Artificial intelligence may be dominating conversations around technology stocks, but the bigger wealth-creation opportunity could lie in the physical infrastructure powering the AI revolution.As India gears up for a massive expansion in data-centre capacity, an estimated $25 billion in investments committed through 2030 could open up significant opportunities across power equipment, electricals, cooling infrastructure, electronics manufacturing services (EMS) and other allied sectors.Atul Suri, CEO & CIO of Marathon Trends, believes India is still at the early stages of this capex cycle. With nearly 75% of data-centre spending flowing into power, electricals and cooling, these segments could emerge as some of the biggest beneficiaries of the AI infrastructure build-out.In an interview with Kshitij Anand of ETMarkets, Suri explains why the current AI rally is increasingly a "picks-and-shovels" story rather than a conventional technology rally, why equipment makers could outperform EPC companies, and how his "buy strength, sell weakness" philosophy is helping navigate this structural market shift.He also discusses the risks of stretched valuations, the potential disruption facing traditional IT services, and how investors should approach the AI opportunity without getting caught up in FOMO. Edited Excerpts -Q) Your newsletter argues that AI has split the equity market into "builders" and "disrupted businesses." Is this a structural shift that could last for years, or merely a cyclical rotation?A) The honest answer is I don't need to have a view on that to make money from it — and that's the whole point of how we work. But if you're asking me what I see, this feels bigger than a cyclical trade, and the reason is the cause.This isn't sentiment or a liquidity wave; it's one physical force — the build-out of AI's infrastructure — pushing capital toward whoever supplies power, chips, grid and cooling, and pushing it away from the businesses AI actually automates.When something has a real-world cause with a multi-year runway behind it, you tend to get years, not months. And what struck me writing this note is that the same fork showed up in America and in India, with the same signs on both sides — that kind of symmetry across two very different markets tells you the driver is structural.That said, I'll never marry the view. My process follows the tape, and the day the tape says the trend has broken, we're out — structural story or not.Q) Have we moved from investing in AI applications to investing in the "picks-and-shovels" supporting AI? Which phase of this cycle are we currently in?A) Yes, and that shift is the single most important thing in the note. This is not a "tech rally" — it's a narrow, physical, picks-and-shovels rally. In the S&P 500, 20 stocks doubled in 2026 and 19 of them are AI infrastructure — memory, storage, semiconductors, optical, electrical equipment, cooling.Meanwhile the mega-caps have actually lagged; six of the Magnificent Seven now trail the average stock, and only Alphabet beats the equal-weight index. So the market has already voted — the money is in the shovels, not the app.We're in the heavy build-out phase, the electrons-and-steel phase. And in India we're even earlier, right at the front of our own capex cycle.Q) Many investors still associate AI with large technology companies. Why do you believe sectors like power, cables, defence electronics and cooling infrastructure are likely to create better alpha?A) Because that's simply where the strength is, and I follow strength — I don't argue with it. Think about what a data centre actually needs: it needs electrons, not headcount. AI compute flows into grid, transformers, HVDC, cables, backup gensets, cooling and servers.And the numbers in India bear it out — Power T&D and electricals up around 41%, Electronics and EMS up 54%, defence electronics up 28%, cables and wires up 26%. Compare that to IT services down 23%. The physical, electrical stuff is beating the asset-light, people-heavy stuff by a mile.The reason this creates alpha is that these are the real earnings-leverage points of the build — in a data-centre capex rupee, power, electricals and cooling are about 75% of the spend. That's where the order books sit, and you're seeing genuine re-rating on multi-year visibility — Hitachi Energy trading at something like 3.6 times book-to-sales. And crucially, this is breadth, not one hero.In Power T&D, 100% of the names beat the benchmark. When an entire theme lifts together, that's a far healthier signal than one stock carrying everything.Q) Looking Forward! India seems to be entering its own data-centre capex cycle with nearly $25 billion of investments committed by 2030. Which listed sectors are likely to be the biggest beneficiaries over the next five years?A) This is the part I'm genuinely excited about, because it's not a forecast — it's already under construction. India's data-centre capacity goes from roughly 1.6 GW today to about 5 GW by 2030 — a tripling of the most power-intensive