구글은 의도적으로 돈을 잃으려 한다 — 그것이 바로 당신이 주목해야 하는 이유다

THE DAILY H.E.A.T. Torching Wall Street’s Obsolete Playbook. Delivered straight to your inbox 5 days a week. Subscribe Home Posts Google Is About to Lose Money on Purpose — And That's Exactly Why You Should Be Paying Attention Google Is About to Lose Money on Purpose — And That's Exactly Why You Should Be Paying Attention Alphabet may be walking into a free-cash-flow trough not because the business is deteriorating — but because it is pre-buying the next AI bottleneck: gigawatts. Jun 10, 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. The AI Buildout Has a Physical Layer News vs. Noise: What’s Moving Markets Today ETF News In Case You Missed It H.E.A.T. The direct trade is not just GOOG. It is the power-delivery stack that has to make those gigawatts usable. Alphabet is not suddenly becoming a bad business. It is becoming an industrial power buyer. That is the real story. On June 9, 2026, thematic research analysts at TD Cowen published a proprietary gigawatt capacity model covering Alphabet's entire datacenter buildout trajectory through 2031. Their model projects that Alphabet's compute capacity — roughly 7 gigawatts in 2025 and an estimated 11 gigawatts in 2026 — could reach 39 gigawatts by 2031, a compound annual growth rate of approximately 29%. Their model further estimates that this buildout will push Alphabet's free cash flow negative in 2027, before a sharp recovery in 2028 and beyond. Most investors will read the negative FCF as deterioration. That is the wrong frame. The cash is not disappearing. It is being converted into compute capacity, data centers, networking equipment, substations, transformers, switchgear, cooling systems, and long-duration power access. By 2031, if the TD Cowen model is directionally correct, that infrastructure will be generating an estimated $317 billion in annual free cash flow — and the incremental revenue-to-incremental-capex ratio will have expanded to a projected 6.6x. The question is not whether Alphabet's spending is rational. The question is which part of the supply chain captures the value first.   "The bottleneck has moved from the GPU to the gigawatt. The trade is not just GOOG — it is everything that sits between the power grid and the data center floor."   THE GIGAWATT STORY NOBODY IS TELLING Let's talk in the language that actually matters here: power. Not software. Not algorithms. Watts. In 2022, Alphabet's total compute capacity was roughly 3.5 gigawatts. By 2025 it had grown to approximately 7 gigawatts. By 2027, according to the TD Cowen model, it will reach an estimated 18 gigawatts — more than doubling in two years. By 2031, the projection reaches 39 gigawatts, representing nearly a tenfold increase over a decade. That is not an IT budget. That is national-grid-scale industrial infrastructure. Wood Mackenzie projects the U.S. data-center electrical-equipment market rising from $20 billion today to $65 billion by 2030. JLL projects approximately 100 gigawatts of new global data-center capacity between 2026 and 2030. Alphabet alone represents a meaningful share of that total. This is not incremental server-room spending. This is the construction of a new industrial order — and the companies that build the substations, transformers, switchgear, and cooling systems to make it functional are sitting at the binding constraint. THE BOTTLENECK MIGRATION THESIS — Applied Here The binding constraint in AI infrastructure has been rotating. Compute was first. Memory followed. Networking and interconnects came next. Now, for the largest hyperscalers, the bottleneck has migrated to power — and specifically to power delivery: the substations, transformers, switchgear, and cooling systems that sit between the grid and the GPU. Alphabet's gigawatt expansion is not just a Google story. It is a demand signal that cascades through every layer of the power delivery stack. When Google doubles its load, every supplier in that chain has to double alongside it — or become the constraint. Follow the bottleneck. THE FCF TROUGH: WHAT THE MODEL ACTUALLY SAYS To be precise about the sourcing here: the negative free-cash-flow projection for 2027 is a TD Cowen model estimate, not an Alphabet management commitment. What Alphabet has publicly disclosed is the direction of travel. In Q1 2026, Alphabet reported $35.7 billion in capital expenditures for the quarter, with the overwhelming majority going to technical infrastructure. Q1 2026 free cash flow was $10.1 billion, and trailing-twelve-month free cash flow was $64.4 billion, down from $73.3 billion for full-year 2025. The company is not hiding the trajectory. The spending is real and accelerating. What the TD Cowen model does is translate that spending rate into a gigawatt capacity forecast and then project when the revenue and operating cash flow will begin to exceed the capex burden. Their estimate: Google Cloud's capex-to-sales ratio peaks in 2026 at approximately 135%, before improving as built capacity starts generating revenue. The critical inflection is 2028. The model estimates that by 2028, for every dollar of incremental capex Google spends, it will generate $6.60 in incremental revenue — compared to a 0.7x ratio during the peak investment years of 2024 through 2027. If that projection is even half right, the investors buying the trough now are buying the cheapest point in the cycle.   "19 billion tokens per minute. That was Google's API consumption in Q2 2026, up from 7 billion just three quarters earlier. The demand is not theoretical. It is already here."   THE TOKEN SURGE IS THE DEMAND PROOF Skeptics of the AI infrastructure buildout thesis have one legitimate question: what if the demand never materializes at the scale the models assume? What if inference gets so cheap that the revenue never catches the capex? Google's own disclosure is the answer. In Q3 2025, its first-party models were processing 7 billion tokens per minute via direct API. By Q4 2025, that number had climbed to 10 billion. By Q1 2026, 16 billion. By Q2 2026, 19 billion. In total, Alphabet is now processing 3.2 quadrillion tokens per month — approximately 330 times the volume it was handling in May 2024. More than 8.5 million developers are building with Google's models, and more than 375 Cloud customers each processed over one trillion tokens in the past year. The infrastructure investment is not speculative. The demand already exceeds the current supply. Google is building to catch up with its own users. GOOGLE IS NOT WAITING FOR THE GRID Here is a detail that the gigawatt model does not fully capture: Google is not passively hoping the grid will scale to meet its needs. It is acting like an industrial power buyer, pulling forward supply through long-duration offtake agreements. Reuters reported in June 2026 that Fervo Energy has signed a 3-gigawatt geothermal power agreement with Google — one of the largest clean-power deals in the history of the technology industry. Geothermal is notable precisely because it is dispatchable, always-on generation: unlike solar or wind, it does not require battery storage to serve as a baseload power source for a 24-hour data center load. The implication for investors: the opportunity is not only in data center construction. It is in the suppliers that shorten time-to-power — the EPC firms building substations, the electrical equipment manufacturers filling the transformer backlog, and the clean-power generators that already have contracts and grid interconnection. Those are the companies that get paid first. WHAT THIS MEANS FOR THE CAPEX-EXPOSED SUPPLY CHAIN The TD Cowen model's most important finding for investors outside of GOOG may be this: current consensus capex estimates for Alphabet are — by that model's reckoning — 23% too low for 2027 and 19% too low for 2028. If that directional view is correct, the spending surprises are not behind us. They are ahead of us. Every major supplier in the power delivery and data center construction chain has more revenue visibility than is currently reflected in consensus estimates. The Wall Street labels that cover these companies — utilities, industrials, niche equipment manufacturers — are lagging their underlying cash flow realities. Categories are lying. The better description for a transformer manufacturer serving hyperscaler load is not 'industrials.' It is AI infrastructure. THE BEAR CASE The supply chain constraints are real. Hardware availability and power interconnection timelines are not fully within Google's control, and any meaningful slip in datacenter completion schedules would push the revenue ramp further into the future. The 2027 FCF trough assumes the buildout proceeds largely on plan — a material delay extends the pain period. There is also the competitive convergence risk. Microsoft Azure and Amazon Web Services are executing similar capacity expansions on similar timelines. If all three hyperscalers land their buildouts within the same window, the AI inference market could become supply-rich faster than demand absorbs it, compressing the pricing power that underlies the 6.6x incremental revenue-to-capex projection. The model efficiency curve cuts in the wrong direction for raw infrastructure plays. As Google's Gemini models and the broader open-source ecosystem become more compute-efficient per unit of useful output, the raw gigawatt demand per dollar of enterprise AI spend could