it regulates stablecoins, projects must follow US law to access US capital. Tether and Circle may become able to freeze stablecoins tied to AML issues at any time, and non-compliant crypto activity could shrink.

SpaceX launch frequency is increasing exponentially. Starlink has only about 9M subscribers today, but if SpaceX raises $1.5T after an IPO, it could take share from traditional networks at geometric speed. That would mark the end of the WWW-era telecom structure.

The missile-war era is changing. EMP, ultrasound, AI, drones and off-road autonomous driving are creating a hegemony fight between traditional defense contractors such as Lockheed Martin and AI defense companies led by Palantir and Anduril.

Traditional software companies such as Salesforce, ServiceNow and Adobe face the same pressure. AI is entering every part of their territory. The likely result is that 2026 becomes the year when new winners emerge from these hegemony battles. Volatility was already high and may rise further. Semiconductors are the one group outside many of these fights. That may explain why Samsung, SK Hynix and Micron are running so hard.

My goal is to reduce risk assets by January or February 2026. I want to bet on a path opposite the market's view: weak first half, stronger second half. It is fortunate that Lighter can TGE before then, and I hope first price discovery happens by January or February. Looking at the market, AI is not a bubble. But whether share prices can keep rising is a different question. I want to lay out negative sentiment while keeping all possibilities open.

1/ Have CSPs already bought enough GPUs? NVIDIA supply-chain growth could slow. 2/ GPUs hoarded since 2024 may face useful-life issues in 2026. When can OpenAI actually make money? 3/ If Gemini catches OpenAI's share, where can OpenAI raise capital? No clear answer. 4/ Can grids outside Virginia support 1GW data centers reliably? No clear answer. 5/ Are current commodity price surges really low enough not to affect inflation and liquidity? 6/ US power generation has barely grown for a decade. Can this really be solved without power-price spikes?

Trump has stopped talking as much about the stock market. He may use the equity market as a card to reverse approval pressure from rising US living costs. Solving items 1-6 requires massive liquidity, and ordinary companies or financial markets may not be enough. The answer is the government. The US government may need to guarantee or nationalize the AI drive in practice. A death valley may appear in 2026 before real revenue arrives, creating a stock-price decline. After that, the best scenario may be government-backed liquidity ahead of the midterms. Crypto has ended a long upcycle and entered winter again after DAT, ETFs and presidential meme narratives were exhausted. Fundamentals may matter more than ever as VC, retail and institutions step away. After 2018, VC entered; after 2022, institutions entered. This time, the key question may be whether nation-state capital can be attracted. For the US, keeping only tokenization infrastructure may still be enough, because Americans' true asset is the S&P 500.

Hypothesis A: reach AGI around 2030 through scale plus engineering. Main candidates are OpenAI, Anthropic, DeepMind and xAI. The idea is that larger models, more data, more compute, RLHF and better safety techniques can get there. Hypothesis B: current systems are weak at hidden-state inference, world models, causal reasoning and persistent reinforcement learning, so a new architecture is needed. Candidates include Imbue and Sakana AI, with a new architecture that combines world model, planner and memory. In practice, big tech companies backing hypothesis A are burning more capital while preparing for the model shape implied by hypothesis B.

The concern is whether Oracle must pay enormous debt costs before RPO converts into earnings, and whether the AI bubble is connected to private credit.

Oracle's debt and private-credit issues matter, but the bigger point is the fight to decide the industrial leader of 2023-2033. 1995-2006 belonged to MSFT. 2007-2023 belonged to AAPL. Apple created the iPhone and iOS ecosystem in 2007 and led the mobile era for a decade. After that, Apple needed fewer assets because it did not need to own factories, and cash was returned to shareholders rather than invested in the AI era. Apple effectively stepped back from the competition. Other cash-rich big tech companies are defining 2023-2033 as the AI era and entering a costly war. They have seen the monopoly profit Apple enjoyed, and they also know what they could lose if a competitor reaches AGI first. No one can step back from this AGI fight, even if it means spending nearly all available capital. Memory companies, power companies and governments face the same logic. A country that succeeds in AGI may open an unbridgeable technology gap. This is a giant chicken game, larger than the early-2000s DRAM price war that bankrupted many Japanese and Taiwanese producers. It can continue until big tech FCF and major US bank capital are fully pulled into capex.

1. Has NVIDIA's GPU scaling law actually broken, or has the market simply shifted toward inference and test-time compute scaling? NVIDIA earnings should provide the first check on whether anything matters more than trillions of parameters. 2. If so, does NVIDIA's non-GAAP margin near 72% face pressure? Is HBM price inflation creating cost pressure? 3. From a TCO perspective, does a strategy of short GPU and long ASIC still work? 4. How real is Jensen Huang's $500B new-order narrative, and how are hyperscalers handling physical constraints such as land and grid capacity? My personal view is that NVIDIA is the god of this cycle, so doubting it may be dangerous. The FOMC minutes before earnings are also important, so market volatility can be extreme. Bitcoin may fall further, but I increasingly wonder how much downside remains. Macro data is negative, yet the AI infrastructure buildout may buy the market more time before any bubble breaks.

1. Buy the dip in AI-related equities, because if AI replaces us in the long run, what else is there besides equities? 2. Coinbase Monad sale 3. Ostium LP farming and product testing 4. Hope for Lighter 5. Watch prediction markets I have always been conservative about airdrop work. If I cannot understand what advantage a project offers, I do not do much. I cannot honestly recommend perp DEX farming broadly, though Ostium looks interesting because it has a special angle. I am skeptical that BTC can make new highs soon and think the timing for broad crypto liquidity has passed. As stock tokenization progresses in 2026, capital may move more toward equities. Crypto projects that can connect to that flow may matter; most other alts are harder to justify. If a product is good, users will come and the flywheel will turn. Projects driven mostly by yapping can become exit benchmarks, so build your own standards and find what is actually good.

