와트 병목 현상

THE DAILY H.E.A.T. Torching Wall Street’s Obsolete Playbook. Delivered straight to your inbox 5 days a week. Subscribe Home Posts The Watt Bottleneck The Watt Bottleneck Wall Street is still pricing power semis as an industrial cyclical. They aren't May 20, 2026 I’ve been a trader and investor for 44 years. I left Wall Street long ago—-once I understood that their obsolete advice is designed to profit them, not you. Today, my firm manages around $5 billion in ETFs, and I don’t answer to anybody. I tell the truth because trying to fool investors doesn’t help them, or me. In Daily H.E.A.T. , I show you how to Hedge against disaster, find your Edge, exploit Asymmetric opportunities, and ride major Themes before Wall Street catches on. Table of Contents H.E.A.T. News vs. Noise: What’s Moving Markets Today ETF News In Case You Missed It H.E.A.T. A once-in-a-generation rewiring of the AI datacenter — from conventional AC distribution to 800-volt HVDC — is about to crown a new tier of compounders. Here’s who gets paid, who gets dragged, and why the constraint has officially moved from the GPU to the wall. THE BOTTLENECK HAS MOVED AGAIN For three years, the binding constraint in artificial intelligence was compute. Then it was memory. Then it was the optical interconnect fabric between racks. We have written about each migration in turn — because each migration creates investable companies, and each one ends with the consensus stuck pricing the old chokepoint while the smart money has already moved to the new one. The constraint has moved again. It is now electricity — specifically, the architecture used to push electricity from the substation to the silicon. A single AI rack today draws roughly 120 kilowatts. By the end of the decade, NVIDIA’s roadmap and the Open Compute Project both contemplate racks pulling over one million watts each. You cannot get there with the low-voltage rack distribution every hyperscale datacenter on the planet was built around. The wiring melts. The breakers trip. The conversion losses eat a quarter of every kilowatt-hour you paid for at the meter. The answer the industry has settled on — quietly, while equity markets were busy debating Blackwell delivery dates — is a wholesale move to end-to-end 800-volt direct current. Introduced by the Open Compute Project in 2024, endorsed and promoted by NVIDIA in 2025, with first deployments expected in 2027. That is not a science project. That is a planned, scheduled, capital-committed architecture transition in the most important capex cycle of our lifetime. And almost no one on the sell-side is treating it as such. THE STAKES, IN FOUR NUMBERS 220 GW 9.4 GW $71M 58% High-confidence U.S. capacity additions — 26% of 2025 peak load Datacenter capacity leased by hyperscalers in Q1 2026 alone — a 5x year-over-year jump Annual opex savings per gigawatt from the 800VDC architecture vs. legacy AC Five-year CAGR for AI datacenter power semis — reaching $13B by 2030 THE PHYSICS PROBLEM NO ONE WANTS TO TALK ABOUT Power equals voltage times current. That is not a metaphor — it is the equation that governs every electron moving inside every server in every datacenter being built right now. When you take power off the grid, it arrives at high voltage. Before it reaches the chip, it must be stepped down — multiple times — to the low voltages that semiconductors actually run on. Every step costs you. The lower the voltage, the higher the current. The higher the current, the more energy bleeds off as heat in the copper. This is not an engineering inefficiency you can software-patch away. It is ohmic loss, and it is as old as the electrical grid itself. Here is what that math looks like at AI scale. At one megawatt of rack power, a 48-volt bus implies over 20,000 amps of current running through copper. That is not power delivery. That is welding. The reason the industry has to move to 800-volt distribution is not a preference. It is an arithmetic floor. In a conventional AC-fed datacenter, between 15% and 25% of every watt you buy never reaches the GPU. It is dissipated, as waste heat, somewhere between the substation and the silicon. At hyperscaler power bills running into eleven figures, that is not a rounding error. That is a structural margin leak. THE ARCHITECTURE SWITCH, IN ONE PARAGRAPH The fix is to keep voltage high for as long as possible inside the building, then drop it down only at the very last stage — right next to the chip. That is what 800-volt direct current architecture does. Heat loss drops from 15–25% to 5–8%. Per-rack distribution copper shrinks by 50–70% (total system copper still rises with rack count). Roughly $71 million per gigawatt per year falls straight to the hyperscaler’s operating line. How that pencils: a one-gigawatt facility at ~90% utilization consumes roughly 7.9 TWh per year. Cutting conversion losses by ten to