moderate. The demand is real. The question is how many gigawatts it ultimately requires. And there is a fourth risk the market has not fully priced: regulatory and cost-allocation backlash. Data-center load is colliding with local power bills, water usage, permitting, and grid-cost allocation in markets across the country. The IEA projects data-center electricity consumption more than doubling to roughly 945 TWh globally by 2030, with U.S. data centers accounting for nearly half of U.S. electricity demand growth between now and 2030. If regulators force hyperscalers to fund more grid upgrades directly — or if communities begin blocking interconnection applications — the buildout still happens, but returns migrate away from cloud platforms toward the utilities and EPC firms absorbing the mandated investment. That is not a reason to avoid the trade. It is a reason to own the right layer of it. THE POWER-DELIVERY SUPPLY CHAIN: A NAMED SCREEN The categories are lying. These companies are indexed as utilities, industrials, and equipment manufacturers. Their underlying cash flows increasingly look like AI infrastructure revenues. Wood Mackenzie projects the U.S. data-center electrical-equipment market reaching $65 billion by 2030, up from $20 billion today — and that growth has to go through this stack.   Layer Names to Watch What They Supply Why It Matters Now Electrical Equipment ETN, HUBB, GEV, NVT, SBGSY, ABBNY Switchgear, transformers, power distribution, grid equipment — the physical layer between utility feed and data center bus Transformer backlog measured in years; hyperscaler capex 23% above consensus for 2027 means order books are understated Data Center Power & Cooling VRT, ETN, TT Rack-to-grid power conversion, thermal management, UPS systems — the infrastructure that keeps compute alive inside the building Cooling density requirements rising with AI GPU power; liquid cooling penetration accelerating, driving replacement cycles Grid Construction / EPC PWR, MTZ, MYRG Substation construction, transmission, interconnection work — the labor and civil infrastructure connecting hyperscaler campuses to the grid Interconnection queues 5-7 years in key U.S. markets; firms with existing utility relationships and permitted sites command premium Firm Power / Clean Generation CEG, VST, ORA; Fervo Always-on, dispatchable generation with hyperscaler offtake potential — geothermal, nuclear, and peaking power that can serve 24/7 data center load Google 3GW geothermal deal with Fervo (Reuters, June 2026) is a template; contracted generation now carries AI demand premium Co-location / Powered Shells DLR, EQIX Existing grid connections, permitted capacity, and built-out data center shells — the fastest time-to-power option for hyperscaler overflow New permits and interconnection take 3-7 years; existing powered capacity increasingly scarce in Tier 1 markets   TIERED WINNERS: BROADER BENEFICIARY MAP   Tier Company / Category Why They Win Catalyst Risk Tier 1 Power Delivery Stack (see screen above) Sits at the binding constraint layer; capex estimates 23% above consensus for 2027 means demand pull already understated Hyperscaler buildout accelerating through 2027; transformer and switchgear backlogs extending Long lead times create revenue lumpiness; project delays shift recognition quarter to quarter Tier 1 Hyperscaler Cloud Platforms (GOOG, AMZN, MSFT) 19B tokens/min and 3.2Q tokens/month validates demand; FCF trough in 2027 = asymmetric entry for patient capital Capital efficiency inflects in 2028; incremental revenue/capex ratio projected at 6.6x vs. 0.7x during trough Model estimates, not management guidance; delay in datacenter completion pushes revenue ramp right Tier 2 Custom Silicon / AI Accelerators Google TPU buildout scales with GW expansion; merchant silicon demand floor elevated regardless of hyperscaler winner Sustained capex above consensus; enterprise AI workload proliferation supports multi-vendor demand Model efficiency gains reduce GPU intensity per unit of useful AI output over time Tier 2 Contracted Clean Power Generation Google geothermal deal (3GW, Fervo) signals long-duration clean offtake demand; nuclear and geothermal assets re-priced More hyperscaler clean-power deals expected; IEA projects data-center consumption nearly doubling to 945 TWh by 2030 Permitting and development timelines for new generation are long; existing contracted assets benefit most near-term Tier 3 Enterprise AI Software / API-Native SaaS Downstream beneficiary as inference costs fall post-buildout; token volume growth supports consumption-based pricing migration Token usage surge creates favorable unit economics for platforms built on metered API access Gross