광학(Optics) 트레이드의 2단계

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 Second Phase of the Optics Trade The Second Phase of the Optics Trade The market knows the ticker. It doesn't know the supply chain. That gap is where the next round of returns gets made. Jun 9, 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.      $1.3B Tower SiPho contract signed for 2027 Up to ~$3.2B NVIDIA–Corning warrant-linked equity; 10x optical capacity expansion $632M AXT raise to double InP capacity >$100M AXT InP backlog, Q1 2026 ~0% CPO attach rate today — early innings   THE CROWDED TRADE WITH THE THIN CROWD Everybody owns the optics theme now. Almost nobody owns the supply chain. That is the opportunity. The first phase of the trade was simple: buy the obvious optical names after NVIDIA told the market that copper's reach was becoming the bottleneck at scale. The second phase is harder. It requires tracing co-packaged optics from the first indium phosphide wafer to the external laser source, the photonic engine, the switch package, the fiber connector, and the final alignment test. That is where the next returns are hiding. CPO is not one product. It is a collision between two manufacturing chains: the compound-semiconductor chain that makes the light, and the silicon-photonics chain that carries the signal. Every place those chains meet is a yield gate. Every yield gate is a potential tollbooth. And most investors who own the names have never traced where those tollbooths are. That silence is the opportunity. This issue closes it.   "NVIDIA wasn't just buying lasers. It was locking up the ability to make them." — The $4B signal the market missed   A SYSTEM YOU CANNOT UNDERSTAND WITHOUT A MAP Co-packaged optics moves the optical interface from the front panel of the switch — where a pluggable transceiver module lives — and embeds it directly into the switch package, next to the ASIC itself. The benefit is power consumption and signal integrity. Moving light over fiber instead of electrical signal over copper inside the chassis cuts watts dramatically at the speeds now being deployed. But to make this work, two separate manufacturing chains have to converge in a single package, and each chain has its own critical materials, fabrication steps, and inspection gates. Stream A makes the light. It starts from a single-crystal indium phosphide ingot, progresses through epitaxial growth of quantum wells via MOCVD equipment, and ends with high-power continuous-wave laser dies that deliver 400mW and above — six to eleven times the output of conventional transceivers. These lasers are assembled into External Laser Source modules that sit outside the switch package because the heat environment near the ASIC would kill them. Stream B handles the signal. It starts from a silicon-on-insulator wafer, moves through a photonic IC foundry where waveguides and modulators are etched, and is completed when an electronic IC fabricated on a conventional CMOS process is bonded on top via hybrid bonding. The result is an optical engine. The two streams meet at packaging. Optical engines mount around the switch ASIC on a common substrate. Light is injected through polarization-maintaining fiber from the ELS. The completed assembly goes through system test and chassis integration, and the switch is done. What matters to investors is this: at every seam where Stream A meets Stream B — and there are multiple — light converts micron-level misalignment directly into signal loss. Every seam is a yield gate. Every yield gate is an inspection event. Every inspection event belongs to a test equipment supplier. The supply chain has more ways to monetize CPO than the market has yet priced.   THE BOTTLENECK MIGRATION THESIS: WHERE LIGHT GETS STUCK Not all positions in this supply chain are created equal. The Bottleneck Migration Thesis — the analytical framework we use throughout Tech & Capital — holds that in any infrastructure buildout, the market consistently undervalues whichever constraint is actually binding at any moment in time. By the time Wall Street discovers it, the scarcity premium is already embedded. In CPO, there are three binding constraints right now that the market has not fully priced. Constraint One: InP. A 400mW-class CW laser has a large die. Large die means more substrate per laser. And indium phosphide substrate supply is not a broad commodity market — it is a concentrated, qualification-heavy supplier base led by AXT/Tongmei and Sumitomo Electric, governed by proprietary crystal growth methods that take years to develop. New entrants cannot replicate that qualification history on a relevant timeline. AXT disclosed InP backlog crossing $100M, completed a $632.5 million raise to fund expansion at its Tongmei