infrastructure this country builds, with $25 billion committed, a 26% capacity CAGR, and names like Google, Microsoft, Reliance and TCS all putting money in.Now, where does that money physically go? About 75% of every build rupee goes into power, electricals and cooling — power and electricals alone are around 45%, cooling another 30%. So the biggest listed beneficiaries are exactly those buckets: power and grid equipment, electricals, cooling infrastructure, EMS, and the AI data-centre hardware and server names.In our note we flag the pick-and-shovels — server, backup and grid names for the AI build, plus defence electronics riding a separate ₹2.2 lakh crore FY27 capex line with a 60% indigenous-content mandate, and EMS on the China+1 and PLI push toward a $500 billion electronics ecosystem by FY31.One nuance I'd stress: equipment beats EPC. The theme being right doesn't mean every vehicle in it works — the equipment makers sharply outperformed the infra and railway EPC names.Q) The report paints a cautious picture on Indian IT services. Do you believe this is a temporary earnings slowdown or the beginning of a structural disruption driven by AI?A) I'll be careful here, because there are two separate questions — what the tape says, and what the long-term story is. What the tape says is unambiguous: IT services down 23%, only 14% of names beat the benchmark, TCS at roughly a six-year low, Infosys at a five-year low. That's not noise, that's the market pricing in a real problem.And the logic behind it is uncomfortable — the people-arbitrage model, that bums-on-seats labour arbitrage, is the very thing AI automates. When Accenture flags weak demand, paused discretionary spend and actual net headcount cuts, that's a demand signal you can't wave away.So structurally, yes, I think AI is a genuine threat to that model, not just a soft quarter. But — and this matters for how we run money — the valuations are now at multi-year lows, which makes them tempting on paper.For a "buy strength, sell weakness" process, cheap is not a reason to buy. There's no relative strength there yet, so for us it's a watch-list, not a position. We stay out until the tape turns. I'd frame IT services as a contrarian debate, not a momentum signal.Q) Your investment philosophy is "Buy strength. Sell weakness." How does momentum investing help investors navigate disruptive themes like AI better than traditional value investing?A) The beauty of a relative-strength process in a moment like this is that it doesn't require me to be a genius about the future. I didn't have to correctly forecast that AI would blow up memory prices or crush the services arbitrage — I just had to follow the tape that was already confirming it, position by position, with a rules-based exit when the trend breaks.That's exactly what happened in our portfolios: we tilted toward the builders of this cycle — power, grid, cables, defence electronics, cooling, the metals that physically house the compute — not because we predicted the theme, but because the tape confirmed it first.Now contrast that with traditional value investing in a disruptive theme. Value would have walked you straight into IT services, because they look cheap at multi-year lows. But cheap in the face of structural disruption is how you get a value trap — the classic falling knife.Momentum protects you from that, because it refuses to buy something with no strength, however attractive the multiple. In a violent rotation like this — where the median stock actually fell and all the return sat in a thin cohort — the only question that matters is which side of the fork you own. A relative-strength process is literally built to keep you on the right side of that line.Q) Has AI changed your stock-selection framework in any meaningful way, or has it simply reinforced your existing process?A) Honestly, it has reinforced it rather than changed it — and I'd be suspicious of any process that needed a total rewrite every time a new theme came along. The whole design of what we do is theme-agnostic: follow strength, step away from weakness, let the tape lead.AI didn't force me to invent a new rule; it just became the biggest live example of the rule working. We didn't sit around forecasting an AI infrastructure supercycle and then build a portfolio around it.The strength showed up in power, grid, cables, EMS, defence electronics — and our process moved us there, position by position, because that's what it's supposed to do. Then you step back and realise all those positions share a common thread: AI's physical build-out.If anything, the lesson it reinforces is humility — the market saw and priced this rotation before any of us could have argued it from first principles, and our job was simply to listen to the tape and stay on the right side of it.Q) The report highlights that 19 of the top 20 performing S&P 500 stocks this year are linked to AI infrastructure. Does this concentration worry you, or does it reinforce the strength of the trend?A) Both — and I don't think those two