AI 트레이드 내부에 숨겨진 100초 병목현상

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 100-Second Bottleneck Hiding Inside the AI Trade The 100-Second Bottleneck Hiding Inside the AI Trade NVIDIA has lined up roughly $6 billion to lock up the optics supply chain. None of it ships until somebody tests it. May 13, 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 $4 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 Disclosure Day: A Playbook For Investors If The Go … H.E.A.T. News vs. Noise: What’s Moving Markets Today ETF News In Case You Missed It Our next webinar will be May 14th at 2PM ET. Disclosure Day: A Playbook For Investors If The Government Confirms It Has Alien Technology Click HERE To Register Live Webinar from Matt Tuttle On February 19, President Trump started the countdown to potentially the biggest government disclosure in history: www.whatifdisclosure.com/webinar/?utm_source=tuttle&utm_medium=general&utm_campaign=mayufodwebinar H.E.A.T. The next chokepoint in the AI buildout is not the chip — it is the equipment that proves the chip works. Six public stocks, one geopolitical wildcard, and a market that has priced this as a trade — not yet as a multi-year bottleneck rent. THE SETUP Every AI cycle has a bottleneck, and the bottleneck moves. First it was compute. Nvidia took that crown. Then it was memory — HBM, the part of the supply chain SK Hynix and Samsung quietly own. Then it was the interconnect — the optical bridge between thousands of GPUs that turns a server room into a single computer. That is where the money has been flowing for the last twelve months: $2 billion to Lumentum, $2 billion to Coherent, $2 billion to Marvell. Roughly six billion dollars from one company lined up to lock down the laser, the optical engine, and the silicon photonics roadmap. And here is the part the market is still underestimating. None of those photonic chips ship without passing test. The bottleneck is not the optics. It is testing the optics. Two definitions before we go further: CPO — co-packaged optics — puts the optical engine directly next to the GPU package instead of cabling it across the room. It is the only way to feed power-hungry AI chips enough bandwidth without melting the building. PIC — photonic integrated circuit — is the photonic chip at the heart of that engine, and it must be aligned and validated at tolerances measured in tens of nanometers. TSMC's COUPE platform — the production version of all of this — enters volume in 2026. A single CPO photonic integrated circuit takes over 100 seconds to fully inspect. That is roughly an order of magnitude longer than testing a conventional logic chip. It requires alignment tolerances measured in tens of nanometers, simultaneous electrical and optical probing, and — in the new TSMC stacked architecture — probing from both sides of the wafer at once. Existing semiconductor test equipment cannot do it. The big front-end equipment names — Applied Materials, Lam Research, KLA — do not play here. This is a different stack, owned by a different group of companies. Seven of them, to be precise. The market has noticed enough to push their stocks up triple-digit and quadruple-digit percentages over the last twelve months — but it has priced this as a trade, not as multi-year bottleneck rent. Six are publicly traded, market caps spanning from $2.9 billion to $128 billion, a 44x range for companies playing in the same niche. That kind of dispersion is where re-ratings hide. And here is the line for anyone who already owns Nvidia and is wondering why this matters. Testing is the gate between AI capex and AI revenue recognition. If the test floor cannot keep up, shipments slip — and the bottleneck rent accrues to whoever owns the test floor. That is the portfolio logic. The compute is yours. The gate is somewhere else. "100 seconds per chip. The fab can build them faster than the test floor can prove they work. That is the new bottleneck, and it has a stock list." BY THE NUMBERS 100+ sec Full optical test per CPO chip $6B NVIDIA optics-supply commitments +1,033% Top performer 1Y stock return Sources: TrendForce industry reporting; company filings (NVIDIA, Lumentum, Coherent, Marvell strategic investments); Yahoo Finance/Google Finance closing prices, May 6, 2026. THE THESIS: THE BOTTLENECK HAS MIGRATED AGAIN If you have been reading Tech & Capital for any length of time you know the bottleneck framework. In any major structural buildout, capital does not get rewarded uniformly. It concentrates wherever the constraint sits at that moment — and the constraint moves. In semiconductors, the constraint has migrated from logic to memory to packaging to interconnect. We are now watching it migrate one step further: from building the optical engine to validating the optical engine at production speed. Why this is different from prior bottlenecks: testing has historically been a low-glamour line item. A few percent of the equipment bill of materials. Conventional digital chips test in milliseconds, on automated probe stations, with margin to spare. The economics of CPO break that model. Run the throughput math at the monthly volume implied by hyperscale deployment — one million CPO units per month — and the equipment requirement gets brutal: • 1,000,000 units × 100 seconds = 100 million seconds of