subsidiary, and is on a path from its $17M quarterly revenue peak to $35M by year-end 2026 and $65–70M by late 2027. For public-market investors, AXT is the cleanest liquid pure-play on this layer. There is no light without InP. Constraint Two: MOCVD Equipment. Between the InP substrate and the finished laser die sits an epitaxial growth step performed on MOCVD equipment. Aixtron is the cleanest public tollbooth on this expansion cycle. Its planetary reactor design dominates the InP batch epitaxy market, and the Lumentum G10-AsP orders disclosed on May 19, 2026 are the first concrete confirmation that high-power InP laser capacity is moving from roadmap to purchase order. Whichever laser fab decides to expand, that decision is likely to show up in Aixtron's order book before it shows up in earnings reports anywhere else in the supply chain. Constraint Three: 3D Stacking. Where the two streams converge, TSMC's COUPE process bonds the electronic IC and the photonic IC into a single package. This is currently the most concentrated single-vendor dependency in the supply chain. Industry sources estimate SoIC capacity tracking toward 30,000 to 40,000 wafers per month by year-end 2026, shared with non-CPO demand — a dynamic identical to how CoWoS created a bottleneck in the HBM era. The yield story here is not the headline engine yield NVIDIA emphasizes — it is the end-to-end yield product across COUPE, assembly, ELS alignment, and connector mating. Those are different numbers.   THE SIGNAL THE MARKET MISSED: MARCH 2, 2026 On March 2, 2026, NVIDIA disclosed a $2 billion investment in Coherent and a separate $2 billion investment in Lumentum, both on the same day, both with what the companies described as "multi-billion dollar purchase commitments" and "future capacity access rights." The market read this as NVIDIA buying laser supply. The correct read is different: NVIDIA was purchasing the right to produce lasers that have not yet been made, from capacity that does not yet exist, at volumes that will only make sense if CPO penetrates at the scale NVIDIA is planning. That is not procurement. That is infrastructure pre-commitment. The difference matters for every company that sits upstream of the laser in the supply chain.   THE ARCHITECTURE WARS ARE NOT OVER — AND THAT'S BULLISH One of the most common errors in CPO analysis is treating it as a single architecture converging on a single winner. The production track record of TSMC's COUPE process leads to this conclusion, but the conclusion is wrong. Tower Semiconductor disclosed a $1.3 billion SiPho revenue contract for 2027 delivery, with $290 million in prepayments already received and larger 2028 commitments on the way. Active SiPho customers exceed 50. That is not a market converging on the COUPE bundle — that is an open camp building its own foundation, and it is doing so with real customer money in hand. The modulation technology race adds a second dimension. TSMC is ramping 200G micro-ring modulators in 2026 with a 400G roadmap. Coherent demonstrated 400G-per-lane InP modulator arrays. UMC is pushing thin-film lithium niobate chiplet production with HyperLight. Lightwave Logic won a foundry process agreement on Tower's PH18 and got onto GlobalFoundries' design kit. What this means for the investment thesis is not confusion. It means that CPO's supply chain is not a winner-take-all architecture bet — it is a multi-camp structure where architecture-independent layers carry lower risk than engine-level plays. InP substrates, MOCVD equipment, and optical fiber carry their revenue regardless of which modulation technology or bundling architecture wins. That is the correct way to build a position during an unresolved architecture war: own the toll roads, not the competing travelers.   THE PLUGGABLE THAT WANTS TO BE BOTH Reading CPO as simply a pluggable killer misses the most interesting position in the market. Applied Optoelectronics disclosed its first volume 1.6T transceiver order from a major hyperscale customer on March 9, 2026. At OFC 2026, it demonstrated a 25dBm-class ELS module, challenging the Lumentum-Coherent-Sumitomo troika directly. This company has structured its business to win regardless of which architecture captures the next three years of switch port deployment. It is selling into the transceiver boom while simultaneously positioning for the light-source socket of the CPO era. This is the archetype for the second phase of the optics trade: not companies that are clearly all-in on one architecture, but companies that carry meaningful revenue in the current regime while owning a credible claim on the next one.   