Second Order Research

Second Order Research is now Hayabusa Research. Thank you to everyone who has followed our work here. We’re continuing at @hayabusares, with the same commitment to independent research, disciplined entry prices, and conclusions we’re willing to revisit. Hayabusa means “peregrine falcon” in Japanese and names Japan’s asteroid sample-return missions. For us, it represents perspective, precision and the discipline to return with evidence. Follow @hayabusares for our next chapter. Curiosity Before Capital. Return with evidence.

EMCOR Group ($EME) and Comfort Systems USA ($FIX): Where I’d Buy AI’s Labor Bottleneck The AI buildout is intensifying a skilled-labor shortage. Every new data center needs electricians, cooling specialists and crews capable of connecting it to reliable power. Those workers take years to train. AGC’s September workforce survey found that data-center demand is keeping construction labor tight even as other parts of the market soften. For investors, the opportunity depends on who keeps the profits. Higher wages can reward workers while squeezing contractors that committed to deliver projects at a fixed price. I’m looking for businesses that can improve productivity, protect contract margins and finish work reliably. After comparing four candidates, my ranking for new money is EMCOR, Comfort Systems, Quanta Services ($PWR), then IES Holdings ($IESC). The entry price changes how interested I am. EMCOR comes first because its valuation asks less of future earnings. At the September 16 close of $739.46, it traded around 20 times published 2027 EPS estimates of $36.75. Applying our assumed 23× multiple produces approximately $845, or 14% upside over a 12–18 month horizon. That is enough to keep researching, but I would prefer an entry around $660–700, where the same scenario offers roughly 21–28%. My main concern is cash conversion. First-half operating cash flow lagged net income substantially. I want to see collections improve as projects progress, alongside evidence that margins remain healthy. Comfort Systems is the stronger growth alternative. Modular construction represented 17% of first-half revenue, giving it a practical way to move some work from crowded job sites into factories. That could improve scheduling and labor productivity. The investment still depends on what those capabilities earn. Published 2027 EPS estimates are approximately $60. At September 16’s $1,587.35 close, FIX traded around 26 times that estimate. Our $60 EPS × 30 valuation scenario produces $1,800, about 13% upside. I become more interested around $1,400–1,500. The condition is that project margins hold up and cash generation remains sound after customer billing advances normalize. Quanta has valuable grid-construction capabilities, but its $619.28 share price already represents approximately 31 times published 2027 adjusted EPS estimates. Even our 32× middle scenario produces only about $631. I would revisit it around $500–525. Fixed-price work accounted for nearly 64% of second-quarter revenue, so rising labor costs deserve close attention. IES offers substantial capacity-expansion potential, with more uncertainty around acquisition integration and sustainable margins. Our own normalized fiscal-2027 scenario uses $12 EPS and a 28× multiple, producing $336 against a $325.02 reference price. My review range is $260–280, contingent on clearer financing and operating evidence. Its estimate is ours, not verified consensus, and deserves less confidence. These entry ranges are conditional valuation judgments, not technical support levels. In our downside scenarios, losses from the reference prices range from approximately 27% for EMCOR to 49% for IES. Those are not maximum possible losses. Timing also argues for patience. All four were below their calculated 50-day averages, and EMCOR was below its 200-day average. I would want selling to stabilize before treating a lower price as an opportunity. We checked expectations and options too. No usable numerical whisper was verified for the two leaders, and their attempted options quotes were incomplete. Neither check supplied evidence to justify rushing an entry. My preference is EMCOR for the balance of valuation and earnings risk, with Comfort Systems close behind for growth. I would reassess that order if their prices or operating evidence change. AI’s demand for skilled labor gives us a durable research question. The investment becomes actionable when the business can retain the economics and the share price leaves room for disappointment. Reference prices: September 16, 2026 closes. Scenarios exclude dividends, fees and taxes. Earnings definitions differ across companies. Curiosity Before Capital. Disclosure: Research is for informational and educational purposes, not personalized investment advice. Investing involves risk, including loss of capital.

