We held our fourth bi-monthly member call yesterday, drawing members from Uruguay to Singapore and upstate New York. One question ran through almost every contribution: the AI capital expenditure wave is real, but it lands in reported numbers in a way that flatters the sellers today and may flatter no one later. Several members offered ways to align with the trend without overpaying, while others shared ideas away from the AI trade.
For those who couldn’t join us, here is a synthesis of the key insights and thesis highlights shared by our community of intelligent investors.
This post summarizes views expressed by participants on an MOI Global member call and is provided for informational and educational purposes only. Nothing here is investment advice, nor a recommendation or solicitation to buy or sell any security. Speakers may hold positions in the securities discussed, and figures cited were stated on the call and have not been independently verified. Please do your own research and consult a qualified adviser before making any investment decision.
The 17% margin that is mostly an accounting artifact
James Emanuel opened with an argument he published earlier this month. Since the Second World War, net margins for the S&P 500 have generally sat in the five to six percent range. They reached roughly 13% after the Tax Cuts and Jobs Act, drifted toward 11% as rates rose, returned to 13% after ChatGPT, and on recent FactSet data now sit near 17%.
His explanation is accrual accounting rather than productivity. The hyperscalers and frontier labs are spending some 800 billion dollars this year and perhaps over a trillion next. That cash lands with the chip and memory makers, who book it as revenue immediately, while the spenders capitalize and depreciate it over five to seven years. The mismatch inflates index margins now and reverses when the build-out slows, as vendor revenues fall while depreciation runs on. Anyone modeling the index off a 17% net margin, James argued, is modeling close to a fiction. If the index-level data is unreliable, what does that say about the constituent numbers we all use?
Elliot Turner pointed out that incremental EBIT margins at Google, Amazon and Microsoft are high and still accelerating even after depreciation, which could offset falling vendor margins later. Barton Hooper asked whether the hyperscalers are themselves over-earning, noting an H100 rents for more today than at launch. How much of that spending is underwritten by the same end demand counted twice is now among the market’s most interesting open questions.
Life science tools: focus on the picks and shovels
Elliot Turner, who hosts a biotech fireside chat with Peter Mantas at Latticework 2026 in November, made the case for life science tools rather than biotech itself. Biotech endured close to a decade of very little, with the XBI starting last year near levels it first reached in 2015. Fundraising has run four consecutive quarters at or above the prior peak, and the regulatory bottleneck has begun to clear. Tools appeal because you need them whichever drug wins, and because approval locks them in: once a component is written into an FDA filing, the manufacturer must keep buying it, and biosimilars inherit the same process.
His first name is Sartorius (Germany: SRT, SRT3), roughly 80% recurring revenue, where the equipment line has been sluggish after the COVID overbuild and five years without meaningful greenfield capacity. He expects that to invert sharply once capacity tightens, and flagged Lonza as the one CDMO he respects. His second is Bruker (US: BRKR), which makes high-precision instruments across genomics and proteomics and is moving into clinical and companion diagnostics. A surprising detail: Bruker’s largest customer is Taiwan Semiconductor, because the industry is shifting to X-ray metrology, and that segment grows 30% at 30% margins against a corporate average of single-digit growth at 12%.
PVA TePla: volume story at a reasonable price
Nitin Sacheti argued that most picks-and-shovels plays in AI have been pricing stories, and a bottleneck price is a poor thing to capitalize into a multiple. He prefers volume growth at a reasonable multiple, in businesses that have not borrowed their future through average selling prices.
The name he highlighted is PVA TePla (Germany: TPE), which he recently visited in Wettenberg, Germany where he sat down with the CEO and CFO. One division builds the furnaces that grow silicon, silicon carbide and indium phosphide boules for wafer makers such as Siltronic and GlobalWafers. The second is metrology, where ultrasound inspection runs fast enough to sit on the line as a first-pass screen ahead of slower X-ray, with TSMC, Intel and Micron as customers. Book-to-bill runs around 2.2 times, with metrology orders converting to revenue over the next few quarters. He sees three-plus euros of EPS in 2028 and four-plus in 2029 against a 29-euro share price, with recent weakness driven by tariffs and order timing rather than anything structural. Nitin will present PVA TePla in depth at MOI’s fully online European Investing Summit in October.