WINNERS & LOSERS: THE CPO SUPPLY CHAIN SCORECARD   TIER NAME / TICKER SUPPLY CHAIN POSITION WHY IT MATTERS NOW WINNER — TIER 1 AXT Inc. (AXTI) Concentrated, qualification-heavy InP supplier base; AXT/Tongmei is the cleanest liquid public pure-play; backlog >$100M $632M raise; capacity doubles 2026, doubles again 2027. No light without InP. Watch: Tongmei execution and STAR Market approval timeline. WINNER — TIER 1 Aixtron (AIXA.DE) MOCVD equipment; dominant in InP epi batch reactors Cleanest public tollbooth on the InP epitaxy expansion cycle. Lumentum's multiple G10-AsP orders (May 2026) are the first visible confirmation that high-power InP laser capacity is moving from roadmap to purchase order. WINNER — TIER 1 Tower Semiconductor (TSEM) Open SiPho PIC foundry; outside COUPE bundle $1.3B SiPho contract signed for 2027 + $290M prepayment already received. $2.8B revenue target for 2028. WINNER — TIER 1 Corning (GLW) Optical fiber & connectors; CPO raises fiber content per rack NVIDIA purchased warrants (aggregate $500M) giving up to ~$3.2B equity exposure alongside a confirmed 10x optical-connectivity capacity expansion and 50%+ U.S. fiber production increase. Warrant-linked conviction signal, not simple procurement. WINNER — TIER 2 Lumentum (LITE) ELS laser dominant; high-power CW moat Every CPO port creates new ELS demand. Listed in NVIDIA SiPho ecosystem; demonstrated 800mW laser at OFC 2026. Greenfield NC fab in progress. CPO cannibalization of pluggable offset by ELS upside. WINNER — TIER 2 Coherent (COHR) Vertically integrated: InP lasers, SiPho, modulators, CPO architectures NVIDIA $2B investment same day as LITE. Demonstrated 6.4T socketed CPO and 400G/lane InP modulator arrays at OFC 2026. Breadth of vertical integration is a hedge against modulation technology uncertainty. WINNER — TIER 2 Onto Innovation (ONTO) Test / metrology / alignment inspection axis Light is analog. Electrical pass/fail logic does not apply — every seam carries a continuous optical loss measurement. Test intensity is structurally higher per device in CPO than in pure electrical chips of equivalent complexity. WATCH — DUAL EXPOSURE AAOI (Applied Optoelectronics) Pluggable 1.6T + ELS challenger First 1.6T volume order March 2026 AND ELSFP at 25dBm at OFC 2026. Meaningful exposure to both current and next-gen architectures; credible foot in both doors, not a guaranteed CPO winner. PRESSURE POINT Front-Panel-Only Optics QSFP/OSFP revenue; no ELS, NPO, or CPO roadmap Pluggables broadly are not dying. CPO, LPO, NPO, and pluggables coexist across speeds and distances. The pressure falls on businesses with no credible path into ELS sockets, near-packaged optics, or the manufacturing and test layers CPO requires. DELAY MECHANISM LPO/LRO Approach Extends pluggable economics at 800G/1.6T by removing DSP If LPO works well at 1.6T, CPO attach rate slows. Timing risk, not structural defeat. CPO's power and density advantages at 3.2T+ are independent of whether DSP removal succeeds at current speeds.   PRESSURE POINTS: THE BINDING CONSTRAINTS DASHBOARD   CONSTRAINT CURRENT STATE BOTTLENECK OWNER WATCH SIGNAL InP substrate supply Backlog >$100M; 6–9 month lead times AXT / Sumitomo Electric; concentrated supplier base with high qualification barriers AXT quarterly revenue > $17M (new record) MOCVD equipment delivery 18–24 month order lead times historically Aixtron near-monopoly on batch InP reactors Aixtron order book inflection / ASP uptick 3D stacking / COUPE capacity TSMC SoIC targeting 30–40K wfr/mo by end-2026 TSMC / SPIL (ASE Group); near single-vendor SoIC capacity milestones + CoWoS share trade-off High-power CW laser volume NVIDIA locked $2B each in LITE & COHR March 2026 Lumentum (dominant) + Coherent + Sumitomo ELS module attach rate per CPO unit shipped CPO attach rate (% of new ports) Starting from sub-single-digit % today NVIDIA Spectrum-X platform deployment pace Hyperscaler capex announcements citing CPO   CREDIBILITY FIREWALL   HOW TO READ THIS SECTION The Credibility Firewall is a standing feature of Tech & Capital. Its purpose is to protect you from the most common failure mode in financial analysis: the drift from sourced fact to confident inference with no signpost in between. The left column below contains only what has been publicly disclosed and confirmed. The right column contains directional inferences drawn from that information — the author's analytical work product. These are different things and must be treated as such.   