🚨 Insider Buying + Political Trade Alert 🚨 $AMRZ just fell BELOW the price its own CEO paid for shares. Rep. Ro Khanna’s family education trust disclosed buying $1K–$15K of Amrize Ltd ($AMRZ) on Aug. 7. That caught my attention, but the political trade itself is not the reason this setup matters. The much bigger signal is what happened next. Since the selloff, Amrize executives and directors have been buying aggressively in the open market. CEO Jan Jenisch bought another 30,000 shares on Sept. 2 for roughly $1.27M at an average price of about $42.47. $AMRZ closed around $41.96. In other words, the market is now offering shares below the CEO’s latest purchase price. And this is not an isolated CEO buy. Multiple executives across finance, strategy, legal, technology, supply chain and operations have bought shares. One current Form 4 dataset shows 13 distinct insiders, 32 open-market purchases, roughly $16.2M purchased and zero open-market sales. That is the signal I care about. The political disclosure was the breadcrumb. The real question is why so many people who actually run the business are putting their own money into the stock while the market continues to sell it. So why is $AMRZ under pressure? Because there is a real fundamental problem, and the market is not imagining it. Q2 revenue grew 8.6% and adjusted EBITDA grew 5.8%, but EBITDA margin fell from 29.0% to 28.2%. Higher freight, diesel and raw-material costs are squeezing profitability, and elevated energy prices could keep that pressure alive. Management cut its 2026 EBITDA outlook to $3.1B–$3.2B while raising revenue guidance to $12.5B–$12.7B. That creates a very interesting disagreement. The market is effectively saying: “Demand may be fine, but inflation is going to eat too much of the economics.” The insiders appear to be saying: “You’re discounting too much of that risk at this price.” That disagreement is where the setup gets interesting. There’s also a second-order angle here that I don’t think should be ignored. Amrize supplies the physical infrastructure behind data centers, energy projects, advanced manufacturing, roads, public infrastructure and commercial construction. That means $AMRZ potentially sits downstream from several major spending cycles at the same time: AI infrastructure, U.S. reshoring, grid and energy buildout, and public infrastructure modernization. The company has already highlighted major data-center projects, including supplying cement and ready-mix concrete into hyperscale construction. That does NOT make the stock automatically cheap, and the government contracts we found are not large enough relative to Amrize’s ~$12.6B revenue base to move valuation by themselves. But they do confirm that the company is participating in exactly the kinds of physical buildout channels we want to monitor. SoR Read This is not: “Ro Khanna bought AMRZ, so buy AMRZ.” The political trade is secondary evidence. The real setup is stronger underlying demand, a stock hammered on margin fears, a broad executive buying cluster, a CEO putting roughly $1.27M into the stock near today’s price, AMRZ now trading below that CEO entry, and meaningful exposure to data centers, infrastructure and reshoring. The market, however, is still refusing to confirm the insiders are right. That matters. Insiders can be correct about the long-term business and still be early on the stock. I do not want to catch a falling knife simply because management is buying. Wednesday Trade Map WATCH ZONE: ~$41–$43 This puts us around or below the CEO’s recent ~$42.47 purchase price. BULLISH CONFIRMATION: I want to see $AMRZ stop making lower lows, hold the CEO-buy zone and begin reclaiming roughly $44–$45 with improving volume. CATALYST: Q3 results — week of Oct. 26. The questions that matter: Can pricing finally outrun freight, diesel and raw-material inflation? Are ASPIRE savings beginning to offset cost pressure? Does management still expect price/cost to turn positive in Q4? Is data-center and infrastructure demand accelerating enough to support another leg of growth? Warning The insider cluster may be completely right about Amrize’s long-term opportunity and still be wrong about timing. If Q3 shows worsening price/cost economics, continued margin compression or another EBITDA reset, the thesis needs to be re-underwritten. That is the falsifier. For now, $AMRZ moves onto the high-priority SoR watchlist. Not because a politician bought it. Because the market is selling shares below the CEO’s own recent purchase price while a broad group of insiders keeps buying into the weakness. That is a disagreement worth investigating. Curiosity Before Capital.