fifteen percentage points saves on the order of 0.8–1.2 TWh. At wholesale power in the $60–$90 per MWh range, that is $50–$110 million in annual savings. The $71M figure sits squarely in that band. At the scale of capex now being committed, that math does not need a spreadsheet. It needs a forklift. “The constraint has moved from the GPU to the wall — and the companies that own Stage 2 power conversion own the toll booth.” WHY WALL STREET IS MISPRICING THIS Power semiconductors have, for decades, been valued the way the analog and industrial buckets have always been valued: 18–22x forward earnings, mid-single-digit revenue growth, cyclically sensitive to autos and factory automation. That is the lens. That is the comp set. That is the discount rate. It is also wrong. The AI datacenter power semis TAM is on track to compound at a 58% annual rate through 2030, reaching $13 billion — 12% of the total power semis market. That is not industrial cyclical math. That is closer to the trajectory of the optical transceiver market in 2022 or the HBM market in 2023, both of which re-rated violently once buy-side analysts noticed the underlying mix was no longer industrial — it was infrastructure for accelerated computing. There is a second, sharper edge to the mispricing. Not every power semi name is exposed to the right part of the stack. The architectural transition matters not in aggregate, but at one specific layer: Stage 2 power conversion, where the dollar content per rack grows the fastest. STAGE TAXONOMY — DEFINED ONCE, USED THROUGHOUT Stage 1: Facility-level conversion. Grid AC arrives, gets stepped down and rectified to high-voltage DC for distribution across the data hall. Stage 2: Rack-level conversion. The 800-volt DC bus is stepped down through intermediate bus converters (IBC) to feed the server boards. Stage 3: Point-of-load. Final conversion from intermediate bus voltage down to the sub-1V rails the GPU actually consumes — vertical power delivery topology lives here. Our thesis sits on the Stage 2 + IBC + vertical-power-delivery axis, because that is where dollar content per rack scales fastest, and where the named winners below have a structural lead. THE WINNERS — TIERED BY CONVICTION TIER NAME THE ROLE WHY IT MATTERS CORE — STAGE 2 PURE PLAY Monolithic Power Systems MPWR Vertical power delivery topology at the point-of-load — the literal last conversion stage before the GPU. Highest TAM materiality of any name in the comp set. Stage 2 exposure is a >100% revenue uplift opportunity by 2030. CORE — SCALE INCUMBENT Infineon IFX (XETRA) World’s largest power semi franchise. Plays across silicon, SiC, and GaN — the full materials stack. The only Western pure-play with end-to-end coverage of the 800VDC value chain. Auto cyclical optics mask the AI datacenter story. HIGH-BETA — COMPOUND SEMI LEVERAGE Allegro Microsystems ALGM Magnetic sensing and isolated drivers — indispensable in HVDC conversion stages. Stage 2 materiality is comparable to MPWR but on a much smaller base. Higher operating leverage if the architecture transition lands on schedule. HIGH-BETA — SiC/GaN onsemi ON Wide-bandgap materials specialist — silicon carbide for the high-voltage front end. Wider bandgap materials handle the higher voltages that 800VDC requires. EV demand is the cover story; AI power is the real upside. PRESSURE POINTS — WHO BEARS THE COST CATEGORY THE EXPOSURE WHERE IT SHOWS UP Legacy AC power infrastructure vendors Suppliers of conventional AC distribution gear watching their installed base get designed out of new AI builds. Service revenue tails will linger; new-build win rates will not. Watch order-book mix shift in 2027 disclosures. Generic analog incumbents Diversified analog names with no specific Stage 2 product portfolio. Will get multiple-compressed as the market separates AI datacenter exposure from industrial exposure. Per-rack copper content HVDC architecture reduces per-rack distribution copper by 50–70%. (Total copper demand at the system level still rises as rack power and rack counts climb.) A mix-shift headwind for some legacy rack-component suppliers; not a thesis to short copper miners on. Utilities without firm capacity Datacenters projected to consume 7–9% of U.S. grid load by 2030E (per analyst forecast), up from 2% in 2024. Utilities with available firm capacity get pricing power; the ones queuing interconnections lose customers to whoever can deliver electrons. CREDIBILITY FIREWALL What is confirmed today, and what is our directional read. CONFIRMED DIRECTIONAL Open Compute Project formally introduced 800VDC architecture in 2024. Adoption begins in 2027, with broad deployment by 2030. Timelines slip; the direction does not. NVIDIA endorsed and promoted 800VDC publicly in 2025. AI datacenter power semis TAM compounds at ~58% to reach $13B by 2030; 12% of total power semis market. 