vs. net revenue distinction matters; usage volatility creates quarter-to-quarter earnings noise   PRESSURE POINTS: WHO GETS SQUEEZED   Company / Category The Problem Pressure Point Bear Case Legacy / Non-AI-Optimized IT Infrastructure Hyperscaler capex rotating to AI-optimized compute; traditional server, storage, and networking share shrinks as percentage of total spend Procurement redirecting budgets to custom silicon and AI-specific networking; refresh cycles skipped in favor of AI-native architecture Multi-year displacement cycle; legacy vendors lose pricing power as installed bases age without replacement Uncontracted / Poorly Located Generators; Utilities Forced to Socialize AI Grid Costs Firms without hyperscaler offtake agreements may see margin compression from rising interconnection demand; regulated utilities face rate-base uncertainty as large industrial loads renegotiate Interconnection queue backlogs of 5-7 years in key U.S. markets; regulators in several states beginning to allocate grid upgrade costs to industrial demand Stranded capacity risk if AI demand moderates; political and regulatory exposure rises as data-center load becomes a visible driver of residential power bills Seat-Based Enterprise Software Without Work-Unit Migration Path AI agent deployments reduce per-seat headcount; seat-only pricing is structurally exposed as enterprises automate tasks previously requiring human licenses Agentic workflows accelerating; CFOs reallocating IT budgets to consumption-based AI platforms; ROI pressure compresses seat-license multiples Revenue model mismatch persists until vendors migrate to work-unit pricing; companies still pricing per seat face structural multiple compression Short-Cycle Capex Investors Reading 2027 FCF Headlines Trailing earnings multiples will misread the 2027 FCF trough as business deterioration; consensus capex estimates already 19-23% too low, meaning D&A ramp is also understated Quarter-by-quarter FCF negative headlines create noise; investors using P/E or FCF yield on current-year estimates will sell the bottom Thesis requires 18-24 month patience; position sizing and drawdown tolerance determine whether this is a trade or a trap   CREDIBILITY FIREWALL   COMPANY-DISCLOSED MODEL-DERIVED (TD Cowen) EDITORIAL VIEW Google first-party API tokens/minute: 7B (3Q25) → 10B (4Q25) → 16B (1Q26) → 19B (2Q26) [company disclosure] Alphabet compute capacity: ~11GW estimated 2026, reaching 39GW by 2031 at ~29% CAGR [TD Cowen gigawatt model, June 9, 2026] The FCF trough is the entry signal, not the exit — investors reading trailing FCF as deterioration are mispricing the cycle Total Alphabet tokens/month: 3.2 quadrillion, ~330x increase vs. May 2024 [company disclosure] Alphabet estimated GW online by 2027: ~18GW — more than doubling from 2025 [TD Cowen] Bottleneck has migrated from compute to power delivery; transformer/switchgear/cooling suppliers are the next binding constraint (our framework) Active developers on Google models: >8.5 million [Google I/O 2026] Google Cloud capex-to-sales ratio: estimated 135% peak in 2026 [TD Cowen] Wall Street sector labels — 'utilities,' 'industrials' — are lagging underlying cash flow realities for power-delivery suppliers (categories are lying) Cloud customers processing >1 trillion tokens/year: >375 [company disclosure] Projected 2026 Alphabet FCF: $30.1B [TD Cowen estimate] Consensus capex estimates being 19-23% too low implies power-delivery supply chain revenues are also underestimated Q1 2026 Alphabet capex: $35.7B, majority to technical infrastructure [Alphabet Q1 2026 earnings] Projected 2027 Alphabet FCF: negative $14.9B [TD Cowen estimate] Regulatory backlash on grid-cost allocation could migrate returns from hyperscalers toward utilities and EPC firms — owning the right layer matters Q1 2026 Alphabet FCF: $10.1B; TTM FCF: $64.4B [Alphabet Q1 2026 earnings] Projected 2028 Alphabet FCF recovery: $50.8B [TD Cowen estimate] Short-cycle investors will misread 2027 FCF headlines; patient capital has asymmetric positioning opportunity at the trough Full-year 2025 Alphabet FCF: $73.3B [Alphabet 2025 annual report] Projected 2031 Alphabet FCF: $317.6B [TD Cowen estimate]   Fervo Energy 3GW geothermal deal with Google [Reuters, June 8, 2026] Incremental revenue/incremental capex ratio: 2.5x (2019-23 avg), 0.7x (2024-27E avg), 6.6x (2028E) [TD Cowen]   U.S. data-center electrical-equipment market: $20B today rising to $65B by 2030 [Wood Mackenzie] TD Cowen capex estimates: 23%/19% above 2027/2028 street consensus [TD Cowen thematic research]   Global data-center electricity consumption projected at ~945 TWh by 2030 [IEA Energy and AI report] ~100GW new global data-center capacity expected 2026-2030 [JLL]     FIVE THINGS TO TAKE AWAY FROM THIS ISSUE   1. The FCF trough is the entry signal, not the exit. TD Cowen's model estimates Alphabet goes free-cash-flow negative in 2027 — not because the business is broken, but because capex is at its peak before the revenue ramp begins. The incremental revenue-to-capex ratio is projected to reach 6.6x by 2028. Investors who understand investment S-curves are buying the trough. Everyone else is selling the headline. 