Intellia Therapeutics ($NTLA): The March FDA Catalyst That Could Reprice the Stock Congratulations to Intellia’s team, the investigators, and the patients who participated in the trials! Bringing a potential one-time treatment for hereditary angioedema to FDA review is a meaningful achievement and an encouraging step for people living with this disease. Intellia now has a March 10, 2027 FDA target decision date for lonvo-z. The FDA accepted the application with Priority Review. According to Intellia, an advisory committee meeting is not currently planned. For shareholders, I’m interested in how this timetable connects to funding and a potential launch. Management already targeted a first-half 2027 launch. March gives that plan a specific regulatory checkpoint. I would update the timing assumptions in a valuation, while keeping approval and commercial adoption uncertain. The clinical foundation deserves attention. In Phase 3, lonvo-z reduced mean monthly attacks by 87% versus placebo during weeks 5–28. That is meaningful evidence of benefit, although it doesn’t establish lifelong durability. The financial position gives Intellia room to pursue the opportunity. At approximately $12.84 in the September 8, 8:10 a.m. Pacific stock snapshot, its basic equity value was roughly $1.8 billion, using the July 31 reported share count. Intellia reported $628 million of cash and investments at June 30. Management expected those resources to fund operations at least into 2028, excluding potential lonvo-z commercial revenue. Then came September 4’s financing announcement. The “up to $400 million” OrbiMed facility provided $75 million upfront. Another $225 million depends on specified milestones, primarily related to lonvo-z. The remaining $100 million requires mutual agreement. That could reduce reliance on additional share issuance as milestones are reached. It also creates debt obligations. Adding the full $400 million to cash today would overstate available funding. I would look for potential alpha in whether lonvo-z can become a reimbursed treatment quickly enough to improve Intellia’s economics per share. Approval would be a major step. Treatment-center capacity, payer access, patient willingness, and launch spending would determine how much shareholders ultimately benefit. There is also a practical options issue. January 2027 calls expire before the scheduled FDA decision. April 16 contracts extend beyond it, but the premium matters. The April $13 call had an indicative $3.30 ask in the September 8 snapshot. Paying that price requires $16.30 at expiration to break even before fees—about 27% above the stock snapshot. That is an expiration breakeven calculation, not a forecast. A delayed FDA decision could also outlast April options. My view: a priority watchlist candidate for the next six months, with an entry price still to establish. At the September 9 and 14 investor conferences, I’d look for concrete answers on launch capacity, payer preparation, spending, and access to the remaining financing. Those answers should feed a conservative valuation before committing capital. A delay, restrictive label, manufacturing problem, or new safety concern would weaken the case. A modest share-price response alone would not prove the market has missed the opportunity. Today’s milestone deserves recognition. The next investment question is how much of that scientific progress can translate into lasting value for each share. Curiosity Before Capital. Evidence First. Disclosure: Research is for informational and educational purposes, not personalized investment advice. Investing involves risk, including loss of capital.

Lululemon Athletica Inc. ($LULU ): The $100 Turnaround Bet Needs an Earnings Floor Lululemon’s September 4 selloff put the shares near $100. That price makes the turnaround worth investigating. Before buying, I would want a firmer estimate of what the business can earn while management works through its product and demand problems. The reported quarter needs an adjustment. Q2 EPS was $2.92, including an $0.86 benefit from tariff refunds and associated interest. Remove that benefit and EPS falls to $2.06 or approximately 34% below last year. Gross margin also looks different after the refund. The reported 60.5% becomes approximately 54.9%, versus 58.5% a year earlier. These calculations remove the identified refund benefit only. They are not a complete normalization of every unusual expense. The company’s supplement describes pressure from tariffs, markdowns, and higher occupancy and depreciation costs relative to sales. Those are operating problems that need an operating recovery. The valuation needs the same treatment. Full-year EPS guidance of $9.48–$9.73 includes the $0.86 benefit. Subtracting it gives $8.62–$8.87, with a midpoint around $8.75. At the September 4 reference price of $100.61, that is roughly 11.5× earnings excluding the refund. I can see why that would attract value investors. But I would hesitate to treat the revised outlook as an earnings floor. Americas comparable sales fell 12%, and management expects total revenue to decline 10–11% in Q3. For the stock to work, investors need a reason to believe that deterioration is approaching its limit. Here is a simple way to frame the payoff: If annual EPS falls to $7 and investors pay 10×, the implied value is $70. If EPS holds around $8.75 and investors pay 12×, it is approximately $105. If EPS recovers to $10 and investors pay 14×, it is $140. Those are hypothetical earnings-and-valuation combinations, not forecasts or price targets. The multiples are assumptions representing continued weakness, stabilization, and an early recovery. They show why the distinction matters. A modest stabilization case offers limited upside from roughly $100. A stronger recovery could produce a worthwhile return, but that requires evidence we do not yet have. There are reasons to keep researching. Lululemon ended Q2 with approximately $1.4 billion in cash, and inventory units were down 7%. That financial flexibility gives management room to work. The investment case still depends on customers responding to the product. Heidi O’Neill is scheduled to begin as CEO on September 8. I would watch her approach to product selection, marketing, and spending, then test it against subsequent results. Also, she was the one with then Nike CEO (John Donahoe) pushed the disastrous direct selling initiative that wrecked Nike, so take this however you want it... The evidence I would value most is improving full-price demand. Selling more through heavier discounts would make the recovery less valuable to shareholders. I would also want sales to meet or exceed the reset outlook, followed by improving margins without another earnings reduction. My current view: watchlist, waiting for evidence of stabilization. I would reconsider even at a somewhat higher share price if the business earns greater confidence. Another guidance cut, persistent discounting, or weaker customer response would keep me cautious. The potential opportunity is to recognize a durable earnings recovery while expectations remain low. The size of the selloff alone cannot establish that opportunity. Curiosity Before Capital. Evidence First. Disclosure: Research is for informational and educational purposes, not personalized investment advice. Investing involves risk, including loss of capital.