SOURCED / CONFIRMED DIRECTIONAL INFERENCE • NVIDIA invested $2B each in Coherent and Lumentum (March 2, 2026) with multi-billion dollar purchase commitments and future capacity access rights (NVIDIA Newsroom) • AXT completed $632.5M raise to fund InP expansion at Tongmei; InP backlog exceeded $100M as of Q1 2026 (AXT investor release) • Tower Semiconductor signed $1.3B SiPho contract for 2027 delivery; $290M prepayment already received; active SiPho customers exceed 50 (Tower press release) • NVIDIA purchased Corning warrants for aggregate $500M; Corning to expand U.S. optical-connectivity manufacturing capacity 10x and U.S. fiber production capacity by 50%+; three new facilities; 3,000+ jobs (NVIDIA Newsroom, Corning 8-K) • Ayar Labs closed $500M Series E at $3.75B valuation with NVIDIA and AMD participation (March 2026); joined NVLink Fusion ecosystem (June 3, 2026) • Marvell acquired Celestial AI at $3.25B initial consideration (February 2026); received $2B NVIDIA investment and joined NVLink Fusion (March 2026) • Lumentum placed multiple G10-AsP MOCVD orders with Aixtron (Aixtron investor release, May 2026) • TSMC confirmed 2026 as COUPE volume production year (TSMC 2025 Technology Symposium; TrendForce reporting) • AAOI received first volume 1.6T transceiver order from major hyperscale customer (March 9, 2026); demonstrated 25dBm ELSFP at OFC 2026 (AOI investor releases) • Coherent demonstrated 6.4T socketed CPO and 400G/lane InP modulator arrays at OFC 2026 (Coherent press release) • CPO attach rate inflects from sub-single-digit % to meaningful share of new datacenter switch ports in 2026–2027 (thesis; no confirmed hyperscaler deployment data at this stage) • Aixtron holds dominant share in InP batch MOCVD reactors with limited credible second sources on relevant timelines — directionally supported by industry commentary; specific share figures are sell-side / private research estimates, not Aixtron-disclosed • TSMC SoIC capacity on path to 30,000–40,000 wafers/month by end-2026 — reported by industry sources and sell-side; not disclosed directly by TSMC • ELS field-replacement creates recurring revenue at scale — directional; depends on CPO reliability data supporting field-swap economics over service life • LPO/LRO is a delay mechanism, not a CPO structural defeat — inference; outcome depends on whether DSP removal meets power budgets at 1.6T+ without unacceptable yield trade-offs • Open SiPho camp (Tower / GlobalFoundries) and COUPE bundle coexist rather than COUPE winning all volume — plausible multi-camp structure supported by Tower contract data; hyperscaler procurement strategy not confirmed • InP supply base is effectively AXT and Sumitomo for public-market purposes; JX Advanced Metals and several other suppliers exist but are less prominent in CPO-grade substrate supply — directional; qualification status of other suppliers not publicly detailed   BEAR CASE: WHAT HAS TO GO WRONG FOR THIS THESIS TO FAIL CPO Penetration Stalls at Single Digits. LPO/LRO solves the power problem at 1.6T without the complexity of CPO integration. Hyperscalers defer CPO until 3.2T or beyond. Attach rate stays sub-5% for two to three more years. InP demand growth is real but the magnitude is a fraction of what the CPO thesis implies, and AXT's capacity expansion arrives into a softer-than-expected market. COUPE Monopoly Becomes COUPE Bottleneck. TSMC's SoIC capacity, shared with CoWoS and other advanced packaging demand, cannot scale fast enough. The CPO supply chain expands InP and laser production only to find the convergence point is the new choke. This is a timing risk, not a structural defeat — but it compresses the 2026–2027 revenue window for the layers above. Architecture War Goes Longer Than Expected. The modulation technology race — micro-ring vs. TFLN vs. EO polymer — remains unresolved through 2027. Design cycles lengthen. Engine-layer companies spend capital on parallel development. Foundry commitments are placed but utilization is split across competing platforms that do not yet have proven yield. InP Capacity Expands Faster Than CPO Attach Rate. AXT and others may execute their expansion programs correctly, but if CPO deployment slips, the market may discover that the first bottleneck was temporary rather than structural. The revenue still arrives — but the scarcity multiple compresses before the new capacity is absorbed. For AXTI specifically, this is the cleanest single bear case: execution without the premium. China/Tongmei Execution Risk. AXT is the cleanest liquid public expression of the InP substrate thesis, but the capacity expansion is tied to Tongmei and Chinese manufacturing. Export permits, STAR Market approval timeline, customer qualification milestones, and governance structure between AXT and Tongmei all represent execution variables that are not present in Sumitomo's Japan-based substrate production. Investors should weigh these asymmetries when sizing a position in AXTI versus thinking about InP exposure more broadly. The architecture-independent layers (InP supply base, MOCVD equipment, fiber) carry the lowest exposure to all five scenarios above. Engine-layer and COUPE-bundled positions carry the highest.   