things contradict each other. On one hand it absolutely reinforces the strength: when 19 of the top 20 doublers in the entire S&P 500 sit in one chain — memory, storage, semis, optical, electrical, cooling — that's not a coincidence, that's a market delivering a very loud verdict about where the value is.And the leadership genuinely broadened away from mega-cap tech toward the infrastructure of AI, which is exactly the rotation a relative-strength process wants to ride. On the other hand, concentration is always something you respect, because the same narrowness that drives spectacular returns on the way up can reverse hard.In India I saw the same thing — the tape was carried by a thin cohort, only 22 names up more than 50% while more than half the market actually fell. So yes, I read it as strength, and I'm positioned with it — but the discipline exists precisely because it's narrow. The rules-based exit is there for the day this leadership starts to crack. Ride it while it's strong, respect that it's narrow, and don't confuse conviction with stubbornness.Q) Within the AI infrastructure value chain, which segments—power equipment, electricals, EMS, cooling, defence electronics or data-centre hardware—offer the best risk-reward today?A) Let me answer the way the tape and the capex maths point, rather than as a personal tip. If you go by where strength and breadth sit together, power T&D and electricals stand out — up around 41% with 100% of names beating the benchmark, so it's broad strength, not one lucky stock.Then Electronics and EMS, which was actually the single strongest theme at 54%. What I like structurally is that power, electricals and cooling together are about 75% of every data-centre build rupee — the earnings-leverage core, so demand visibility is highest there. Within that, one distinction I'd really push: equipment beats EPC.The equipment and components makers were up around 31% while infra, EPC and railways were actually down 3% — so even inside a winning theme, you want the equipment maker, not the contractor. Defence electronics is its own attractive pocket, riding a separate ₹2.2 lakh crore capex line and an indigenisation mandate, so it's not purely dependent on the data-centre story.The one caution I'd add is that several of these winners have already re-rated hard after these runs — so the risk-reward today is less about "is the theme right" and more about entry discipline. The theme is right; the price you pay is now the variable.Q) As more investors crowd into AI infrastructure stocks, what risks should they be mindful of over the next 12-18 months?A) The first risk is the one the note itself hints at: this leadership is narrow. Concentrated, momentum-led moves can unwind quickly, and when everyone is crowded on the same side, the exit gets small.Second, valuations — many of these names have already re-rated sharply, so a lot of good news is now in the price, and any disappointment on order inflow or execution gets punished hard.Third, the theme being right doesn't mean every stock works — we saw that with equipment beating EPC, so poor vehicle selection inside a good theme is a real risk.Fourth, a genuine risk to the whole narrative is if the AI capex cycle itself slows, or if there's any pause in the global hyperscaler build-out — the Indian names are ultimately geared to that demand.And fifth, capex cycles create bottlenecks and sometimes over-ordering, which can bite two years out. The way I'd manage all of it is exactly our discipline: don't fall in love, size sensibly, and keep the rules-based exit ready for when the tape turns.Q) If an investor wants to build an AI-focused portfolio today, how should they think about balancing growth opportunities with valuation discipline?A) First, I'd gently challenge the framing — I don't build an "AI portfolio," I build a strength portfolio that happens to be full of AI infrastructure right now because that's where the strength is. That distinction matters, because the day the strength moves elsewhere, so should you.On balancing growth and valuation, my honest position is that in a momentum framework, valuation is a risk-management tool, not an entry signal. I won't buy something just because it's cheap — that's how you end up in IT services catching a falling knife at multi-year lows.But equally, I'd be cautious about chasing names that have already tripled and now sit on very stretched multiples, because the risk-reward there turns asymmetric. So the practical answer: stay with the leading themes — power, grid, electricals, cooling, EMS, defence electronics, data-centre hardware — but respect breadth over hero stocks, prefer equipment over EPC, and let entry discipline, not FOMO, decide when you add.Build in the exit before you build in the position. Buy strength, sell weakness, stay disciplined — that's not a slogan, it's the entire risk framework. (Disclaimer: Recommendations, suggestions, views, and opinions given by experts are their own. These do not represent the views of the Economic Times)