Japan’s Yen Surge: What I’d Watch Before Buying The Tech Dip The yen touched ¥152.89 per dollar on September 8. For investors who borrowed yen to buy overseas assets, that appreciation increases the cost of repaying their debt. Unwinding those positions can force selling even when company earnings remain intact. That puts U.S. tech on my risk screen. Japan also reported a $79.6 billion decline in August reserves. Valuation changes affect that figure, so it cannot establish how many Treasuries were sold. SoR Trade Map I’m watching Invesco QQQ ETF ($QQQ) against its September 4 low. If USD/JPY keeps falling—meaning the yen strengthens—and $QQQ closes below that low, I’d reduce tactical leverage and pause fresh dip buys. Together, those moves would raise my concern about selling spreading across markets. If the yen stabilizes and $QQQ closes back above that level after breaking it, I’d start looking for small, staged entries in businesses whose earnings outlook and valuation still support buying. Another close below that low would invalidate the rebound setup. If $QQQ holds firm despite further yen strength, the immediate case for cutting tech exposure weakens. I want buying power available if funding stress pushes good businesses into attractive entry ranges. These triggers still need validation; I’d use them to organize risk decisions. Sources: Japan MOF, Reuters, BIS. Curiosity Before Capital. Evidence First. Disclosure: Research is for informational and educational purposes, not personalized investment advice. Investing involves risk, including loss of capital.

Nebius Group N.V. ($NBIS ): The GPU Pricing Signal That Could Lift 2027 Earnings Nebius has my attention because its contract economics could support a stronger 2027 earnings outlook. In its August 12 shareholder letter, management reported Q2 pricing for older-generation GPUs more than 30% above Q1. Roughly 70% of deals included customer prepayments, covering 50–60% of the associated capital spending. If those terms persist, Nebius could earn more from its hardware while customers finance a larger portion of the buildout. Most of these deals involve capacity arriving in late 2026 and contribute primarily to 2027 revenue. That timing is where I’d focus the research: how much improvement reaches next year’s earnings after financing costs and dilution? Expectations have already moved. August’s adjusted loss of $0.12 per share beat the published $0.52-loss whisper. That is historical; I couldn’t verify a whisper for the next report. Shares also rose 7.5% on September 4, closing at $226.39. Options make the timing question more immediate. IBKR’s frozen quotes retrieved before the September 8 open put the September 11 $225 call-plus-put midpoint at $19.65, roughly 8.7% of that closing share price. That premium reflects substantial priced movement in either direction. It provides no bullish confirmation, and the quotes need refreshing after the open. SoR Read The September 8 Goldman Sachs and September 9 Citi appearances are the next checkpoints. I want evidence that stronger GPU pricing is continuing into Q3 and specific dates for when new capacity starts generating revenue. If pricing holds and delivery stays on schedule, I’d test a staged share entry against a conservative 2027 valuation that includes potential dilution. Weaker rental terms or delayed deployments would make me step back. My current decision: watchlist through those updates. The contract evidence earns NBIS a closer look, but I haven’t established an attractive entry price. A 2027 earnings thesis also needs a separate timing and pricing case before I’d buy this week’s calls. Sources: Nebius Q2 disclosures and conference announcements; Earnings Whispers; IBKR. Prepared before the September 8, 2026 U.S. market open. Subscribe Free: Curiosity Before Capital. Evidence First. Disclosure: Research is for informational and educational purposes, not personalized investment advice. Investing involves risk, including loss of capital.