total monthly test time, or roughly 27,800 equipment-hours. • A single test cell at 85% uptime delivers about 612 hours per month, which translates to roughly 22,000 units per cell per month at the 100-second inspection rate. • Result: roughly 45 test cells running flat out to keep up — before counting retests, multiple insertion stages, and multi-engine architectures. The actual equipment requirement is materially higher. Each cell — combining burn-in, probing, ATE, optical alignment, and automation — costs millions. That is the punchline: the fab can build them faster than the test floor can prove they work. That is the bottleneck rent. That is not a small market hidden inside a bigger one. That is a market being created in real time, and it does not have an incumbent to defend it. Applied Materials, Lam, and KLA — the $200-to-$370 billion front-end equipment titans — are not the answer. The tools they build do not solve this problem. The companies that solve it are smaller, more specialized, and trading at a fraction of the front-end multiple, even after the run-up. THE FOUR-STAGE TEST STACK CPO does not get tested once. It gets tested four times. Plus a burn-in step before the rest. Each stage is a different problem. Each problem has its own equipment leaders. Mapping the stack is how you tell the pure-plays from the optionality from the noise. Stage What gets tested Why it is hard Burn-In InP lasers, before anything else Screens out infant mortality. Lasers fail early or run for a decade — the whole point is to find the ones that fail before they ship inside a $40,000 GPU module. Insertion 1 PIC wafer-level test Probe each photonic chip on the wafer, simultaneously driving electrical signals and coupling laser light through nanometer-precise alignment. Insertion 2 EIC + PIC combined wafer test After the electronic chip is bonded to the photonic chip, you have to test both layers together — and in stacked architectures, that means probing top and bottom of the wafer at the same time. Insertion 3 Optical engine assembly test After the engine is assembled with fiber arrays, test the full optical link. This is where assembly precision and test converge — and where the equipment competition is fiercest. Insertion 4 Module / system test Final validation of the full CPO module before it ships into a switch or GPU tray. Mostly an electrical-test problem with optical signal-integrity overlays. SPOTLIGHT: THE GEOPOLITICAL WILDCARD One of these companies is German. Its parent is Chinese. ficonTEC — the private German company that dominates Insertion 3 optical engine assembly and test — is a subsidiary of China-based Robo Technik. In any other moment, that would be a footnote. In 2026, with US-China technology decoupling extending deeper into the semiconductor equipment supply chain, it is not a footnote. It is a vendor-selection variable. Hyperscale customers building U.S.-government-adjacent AI capacity have a clear incentive to dual-source — and an even clearer incentive to avoid single-supplier exposure to a Chinese-owned entity at a chokepoint stage. This is the underpriced variable in the trade. The technical merits of ficonTEC versus competitors matter less than the structural pressure to qualify a non-Chinese alternative. That pressure flows directly to the public companies that occupy the same insertion. THE INVESTABLE SETUP Six public names occupy the four-stage stack, each with a different exposure profile. The way to play this is not to pick the winner — it is to build a basket weighted toward the companies whose insertion has the highest throughput pressure and the smallest existing revenue base. That is where the operating leverage lives. WINNERS — THE FOUR-STAGE TEST STACK Company Ticker Stage Mkt Cap The Setup Aehr Test Systems AEHR Burn-In $2.9B Smallest pure-play in the basket. $41M record order from a hyperscale AI customer in April 2026 — first proof point that AI is now a meaningful revenue line, not just a hope. Highest beta to volume ramp. FormFactor FORM Insertion 1 (PIC wafer) $11.6B The cleanest pure-play on photonic wafer probing. Q1 2026 revenue +32% YoY. Raised CPO revenue guide to the high end of the $10-20M range — small in absolute terms, large as a leading indicator. NVIDIA disclosed at >10% of revenue for the first time. Teradyne TER Insertion 4 (system) $56B Q1 2026 revenue +87% YoY, with AI revenue now ~70% of the total. Less of a CPO-test pure play, more of a backbone of system-level AI test. Photon 100 product positions them for the optical migration. Keysight KEYS Cross-stage instrumentation $62B The instrumentation layer that sits underneath every other company on this list. Optical test gear, signal analyzers, the picks-and-shovels play within picks-and-shovels. Lower beta, higher floor. Advantest 6857.T Insertion 2/4 (combined + system) $128B Largest market cap in the group. Already the dominant ATE name for HBM. Will own a meaningful slice of CPO module test by virtue of incumbency. Less explosive — more durable. Chroma ATE 2360.TW Insertion 3 (optical engine) ~$30B Taiwanese listing. Direct competitor to ficonTEC at Insertion 3 — and the most obvious beneficiary if hyperscale customers decide to dual-source away from Chinese-owned ficonTEC. The geopolitical asymmetry stock. Note on ficonTEC: private, German-headquartered, Chinese-owned through parent Robo Technik. Not investable directly. Its competitive position is the wildcard variable inside the public names — specifically Chroma — rather than a stock to own. PRESSURE POINTS — THE OTHER SIDE OF THE TRADE Bottleneck migrations create losers as well as winners. As the constraint moves from logic to memory to optics to optical test, the names whose narrative was tied to the old constraint face multiple compression even if absolute revenue holds up. Pressure Point The Issue Front-end equipment titans Applied Materials, Lam, KLA. Not losing money, not losing share in their own segments. But their tooling does not solve the optical-test problem, and the next leg of AI capex growth flows past them. Risk is multiple compression as the narrative migrates. Pluggable-optics legacy plays CPO is, in the limit, a replacement for pluggable transceivers in hyperscale switches. Companies whose AI exposure is concentrated in pluggable-module shipments rather than co-packaged architectures face a longer-dated headwind. Single-axis CPO bets Any name being marketed as a "CPO play" but only exposed to one of the four insertion stages. Throughput compresses, recipes change, and single-stage exposure is fragile. The basket beats the single-name bet. CREDIBILITY FIREWALL: CONFIRMED VS. DIRECTIONAL Confirmed Directional 100+ second per-PIC test time figure (TrendForce industry reporting). Throughput math implying 45+ cells per million-chip-month — derived calculation, not an industry-published figure. NVIDIA strategic investments: $2B Lumentum, $2B Coherent, $2B Marvell (per company filings/disclosures). Conversion timing of those $6B commitments into actual production volume — uncertain, sentiment-driven. FormFactor Q1 2026: revenue $226.1M (+32% YoY), CPO guide raised to high end of $10-20M (per Q1 2026 earnings). Whether CPO test scales into a multi-billion-dollar annual market comparable to front-end equipment — open question, dependent on COUPE volume ramp. Aehr $41M order from hyperscale AI customer (April 2026 disclosure). Whether the geopolitical pressure to displace Chinese-owned ficonTEC translates into measurable share gains for Chroma — directional read on dual-sourcing behavior. Teradyne Q1 2026: revenue $1.282B (+87% YoY), AI ~70% of total. TSMC COUPE entering production in 2026 per public roadmap — schedule risk is non-zero. WHY NOW: FOUR CHECKPOINTS TO WATCH The thesis is structural, but it re-rates around discrete catalysts. These are the prints and disclosures that, if they break the right way over the next two to three quarters, force the Street to mark the basket higher. The next 90-180 days TSMC COUPE volume ramp confirmation. Any earnings-call language from TSMC or its supply chain partners on actual CPO unit volumes pulls every name in this basket. Slippage cuts the other way. FormFactor's CPO revenue trajectory. The current $10-20M guide is a leading indicator. If the company raises again — particularly if they put a dollar figure on 2027 — the multiple expands. Aehr's order book breadth. April's $41M order was from one hyperscale customer. The question is whether others follow. Each new disclosed customer is a step-change in the bull case. Hyperscale dual-sourcing announcements. Any disclosure from a U.S. hyperscaler explicitly qualifying a non-Chinese alternative at Insertion 3 is a direct read-through to Chroma. This is the geopolitical asymmetry trade — silent until it isn't. And the structural moat to remember: throughput is destiny in this market. Optical test does not scale by adding more wafers. It scales by adding more cells, more floor space, more capex. The company that compresses 100 seconds to 50 wins not because it is better — but because it is the only way to keep up with COUPE volume. That is the bottleneck-within-the-bottleneck, and it is why the equipment vendors that solve it earn a premium. THE BEAR CASE What breaks this thesis COUPE slips. TSMC's production roadmap is the keystone. A six-month delay in volume ramp pushes equipment capex right with it, and these stocks — already up triple-digit percentages — are not priced for slippage. Test time gets compressed faster than expected. Counterintuitive, but real: if FormFactor or ficonTEC crack the throughput problem in 2026 rather than 2028, the equipment-cells-needed math collapses. Fewer cells means lower capex, even if volume ramps. Pluggable optics holds its own. CPO is the consensus future, but pluggable transceivers are not dead. If the cost-per-bit gap between CPO and 1.6T pluggables stays narrow, hyperscalers may delay CPO adoption at the margins. Geopolitical wildcard cuts the other way. If U.S.-China tech tensions ease — unlikely but possible — the dual-sourcing premium for non-Chinese alternatives compresses, and Chroma's geopolitical asymmetry fades. Already-priced risk. Aehr +1,033%, Chroma +757%, FormFactor +468% over twelve months. The basket has run. These are not value stocks — they are bottleneck rents. Time your entries. Buying on a pullback is materially different risk-reward than chasing recent highs. FIVE TAKEAWAYS 1. The AI bottleneck has migrated again — from compute to memory to optics to optical test. The companies that solve the testing problem are smaller, more specialized, and not the same names that owned prior bottlenecks. Front-end equipment titans (AMAT, Lam, KLA) do not play in this market. 