A basic thesis may be getting missed. Semiconductor stocks do not receive NVIDIA, Google or Broadcom-level P/Es because they are cyclical. They can make money, but their cycles are so violent that share prices move wildly. Samsung Electronics and SK Hynix were not fabless designers, so they were not exceptions. Add the Korea discount and they had been price takers, with pricing power held by chip designers and buyers. That is changing. Samsung and SK Hynix are beginning to gain pricing power in HBM, DRAM and NAND, the key chips needed for AI infrastructure. The secret is capex discipline. After past cycles, companies that overinvested during upturns disappeared, and few firms now want to challenge Samsung and Hynix in another money game. US hyperscalers are in a hurry to build data centers and may spend heavily on semiconductors. Supply is limited, so prices rise. I do not know what 2026 earnings will be, but the setup is clear.

People ask why coins are not rising, but when asked why they should rise, there is no good answer. Across many cycles, hundreds of thousands of coins have been issued, but among thousands of projects perhaps fewer than 50 are still doing real work. A better industry cycle requires at least the will to keep building products. Crypto may rally again when the 2026 rate-cut cycle comes, but can it rise as explosively as AI equities? I doubt it, because it is not as new, innovative or useful as AI. The next liquidity wave may come from equity, bond and real-world asset tokenization, but most existing altcoins do not obviously deserve it. This is not a claim that crypto dies. It means liquidity into coins is not obvious. Focus on AI, find viable perp DEX projects if any, and diversify with clear standards.

My view is that the 401(k) executive order was ultimately for bitcoin. By opening retirement accounts to alternative assets, it created a path for bitcoin to enter the 401(k) system. The S&P 500's long-term rise was possible because of US corporate competitiveness plus continuous inflows. A structure with endless inflows stops looking like a Ponzi and becomes something other than a zero-sum game. That foundation is the scale of US retirement assets and the high equity share of household assets. US defined-contribution assets are around $12.2T, and total retirement assets are around $43T. Most of that DC money is already invested in US equities, which is one reason the US remains such a strong consumption economy.

In 2019, China tried to ban bitcoin and blockchain activity partly because Tether was being used for capital flight. Before that, China had been the center of bitcoin mining and price influence. China is a tightly controlled capital market. Property tycoons needed ways to move money abroad, and USDT became one channel. During the Evergrande crisis, rumors circulated that Tether held Evergrande commercial paper. Some of that may have been spread by Wall Street players who wanted stablecoin share or bitcoin bulls who wanted the US to become bitcoin's center, but I think the possibility was real. China did not simply allow Evergrande to go bankrupt because it likely saw signs that valuable assets had already been moved offshore through channels including Tether. Instead of liquidation, it effectively trapped the company until debts were repaid.

Russia, the world's second-largest military power, has continued the war for more than three years. Western support made that possible, but the early economic sanctions shocked the world. Dollar-denominated bonds could become paper. From then on, China began accumulating gold. It used Russia as a live experiment and adjusted the roadmap for its own economy. Central banks, especially in China-aligned countries, accelerated gold accumulation.

Trump may pursue personal interests, but he is also a conventional US nationalist. His slogan was Make America Great Again. He is loosening rules and reshaping institutions to make bitcoin a new form of gold. The endgame could be 401(k) inclusion and bitcoin as a national strategic asset. By institutionalizing stablecoins, pulling demand into short-term US Treasuries and weakening China's gold-reserve strategy, bitcoin can become a powerful card for pulling global capital back toward the US. AI and bitcoin may create a new paradigm shift like the industrial revolution's new bourgeoisie. The question is what position to take within that wave.

The point is simple: crypto is facing another test, and many people left in the market no longer try to measure real value. They only look for new packaging methods for exit liquidity. "There is not a single altcoin I can recommend." Most individual investors do not need to know every macro detail to live their lives. But for the small group reading this, macro matters because it is more useful than another post saying BTC support is 105K and resistance is 110K. Macro is the base layer of bottom-up investing.

The S&P 500 is not just an index that goes up, hits all-time highs or drops 10% in a day. It tracks the equity performance of 500 large US-listed companies, and the meaning of those 500 leaders changes by era. The leaders were once Intel, Exxon Mobil, Walmart and GM. Apple may have been a defining S&P 500 company for the past decade, but that does not guarantee the same role for the next decade.

Today the S&P 500 leader is NVIDIA. In April 2023, NVIDIA traded near $267 pre-split, with trailing P/E around 156x and forward P/E around 60x. Back then, the market still remembered NVIDIA through gaming and bitcoin mining more than data centers. Its GPUs then became central to high-performance AI compute, and the S&P 500 market-cap ranking began to change. The small altcoin tied to that thesis was Render, which could be explained through Otoy's rendering network and decentralized compute narrative. The best valuation framework was not a static multiple. It was the ability to read the direction of the era and accept how the world was changing.

Past altcoins could sometimes be valued through peers and potential value creation, even if imperfectly. Looking at today's market, especially the humanity-coin style of assets, that is much harder. It often feels faster to play a casino game. Few projects create value; many simply struggle to exit without value. The surprising part is that few people say it directly. The era is changing, and it is sad to see how little room may remain for altcoins outside blockchains issued and operated by centralized companies. There is not a single altcoin I can recommend.