WaFd, Inc. ($WAFD): The $4 EPS Merger Case That Could Reprice the Stock WaFd’s Labor Day merger announcement has my attention. Management expects approximately 29% earnings-per-share accretion for WaFd shareholders. I went to the appendix to understand when those earnings are supposed to arrive. The proposed combination with EverBank is an all-stock transaction. Existing WaFd shareholders would own approximately 40.8% of the combined company, with closing expected in early 2027, subject to approvals. For someone buying WaFd, the investment depends on the future bank’s earnings, credit quality, and execution. There is no fixed cash payout to anchor the return. Here’s the distinction I would build the valuation around. Management’s $4.34 calendar-2027 EPS illustration assumes fully realized cost savings. Its integration schedule assumes only 40% of savings during 2027, with full phase-in by the second half of 2028. The appendix also shows $729 million of pro forma earnings before full savings. Divide that by 182 million diluted shares: Approximately $4.01 EPS. That is roughly 19% above the $3.37 standalone estimate. This is my calculation from management’s model, not separate company guidance. The model assumes a January 1 closing and reflects $280 million of one-time pretax costs in capital at closing. Even the phased earnings estimate could support an interesting investment. What matters now is the price and how much confidence those earnings deserve. Using approximately $4.01 EPS: $36 represents roughly 9× earnings. $40 represents roughly 10×. $44 represents roughly 11×. These are valuation sensitivities, not price targets. At the September 4 pre-announcement price of $36.30, the phased estimate implies approximately 9.1× earnings. A substantial rally would change the opportunity before the bank earns a dollar of merger savings. I would spend the next round of research on deposit pricing and credit costs. Savings have less value if retaining customers becomes expensive or loan losses rise. I also want to understand how much of the earnings improvement can persist as acquisition-accounting effects run off. That is where management needs to earn investors’ confidence. A convincing explanation of funding economics, underwriting standards, and integration milestones would matter more to me than another statement about the size of the combined bank. My current view: watchlist. I would become more interested if the shares remain near their pre-announcement valuation and the evidence supports the earnings quality. If the price quickly assumes successful integration and a higher valuation multiple, I would wait. The opportunity is to buy a stronger earnings stream at a price that leaves room for delays and mistakes. We still need to establish whether that opportunity exists. Next catalyst: the September 8 merger call at 5:00 a.m. Pacific. This is pre-call analysis. Source: WaFd/EverBank’s September 7 announcement and merger presentation. Curiosity Before Capital. Evidence First. Disclosure: Research is for informational and educational purposes, not personalized investment advice. Investing involves risk, including loss of capital.

Goldman Sachs Made Its Optical-Networking PDF Free. I Expected Fabrinet ($FN) to Win. Celestica ($CLS) Did. First, here is Goldman’s full report. It is free: goldmansachs.com/pdfs/insigh… Goldman estimates that AI networking could become a $154 billion market—roughly nine times its current size. The report works through copper cables, pluggable optical modules, co-packaged optics, PCB midplanes, scale-up networks and scale-out networks. It is a useful map of where the money may flow as AI clusters become larger and more difficult to connect. It is also the kind of TAM forecast that can make investors forget to check the price. I went into the research expecting Fabrinet to be the winner. It has specialized manufacturing capabilities, a strong balance sheet and direct exposure to optical networking without forcing investors to predict whether one specific architecture wins. Then I compared the numbers. Where Fabrinet Gets Complicated Fabrinet’s latest quarterly revenue increased 45%, while non-GAAP operating margin reached 10.9%. For a manufacturer, that margin is excellent. FN is not a broken company. At approximately $407, though, it trades around 22 times current FY2027 consensus EPS of $18.20. The number that bothered me was free cash flow. FY2026 free cash flow fell from $207 million to approximately $4 million as capital spending, inventory and receivables increased. Management expects capital expenditures to rise again in FY2027. There is also an interesting customer-mix issue. Based on Fabrinet’s disclosed customer percentages, NVIDIA-related revenue appears to have declined approximately 20% in FY2026, even as total company revenue grew 36%. That does not prove NVIDIA is leaving Fabrinet. Customer programs and product timing can move around. It does tell us the cleanest version of the story—AI demand goes up, therefore every NVIDIA-related number goes up—is incomplete. This also helps explain the post-earnings selloff. The market was not rejecting the quarter. It was reconsidering how much future growth had already been priced in while cash conversion weakened. The alternatives are not obviously cheap either. Lumentum ($LITE) just reported 109% quarterly revenue growth and a 36.6% non-GAAP operating margin. It is probably the strongest direct optical operating story, but investors are paying approximately 41 times FY2027 consensus earnings. Coherent ($COHR) trades near 30 times FY2027 earnings. Its optical exposure is attractive, but the capacity build has pushed free cash flow deeply negative. Jabil ($JBL) is closer to 18.5 times FY2027 earnings and expects at least $1.4 billion of FY2026 free cash flow. I like the cash conversion, but optical networking is a smaller part of the overall business. Then there is Celestica. Why I Landed on Celestica Celestica does not offer the same pure optical exposure as Lumentum or Coherent. It builds the broader physical infrastructure—networking platforms, switches, compute systems and other equipment that turns chips and optical components into working data centers. That makes CLS a less precise bet on optical technology, but a broader bet on the spending Goldman expects. Second-quarter revenue increased 62% to $4.7 billion. Connectivity and Cloud