2. 100 seconds per chip is the number that defines the trade. It dictates how many test cells are needed per million units of CPO production — and the math implies a step-function increase in equipment capex that has not yet shown up in consensus revenue estimates for the test names. 3. Six public stocks own the four-stage test stack: Aehr (burn-in), FormFactor (Insertion 1), Advantest (Insertions 2 and 4), Teradyne (Insertion 4), Chroma (Insertion 3), Keysight (cross-stage). Market caps span $2.9B to $128B — a 44x dispersion that implies very different beta to the same underlying volume ramp. 4. The geopolitical wildcard is real and underpriced. ficonTEC — the dominant private player at Insertion 3 — is Chinese-owned. The pressure to dual-source flows directly to Chroma. This is the asymmetry inside the basket that the market has not paid for yet. 5. Six billion dollars of NVIDIA optics-supply commitments are the leading indicator. Every one of those Lumentum, Coherent, and Marvell-supplied components passes through the four-stage test stack before it ships. The capital is committed. The volume is coming. The test floor is the gate. Every AI dollar eventually has to walk through a test cell. The basket that owns those cells is the next leg of the trade. 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 A couple of troubling developments yesterday: CPI came in hotter than expected. What the data also showed is that consumers are pulling back on other purchases to buy gas. That shouldn’t surprise anyone, but it probably does. I continue to think the average consumer is in rough shape, regardless of what manipulated data says, we also tend to stay away from consumer stocks. Why do I say “manipulated”? For example, the report showed gasoline up 5.4% month over month and up 30% since January. IMHO this greatly understates gas prices, I think they are up more than 50% since January. I could go on, but don’t want to bore you. South Korean policy chief Kim Yong-beom floated the idea of a “Citizens Dividend” that would be funded by excess tax revenue from AI. Market’s didn’t seem to like the word “excess” here. Whenever anyone in government starts calling your profits “excess” you have a notion of what could come next. Market’s sold off but then recovered most of the losses into the close. My sense continues to be that traders believe higher oil prices and higher inflation are temporary and the AI buildout is a long term structural shift unlike anything we have seen since at least the 1990’s. Or investors just have FOMO and buy every dip with both hands. So far this morning the market is retracing yesterday’s losses. Meanwhile, Trump is in China. Tensions, especially over tariffs, have taken a back seat to Iran. I’ve been thinking Trump wants some good news to change the news cycle, which could be a positive for Chinese stocks. So far this morning they are selling off a bit. I’ll give it a little rope, but this is such a target rich environment if you are wrong on something you cut it and move on. 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 $MEMY Holdings Update: We replaced $FNV ( ▲ 1.6% ) and added $WOLF ( ▲ 16.53% ) All 5% positions. For a full list of MEMY holdings, visit: https://incomeblastetfs.com/etf/memy Distributor: Foreside Fund Services, LLC A Stock I’m Watching $WOLF ( ▲ 16.53% ) There are so many stocks to watch that I often struggle with what name to put here. Should have done this one yesterday as it looks to be up big pre market. Assuming that’s real, I wouldn’t chase, but keep this on your radar screen for a possible dip buy. The company emerged from Chapter 11 last summer with bondholders owning the equity free-and-clear of the debt that bankrupted the old company, and they inherited the only 200mm silicon carbide fab in the world plus the largest SiC wafer facility ever built — combined replacement cost north of $6.5 billion against today's $2.2 billion market cap. In other words, you're buying irreplaceable physical infrastructure for roughly a third of what it cost to build. Here's the kicker: Nvidia's new 800-volt rack architecture for next-gen AI GPUs runs on the exact same wide-bandgap silicon carbide supply chain that was originally scaled for electric vehicles. The capacity that bankrupted Wolfspeed when EV demand disappointed now has a second customer cohort — AI data centers — walking in the door at exactly the moment the first one wobbled. The Street is still pricing WOLF as a busted EV story; we think it gets repriced as the only domestic pure-play in a critical AI infrastructure bottleneck, with 300mm SiC capability (Wolfspeed is one of only two companies on Earth that have demonstrated it) sitting on top as a free option. 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. 1 Keep Reading May 7, 2026 • 14 min read AI's Last Tactical Mile Anduril and Nokia just announced something Wall Street misread. The investable trade isn't the names on the announcement. May 12, 2026 • 13 min read The Convergence Trade That Everyone Is Missing Defense, telecom, cloud, and robotics just collapsed into one stack. Wall Street still prices them as four Apr 23, 2026 • 10 min read Beijing Is Not Blinking View more 2