2. 39 gigawatts by 2031 is a purchase order for the entire power-delivery stack. Google's compute capacity growing at a projected 29% CAGR through 2031 is not a forecast in isolation — it is a demand mandate for every layer of the power-delivery supply chain. Transformer backlogs, substation construction, cooling density, and grid interconnection capacity are all being consumed faster than current consensus pricing reflects. 3. The token data is the demand proof. 19 billion tokens per minute in Q2 2026, up from 7 billion in Q3 2025. More than 3.2 quadrillion tokens per month. More than 375 Cloud customers each processing over one trillion tokens per year. This is live, paying consumption — not a model assumption. The buildout is chasing real demand. 4. Consensus capex estimates are wrong — and that means supply-chain revenues are also understated. The TD Cowen model puts 2027 and 2028 hyperscaler capex 23% and 19% above current street estimates. If that analysis is directionally correct, every supplier in the power-delivery chain has revenue upside that has not yet been priced into consensus. 5. Own the right layer — the regulatory risk is real. The IEA projects data-center electricity consumption nearly doubling to 945 TWh by 2030. That scale of demand is beginning to attract regulatory scrutiny on grid-cost allocation. If regulators force hyperscalers to fund more grid infrastructure directly, returns will migrate toward utilities and EPC firms. The trade is not 'buy everything in power.' It is 'buy the layer that captures the mandate regardless of who pays.'   The AI Buildout Has a Physical Layer Many of today’s data centers are still using copper wiring. The same metal we’ve been using for a hundred years. At the speeds AI demands with data moving between thousands of GPUs, billions of times a second, copper doesn’t just slow down. It turns that data into heat. The more you push through it, the worse it gets. There’s no software for fix for that. So what’s the answer? Explore the Photonics Layer….. Tuttle Capital Pure Play Photonics ETF (FOTO) Distributor: Foreside Fund Services | Investing involves risk including possible loss of principle. News vs. Noise: What’s Moving Markets Today Definitely seeing some toppiness in the market here…. Yesterday was kind of wild, which is why I like the visual of candlestick charts. At one point semi’s where getting crushed, only to finish down 1.2%. Semi’s (SMH) are down about 2.5% as I write this. Are we at THE top, A top, or just pausing? Is this investors lining up to buy SPCX? We will know with hindsight. Should you be more cautious here? Probably. We get CPI this morning ahead of Warsh’s first Fed meeting next week. That ought to be important. Where Does the Money Go When AI Hits a Wall? When capital chases a tech theme, it tends to pile into the most obvious layer and miss the one underneath. AI spending is now bumping hard against memory. Hyperscalers — the big cloud builders like Amazon, Google, and Microsoft — have shifted memory from 8% of their build budgets to an estimated 30% in a single cycle. That capital has to go somewhere. If the constraint is memory, and the build can't move without it, shouldn't an investor own the layer AI runs on? View HBMX fund holdings → Distributor: Foreside Fund Services | Investing involves risk including possible loss of principal. <Link = http://www.hbmxetf.com/> ETF News $UFOD ( ▲ 4.66% ) $TDAX ( ▲ 3.57% ) TSYX A Stock I’m Watching $GOOGL ( ▲ 0.39% ) Topped out in May and has settled into it’s 50 day moving average. If this holds it could be buyable here. In Case You Missed It We invited Dan Ferris to the Financial Heat Podcast….. 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 Jun 5, 2026 • 11 min read Five Forces That Could Make You Rich---Or Leave You Behind Bain's new report says AI isn't working. They're not wrong. They're just measuring the wrong layer. Jun 2, 2026 • 1 min read The memory bottleneck Money follows the constraint Jun 2, 2026 • 9 min read NVIDIA Just Bet $6.5 Billion On This Jensen Huang committed $6.5 billion to a supply chain that does not yet exist at required scale — the optical interconnect layer that makes million-GPU AI factories physically possible. View more 6