Solutions grew 84%, and Hardware Platform Solutions grew 58%. Adjusted operating margin improved from 7.4% to 8.2%. Management expects approximately 8.4% in the third quarter and raised its 2026 free-cash-flow outlook to $600 million. Management then made a remarkable forecast: revenue growth should accelerate beyond 65% in 2027, with adjusted EPS growing even faster. Starting from the company’s $20.5 billion 2026 revenue outlook, another 65% would put 2027 revenue above $33.8 billion. Shortly after that forecast, Celestica sold 11.1 million shares at $310 and raised approximately $3.45 billion for working capital and capital expenditures. The offering increased the basic share count by roughly 9.7%. I don’t love the dilution. But the stated purpose matters. Celestica is financing customer programs and physical capacity during a major demand ramp, rather than raising money to repair a failing balance sheet. The bet is that those proceeds become productive capacity quickly. If inventory and receivables absorb the money without producing the promised earnings, shareholders simply paid for an expensive expansion. CLS closed September 4 at $312.35—almost exactly where the offering was priced. That $310 price is not a guaranteed floor. Bankers do not defend it for us. It is simply a useful level around which to evaluate the risk. Celestica’s earlier $11.30 FY2026 EPS outlook assumed no new share issuance, so I rebuilt the forward earnings estimate using the post-offering share count. Assuming approximately $33.8–$34.5 billion of 2027 revenue, an 8.8%–9.0% adjusted operating margin, a roughly 20% tax rate and about 127 million diluted shares, I get normalized adjusted EPS of approximately $18.50–$19.50. At $312, that is roughly 16–17 times normalized FY2027 earnings. That is cheaper than Fabrinet, Lumentum and Coherent, despite Celestica currently growing faster. There is a reason for the discount. Three customers represented 63% of Celestica’s second-quarter revenue. Its top ten customers represented 83%. Margins remain lower than the optical-component companies, and the 2027 expansion will require substantial capital and working capital. I also would not treat current free cash flow as normalized. Earnings will probably run ahead of cash conversion while Celestica builds capacity. The Trade Research entry zone: $300–$315. Preferred expression: a small position in common shares. Short-dated options require getting both the thesis and the timing right, and the earnings date has not been officially confirmed. No chase above approximately $330 before the next quarterly evidence. My base case assigns a 45% probability to approximately $370–$390. That assumes normalized EPS around $19 and a 20-times multiple—above Jabil because Celestica is growing faster, but below the direct optical winners. The bull case receives a 30% probability and produces approximately $450–$465. Celestica would need EPS around $21, continued margin expansion and no meaningful customer-program delays. The remaining 25% is the risk case. If normalized EPS falls toward $16 and the market applies a 14-times manufacturing multiple, the shares could trade near $225. Those assumptions produce a rough probability-weighted value around $365, approximately 17% above the September 4 close. That is enough to support a starter-sized research position near $310. It is not enough to ignore a downside case approaching 30%. What Would Change My Mind Celestica’s next report is currently estimated for October 26, but the company has not confirmed that date. The quarter needs to do more than beat the headline EPS estimate. I want to see revenue near or above the $5.4 billion guidance midpoint, adjusted operating margin of at least 8.4%, and no retreat from management’s claim that 2027 growth will exceed 65%. Management also needs to explain where the offering proceeds are going and how quickly the new investments can generate revenue. Stronger evidence would justify increasing the research position. A fast rally toward the base valuation without corresponding estimate increases would argue against chasing the shares. The thesis fails if management materially reduces its 2027 growth outlook, operating margin falls back below 8%, a major customer program is delayed or cancelled, or another large capital raise becomes necessary before the current proceeds produce an earnings return. SoR Read Fabrinet may still be the better business. Celestica appears to be the better-priced security. Lumentum and Coherent offer more direct optical exposure, but investors are already paying heavily for it. CLS provides broader exposure to the physical AI build at a lower forward multiple, with the recent $310 offering creating a timely reference point. Near that level, the setup supports a small position with a six-to-twelve-month horizon. The next earnings report determines whether it deserves more capital. Curiosity Before Capital. -------------------------------------------------------- Primary Sources Goldman Sachs — Optical Networking: The Next Mega Trend in AI Infrastructure goldmansachs.com/insights/go… Celestica — Q2 2026 Financial Results corporate.celestica.com/news… Celestica — August 2026 Equity Offering corporate.celestica.com/news… Celestica — Q2 2026 SEC Filing sec.gov/Archives/edgar/data/… Fabrinet — FY2026 Financial Results investor.fabrinet.com/news-r… Fabrinet — FY2026 Form 10-K sec.gov/Archives/edgar/data/… Lumentum — FY2026 Financial Results investor.lumentum.com/financ… Coherent — FY2026 Financial Results coherent.com/news/press-rele… Disclosure: As of publication, the author owns shares of Celestica ($CLS) and therefore has a financial interest in the company’s share-price performance. This is independent, impersonal investment research for informational and educational purposes only—not individualized investment advice or a solicitation to buy or sell any security. The securities discussed may be volatile and involve substantial risk, including possible loss of capital. Any prices, scenarios, catalysts, entry levels, targets, or risk levels are research frameworks used to evaluate a thesis, not instructions tailored to any reader’s circumstances. Readers should conduct their own due diligence and determine independently whether any investment is appropriate for them. Any additional material position or financial conflict relating to a security discussed will be disclosed.