Copper: four demand drivers against a supply chain with long lead times
Will Thomson, who hosts the copper panel at Latticework 2026 in November, called copper the most macro of metals, pulled by four secular demand drivers: energy transition, AI, core economic demand and defense. Data centers alone consumed 1.1 million tons last year out of roughly 28 million globally. Supply is growing about 1%, grades are falling, and the capital cost of new production has risen to some 20,000 to 30,000 dollars per ton from 15,000 to 20,000.
That makes brownfield restarts attractive, since an old mine at 1% copper is now a good orebody. Magna Mining (Canada: NICU) is restarting past-producing nickel, copper and platinum-group mines in Sudbury using existing infrastructure; its shares have fallen from about 3.54 Canadian dollars to about 2.50 on sector sentiment rather than anything company-specific. Blue Moon Metals (Canada: MOON, US: BMM) brings on a sedimentary-hosted deposit in Norway in the fourth quarter of next year, capital efficient and long-lived, with a deep water port at the mine mouth, and the first new European copper mine in a decade or more. Copper near six dollars eighty sits above the roughly five-fifty incentive price, but it is also where substitution toward aluminum starts to interest buyers.
European banks, where expectations remain quite low despite stock price recovery
Patrick Brennan observed that European bank shares have outrun the Mag Seven over the past four or five years, which says more about the starting point than the businesses. The sector had been priced for a return to negative rates that forward curves no longer support.
Metro Bank (UK: MTRO) trades around 170 pence with average deposit costs below 1%, striking in the UK, and he sees roughly 40 pence of EPS by 2028 under a controlling shareholder with a long record of selling banks well. CAB Payments (UK: CABP) is the more unusual situation: a London-listed payments business focused on Africa and expanding into Latin America, growing around 20% at the top line, listed at 330 pence and trading near 82. Its 40% holder has a bid in at 84, StoneX has bid 110, and an earlier approach came nearer 145. The bid needs 75% approval, which Patrick does not expect it to get. With a new dividend and a CET1 ratio in the low 20s against a 14 to 15% target, he sees ample capacity to return capital if it stays independent.
Allocating capital away from the AI trade
Patrick Brown holds coal as an inflation hedge and a sector in which he can own the low-cost producers. Warrior Met Coal (US: HCC) has brought on a new mine well below the cost of its existing production. Core Natural Resources (US: CNR), formed when CONSOL Energy merged with a metallurgical producer and now split roughly evenly, was the company whose exports were disrupted by the Maryland bridge collapse. Both sit very low on the cost curve, so competitors fail first, supply leaves, and prices recover before the pain reaches them. Neither carries meaningful debt. He invoked Nassim Taleb’s antifragility, and noted that with the Strait of Hormuz closure affecting gas, some countries weighing gas-fired plants are reconsidering coal.
Keith Smith has been working through small banks and insurers on a Berkshire-style float model. His example was Northeast Bank (US: NBN), effectively a debt fund inside a bank charter, buying loan portfolios from sellers in distress or in merger, alongside capital-light SBA and insured-loan businesses. It earns ROEs around 20% at under ten times earnings, run by Rick Wayne, who has executed this playbook since the 1980s. Javier López Bernardo of BrightGate Capital presented it at Best Ideas 2024.
Can you still be a pure stock picker?
A member raised an interesting question: whether it is still possible to be purely bottom-up given how much the macro has driven the market, and whether a more technological, more international S&P still represents what it once did. Our continued belief: A bottom-up research focus remains entirely valid, and several people on yesterday’s call are proof of it. However, the further an investor wades into the fastest-moving sectors, the more of a macro thesis he is underwriting, whether explicitly or implicitly.
Yifan Zhang cited a fascinating example: AI-related security spending is rising, but not as he expected: models find vulnerabilities in human-written code faster than teams can approve fixes, so firms are hiring more engineers to keep up. Demand is being created by the tool meant to reduce it.
We closed with a timely insight by Wyatt Sparks: After three months in which the market chopped sideways while rotating violently underneath, his “unreasonable” rule for portfolio management is that over-trading an account is probably not the way to go. Go touch grass, he suggested, and enjoy the rest of the summer.
I found the call incredibly valuable and hope this summary provides a few worthwhile insights for your own capital allocation process.
Our next live call will be on October 21 at 2:00 pm ET. No formal registration is needed; simply click the following link at the scheduled time to join.