220 GW of high-confidence U.S. capacity additions — ~26% of 2025 peak load. Stage 2 power conversion accrues the disproportionate share of dollar content per rack. Hyperscalers leased 9.4 GW of U.S. datacenter capacity in Q1 2026 — 5x year-over-year. Re-rating of pure-play Stage 2 names from analog multiples to infrastructure multiples within 12–24 months. AC-to-chip power delivery loses 15–25% as heat; 800VDC reduces this to 5–8%. Silicon retains the majority of value; SiC and GaN take the highest-voltage stages. WHY NOW — THE CATALYST CALENDAR 2024: Open Compute Project formalizes the 800VDC reference architecture. 2025: NVIDIA publicly endorses 800VDC at its GTC keynote and in subsequent reference designs. 2026 (now): Hyperscaler capex committees are sizing 2027–2028 builds. Power architecture decisions are being locked in this quarter and next. 2027: First production 800VDC datacenter deployments expected. Stage 2 design wins disclosed in vendor revenue mix. 2028–2030: Industry-wide architectural transition. Power semis TAM compounds; the laggards retrench to industrial markets. THE BEAR CASE — HONESTLY WHAT COULD BREAK THIS THESIS Timeline slippage. 2027 first-deployment expectations are confident, not certain. Compute roadmaps slip routinely; power architecture transitions historically slip more. A two-year delay would not kill the thesis but would test the duration of investor patience. Hyperscaler bargaining power. The hyperscalers buying these chips run the most sophisticated procurement organizations on the planet. They will demand multi-source qualifications and pressure margins. The TAM may compound at 58%; gross margins for individual suppliers may not move in a straight line. Auto cyclical drag. Infineon and onsemi carry meaningful auto exposure. A sharp global auto correction could overwhelm the AI datacenter tailwind in any given quarter, creating optical setbacks that obscure the underlying mix shift. Architectural alternatives. 400VDC, intermediate-bus topologies, and on-package power delivery innovations are all live engineering paths. The industry is unlikely to converge on a single solution as cleanly as the consensus 800VDC narrative implies. Safety, standards, and serviceability friction. HVDC inside the building introduces non-trivial engineering challenges: arc fault behavior, DC breaker availability, electrical code compliance, and field-service protocols for technicians trained on AC systems. These rarely kill an architecture transition. They reliably slow it. WHAT TO WATCH — THE THREE SIGNALS 1. First disclosed 800V design wins inside power-semi vendor revenue mix. The earliest place this shows up is segment commentary, not headline guidance. 2. Hyperscaler rack power targets pushing above 250 kW in public roadmaps or RFPs. That is the threshold where the legacy architecture stops being viable on engineering grounds, not just economic. 3. Open Compute Project specification updates and supplier qualification language. OCP drives the timing here; their spec releases are the leading indicator the rest of the supply chain follows. FIVE TAKEAWAYS 1. The bottleneck has moved from the GPU to the wall. Compute, then memory, then optics — and now power architecture. The investable layer in the AI buildout shifts roughly every 18 months, and Wall Street prices each shift roughly 12 months late. 2. 800VDC is not a forecast. It is a scheduled transition. Endorsed by the Open Compute Project, promoted by NVIDIA, with first deployments in 2027. That is the kind of architectural inflection that creates a 5–7 year revenue cliff for the winners and a slow bleed for the legacy stack. 3. Stage 2 power conversion is where the dollar content lives. Not all power semi names are exposed equally. Companies aligned to rack-level intermediate-bus and vertical power delivery capture disproportionate TAM. MPWR and Infineon lead; ALGM and onsemi offer leveraged second-tier exposure. 4. The mispricing is the multiple, not the earnings. Power semis still trade on industrial-cyclical multiples. A 58% CAGR segment inside an analog book argues for a meaningful re-rating once buy-side analysts disaggregate the AI datacenter mix. 5. Watch the utility side of the trade too. If datacenters grow from 2% to 7–9% of U.S. grid load by 2030, every gigawatt of firm capacity is a strategic asset. The companies that can deliver electrons — not the ones queuing for them — set the price. The race for AI supremacy is no longer being run inside the chip. It is being run inside the rack — and the company that owns the last conversion stage owns the toll booth on the most important capex cycle of our lifetime. Space Exposure Designed For Potential Weekly Payouts The space economy is no longer just a government program — it’s becoming part of a global industry that is gradually advancing. But why view long-term growth as the only potential path