AI’s Power Buildout: Where I’d Start Looking For Returns Goldman Sachs forecasts global data-center power demand will rise 170% by 2030 from 2025. That would put demand at roughly 2.7 times its 2025 level. I’d start with the companies selling electrical equipment and building the grid. What appeals to me is the possibility of getting paid during the buildout, while the eventual returns on all that AI spending are still being worked out. Equipment suppliers and contractors can have a more direct connection between customer spending, contracted work, and cash receipts. That connection deserves investigation. It can also break if customers delay projects, financing falls through, or the supplier takes on work at the wrong price. Follow The Contracts Eaton ($ETN) supplies power distribution and backup-power equipment used in data centers. It has an understandable role in turning a building full of computing equipment into something that can actually operate. Before becoming comfortable with the investment, though, I’d want to see what Eaton has to spend to meet that demand. More factories, inventory, and working capital can absorb a substantial amount of cash. Rising sales would be more convincing if cash generation improved alongside them. With Quanta Services ($PWR), the questions move toward project readiness and execution. Quanta builds transmission lines, substations, and distribution infrastructure. A grid that needs investment creates potential work, but someone still has to fund the project, secure the approvals, and get skilled crews onto the site. I’d pay close attention to which projects are ready to proceed and how the contracts allocate the risk of delays and rising costs. A large pipeline looks different when you know how much of it can realistically become profitable work. GE Vernova ($GEV) adds another consideration: the manufacturing cycle. Its gas-generation equipment and grid technology fit the power buildout, and long equipment lead times can support demand visibility and pricing. Those same lead times make expansion decisions difficult. Capacity added to address today’s shortage may arrive in a different market. I’d examine firm orders separately from capacity reservations, then look at delivery schedules, margins, and the cash required to expand production. These businesses give us concrete things to investigate. Their exposure to the theme alone doesn’t establish an attractive entry price. The Capital Cycle Is The Harder Part Goldman’s comparison with the shale investment cycle is the part of its report I keep coming back to. A technological breakthrough can work remarkably well and still produce disappointing shareholder returns. If companies keep adding supply as the economics weaken, the benefits can flow to customers while investors absorb the cost. That is the lesson I’d carry into AI infrastructure. The comparison does not establish that the two industries will follow the same path. Today’s constraints could persist for years. Equipment shortages, grid connections, and construction capacity may continue to limit how quickly demand can be served. But as those constraints ease, the investment question changes. The supplier that benefited from scarcity must then show it can earn attractive returns with more competition and more capacity available. Efficiency makes the timing harder to judge. Cheaper AI could encourage enough additional usage to keep total spending and electricity demand growing. At some point, customers might instead retain more of those savings and slow their spending. I don’t believe the demand forecast, by itself, tells us when that transition happens. SoR Read My preference is to investigate businesses whose opportunity can be traced through funded projects, enforceable contracts, and improving cash returns. I’d become more confident if orders converted into cash without a disproportionate increase in capital spending or working capital. Repeated delays, cancellations, or weakening margins would make me more cautious—even if the headline power-demand forecast remained strong. The unresolved part is price. I haven’t established attractive entry levels for Eaton, Quanta, or GE Vernova here. That requires company-specific underwriting, including how much growth the current valuation already assumes. For now, this is where I’d spend the research time: finding who can earn well from the buildout, and what we would have to pay to participate. Source: Goldman Sachs’ August 6, 2026 GS SUSTAIN report and September 1 Exchanges episode: goldmansachs.com/insights/go… Explore SoR research and select “Subscribe Free”: secondorderresearch.com/ Curiosity Before Capital. Evidence First. Disclosure: Research is for informational and educational purposes, not personalized investment advice. Investing involves risk, including loss of capital.