FIVE TAKEAWAYS   1. The CPO market is not a single architecture converging on a single winner. Two camps — the COUPE bundle and the open SiPho camp — are building simultaneously with real customer prepayments. Own the architecture-independent layers. 2. InP is the bottleneck nobody talks about. No InP, no laser. No laser, no CPO. The substrate supply base is concentrated, qualification-heavy, and years-long to expand — and the backlog is already over $100M. AXT's $632M raise is one of the most important capital deployment decisions in the AI infrastructure supply chain right now. Note the execution variable: the capacity expansion runs through Tongmei in China. 3. Aixtron is the cleanest public tollbooth on the InP expansion highway. Its planetary reactor design dominates the InP batch epitaxy market, and the Lumentum G10-AsP orders in May 2026 are the first visible proof that high-power laser capacity is moving from roadmap to purchase order. Equipment orders are the leading indicator; revenue follows. 4. The NVIDIA–Corning and NVIDIA–Lumentum/Coherent capital commitments are demand signals, not supplier relationships. When a platform owner pre-commits billions to secure capacity that does not yet exist, that is the market itself telling you where the constraint is. Follow the pre-commitment capital, not the current revenue. 5. The Bottleneck Migration Thesis predicts that constraints rotate as each is relieved. InP and MOCVD are today's binding constraints. If the expansion programs at AXT and Lumentum execute on schedule, the constraint migrates to COUPE packaging capacity by 2027–2028. The companies positioned at both current and next-generation bottlenecks capture the premium in both legs of the trade. That is the "second phase" playbook.   CATALYST WATCHLIST   WHEN WHO WHAT TO WATCH Q2 2026 Earnings AXT (AXTI) InP revenue run rate vs. $17M historical peak; Tongmei capacity milestones Q2 2026 Earnings Tower (TSEM) SiPho customer count vs. 50+; 2027 prepayment schedule update Q2 2026 Earnings Aixtron (AIXA.DE) G10-AsP order book growth; backlog composition shift toward InP tools Q2 2026 Earnings Lumentum (LITE) ELS revenue line emergence; NC fab groundbreaking timeline Q3 2026 TSMC SoIC monthly capacity print; COUPE yield metrics in earnings materials Q3/Q4 2026 Hyperscalers Capex guidance language citing optical interconnect or CPO specifically End-2026 NVIDIA Spectrum-X deployment volumes and CPO attach rate implied by ELS pull Jan 2027 Tower (TSEM) 2028 prepayment receipt ($290M tranche due); signals whether demand commitment holds Mid-2028 Lumentum NC InP fab operational — supply-side test of whether domestic InP capacity arrives on schedule   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 So far looking like day 2 of the recovery. Some market thoughts from me and the team…… Yesterday’s rally was a bit misleading, decliners led advancers within the S&P 500 at a pace of 1.8 to 1. Investors also seemed underwhelmed by Apple’s AI pivot for Siri…. With everything else going on in the market it’s hard for Apple to stay really relevant anymore. In other Mag 7 news, headlines hit that Meta was considering a capital raise on the heels of Google’s announcement last week. With OpenAI filing for an IPO you would expect the other hyperscalers to join in on capital raises before SpaceX, OpenAI, and Anthropic hoover up all the investor attention. For the bulls…… Jim Cramer warns key pillars of the bull market are beginning to crumble CNBC's Jim Cramer said that he's becoming more cautious on stocks after several pillars of his bullish outlook have come under pressure. CNBC • Alexa LoMonaco 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 Tuttle Capital Launches “Permanent Portfolio”-Focused, Multi-Asset ETF Tuttle Capital has partnered with Porter & Company to launch a new multi-asset ETF that may appeal as a balanced overall holding. ETF Trends • Nick Peters-Golden A Stock I’m Watching $AMBA ( ▼ 4.32% ) At the moment there is no pure play robotics stock, so those looking for exposure have to play around the edges. AMBA is one such name and it had an undercut and rally at the 50 day yesterday. 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 3, 2026 • 14 min read The $1M Missile Has A $5 Problem Capability used to be the moat. Cost-per-kill at scale is becoming the moat. The Pentagon just made that math explicit at SOF Week 2026 — and three public companies are sitting directly in the procurement stream it just created. Jun 2, 2026 • 1 min read The memory bottleneck Money follows the constraint Jun 8, 2026 • 1 min read Bonds failed. Stocks are a rollercoaster. Now what? The old playbook is dead. View more 3