forward? SPCI seeks to offer approximately 100% concentrated exposure to the space industry while striving to generate weekly income through a disciplined put credit spread options strategy. Gain exposure to the “final frontier” and a potential weekly distribution while you wait for the theme to play out. Discover SPCI. Distributor: Foreside Fund Services | Investing involves risk including possible loss of principle. News vs. Noise: What’s Moving Markets Today — # (#) Markets are green this morning after a 3 day selloff, but we have Fed minutes later today and NVDA earnings after the bell. With hopes of a rate cut all but done, it will be very interesting to see where the Fed’s head is at. Obviously NVDA will give great insight into the AI ecosystem. What Iran Tells Us About UFO Disclosure When governments confront unknown threats in their airspace, defense budgets surge and the same aerospace and surveillance companies move hardest. On March 2nd, Northrop jumped 6% and Lockheed 3.3% on the Iran news — and President Trump has since ordered the formal release of government UAP files, with the Pentagon confirming compliance. So if a conventional conflict can move these stocks this fast, what happens when the bigger story breaks? See the UFOD holdings: [thetruthisoutthereufod.com ETF News A Stock I’m Watching $WMB ( ▼ 1.91% ) Williams Companies (WMB) is the kind of stock that doesn't get pitched on CNBC at 9:30 AM but probably should. Williams owns Transco — the largest interstate natural gas pipeline system in the United States — and Transco runs straight through Virginia, Georgia, and the Carolinas, which happens to be the exact corridor where hyperscalers are building datacenters at a pace that would have sounded insane three years ago. Here's the thing nobody wants to say out loud: AI doesn't run on vibes. It runs on electricity, and a huge chunk of that electricity is going to come from natural gas turbines for the next decade because solar and wind can't deliver the 24/7 baseload that a GPU cluster demands. Williams moves the gas that powers the grid that powers the AI. The pipeline was built decades ago, it's already paid for, it can't be obsoleted by a software update, and the regulatory moat around building new interstate pipelines is so thick that competitors essentially cannot replicate it. It throws off a fat dividend while you wait. This is the HALO thesis in one ticker — a heavy asset that doesn't go obsolete, hooked directly into the most important demand story of the decade. The market keeps treating it like a boring utility. We think that's exactly the point. In Case You Missed It Great talk on with the Acquirers Podcast on markets, value investing, inverse Cramer, and Michael Gayed joins me to talk about taking income from your portfolio and how to get more than 4%…… The H.E.A.T. (Hedge, Edge, Asymmetry and Theme) Formula is designed to empower investors to spot opportunities, think independently, make smarter (often contrarian) moves, and build real wealth. The views and opinions expressed herein are those of the Chief Executive Officer and Portfolio Manager for Tuttle Capital Management (TCM) and are subject to change without notice. The data and information provided is derived from sources deemed to be reliable but we cannot guarantee its accuracy. Investing in securities is subject to risk including the possible loss of principal. Trade notifications are for informational purposes only. TCM offers fully transparent ETFs and provides trade information for all actively managed ETFs. TCM's statements are not an endorsement of any company or a recommendation to buy, sell or hold any security. Trade notification files are not provided until full trade execution at the end of a trading day. The time stamp of the email is the time of file upload and not necessarily the exact time of the trades. TCM is not a commodity trading advisor and content provided regarding commodity interests is for informational purposes only and should not be construed as a recommendation. Investment recommendations for any securities or product may be made only after a comprehensive suitability review of the investor’s financial situation.© 2026 Tuttle Capital Management, LLC (TCM). TCM is a SEC-Registered Investment Adviser. All rights reserved. Keep Reading May 19, 2026 • 2 min read I just launched something This is the trade I've been building toward for 45 years. May 15, 2026 • 10 min read The $24 Trillion Robot Has a Wrist Problem Wall Street is buying the platforms. The bill of materials says the early money is somewhere else — in four precision components that almost no Western company makes at scale. May 8, 2026 • 13 min read The WWII Era Move That's Shaping Today's AI Race Why the AI buildout is gated by a one-year grinding cycle in southern Germany — and where the bottleneck rent actually flows. View more