Fervo Energy ($FRVO) & Ormat Technologies ($ORA): Who Gets Paid for Geothermal Growth? Google signing up for Fervo’s power makes the geothermal opportunity feel tangible. A major customer is committing to a project that still needs to be built. For investors, the harder question comes afterward: how much cash reaches shareholders once construction, financing and reinvestment are accounted for? That’s the question behind our latest Investment Brief. What I’m Watching At Fervo Fervo’s September 1 agreement with Google covers 396 MW, with the project expected online in 2028. There’s also an option for approximately 600 MW more, which we treat separately from committed demand. The agreement is meaningful. I’d still be cautious about using contracted megawatts alone to justify the stock’s valuation. A well can perform as designed while transmission constraints delay revenue. Faster drilling can lower costs, yet financing terms may leave common shareholders with less of the project’s cash than they expected. So the details I want to see are practical: when power reaches the customer, what it costs to deliver, and who gets paid along the way. Our brief does not establish a supported buy price for FRVO. There are still too many unresolved inputs in the project economics and funding plan. Ormat Is Easier To Analyze. The Price Still Matters. Ormat gives us operating assets and an earnings history to work with. Of the two businesses, it’s the one I can underwrite with more confidence. That doesn’t make it an obvious bargain. In our September 6 brief, we used ORA’s September 4 closing price of $105.45. Our base scenario, including the funding needed for growth, worked out to roughly an 8% annualized return through year-end 2028. Our research hurdle was 12%. Those returns depend on our earnings, financing and valuation assumptions. The earnings also need some unpacking. Selling a plant brings in real cash, but we can’t assume that profit repeats every year. Strong storage results deserve similar scrutiny before we extend them into the future. I’d want a better entry price or stronger evidence of recurring cash generation to become more interested. SoR Read Ormat offers the more established business. Fervo offers more direct exposure to enhanced-geothermal development, with greater uncertainty around delivery and financing. Neither is a demonstrated bargain at the prices analyzed. The full brief walks through the valuation scenarios, funding requirements and evidence that would change our view: secondorderresearch.com/ferv… Select “Subscribe Free” on the site to join Explorer and receive future research. No credit card required. Curiosity Before Capital. Evidence First. Disclosure: The author owns neither FRVO nor ORA. Research is for informational and educational purposes, not personalized investment advice. Investing involves risk, including loss of capital.

Second Order Research is live. Explorer membership is free. Independent investment research across public equities, macro, and market structure. We focus on questions that can change an investment decision: • What does the market already expect? • Who captures the economic value? • What could invalidate the thesis? • At what price does the opportunity become attractive? What You Can Read Investment Briefs: focused analysis of an investment opportunity, its valuation, catalysts, and risks. Full Underwriting: deeper research into the business, competitive position, financial economics, valuation scenarios, and evidence that could change our conclusions. Our approach separates what is known, what is inferred, and what remains uncertain. Join Explorer Explore the research and select “Subscribe Free” to receive research updates: secondorderresearch.com/rese… Free to join. No credit card required. We don’t sell your personal information. Unsubscribe anytime. Curiosity Before Capital. Evidence First. Disclosure: Research is for informational and educational purposes, not personalized investment advice. Investing involves risk, including loss of capital.

Boeing and Airbus Can’t Build Enough Planes. AerCap Holdings ($AER) Is Getting Paid. The obvious aircraft-shortage investment is a manufacturer recovery. The less obvious one is owning the aircraft airlines already need. AerCap is the world’s largest aviation leasing platform, with approximately: • 1,589 owned and managed aircraft • 1,300 engines • More than 300 helicopters • Roughly 400 assets on order When Boeing and Airbus miss delivery targets, airlines cannot simply stop flying. They extend existing leases, compete for available aircraft and pay more for scarce delivery slots. That creates a favorable chain for AerCap: OEM delays → aircraft scarcity → stronger lease economics → higher asset values → profitable aircraft sales → cash available for debt reduction and share repurchases The results are compelling. In Q2 2026, AerCap reported: • 18% adjusted ROE • 3.9% net spread after depreciation • $1.5 billion of operating cash flow • Book value of $119.21 per share, up 16% year over year But the investment question is no longer whether aircraft scarcity is real. It is how much of that advantage is already reflected in the stock. At our $145.40 decision price, AER traded at approximately 1.22× book value. The headline valuation—8.7× management’s $16.80 adjusted EPS guidance—looks cheap. But that guidance includes substantial first-half aircraft-sale gains. Removing their estimated contribution produces roughly $14 of normalized EPS, or closer to 10.4× earnings. Still reasonable. No longer obvious deep value. Our 18-month scenario range: • Bear: $112 • Base: $166 • Bull: $198 • Probability-weighted value: approximately $160 before dividends That produces positive expected value, but not enough margin of safety for an aggressive position at the current price. Our posture: • At or below $125: accumulate • $126–$135: starter zone • Around $145: watchlist • Above $150: require stronger proof The thesis ultimately depends on one question: Can AerCap continue producing low-to-mid-teens returns and compounding book value after aircraft production begins to normalize? If it can, AerCap may deserve a durable premium to book. If today’s returns are primarily a scarcity windfall, investors buying at the current valuation could be underwriting the strongest part of the cycle. The asset scarcity is real. The margin of safety isn’t—yet. We published the complete SoR Investment Brief, including the business model, normalized earnings, valuation scenarios, catalysts, entry bands and falsifiers: secondorderresearch.com/aerc… Curiosity Before Capital.