Nvidia stock (NASDAQ:NVDA) has returned roughly 1,200% since ChatGPT launched in late 2022. Most investors know that number by now. Fewer have followed the cascade of winners the same AI buildout created downstream, across sectors that don’t always appear in technology indices.
The chip layer itself runs deeper than one company
ASML, the Dutch company with a global monopoly on extreme ultraviolet lithography machines, supplies every advanced chip fab on the planet: TSMC, Samsung, and Intel. Without ASML’s equipment, none of them can manufacture at cutting-edge geometries. TSMC manufactures the chips that train every major AI model. ARM Holdings designs the architecture running 99% of the world’s mobile devices and, increasingly, data centre CPUs. These are structural businesses with moats built over decades, and the market has priced them accordingly.
What the market was slower to price is what a $750 billion AI infrastructure buildout creates downstream of the chip.
Every constraint reveals the next
The AI buildout did not stay a chip problem. It became a networking problem, then a power problem, then a raw materials problem, each layer stressed in sequence, each stress generating a new cohort of winners.
Networking and custom silicon
The hyperscalers are committing close to $750 billion in AI infrastructure capex in 2026, up nearly 70% from 2025, after every major tech company raised guidance on its late April earnings calls. Every GPU cluster needs high-speed switching to operate at that scale. Arista Networks raised its 2026 AI revenue target to $3.5 billion on its Q1 2026 earnings call.
Broadcom’s AI revenue hit $8.4 billion last quarter, up 106% year-on-year, backed by a signed backlog of $73 billion covering roughly the next 18 months of deliveries. Six of the world’s largest AI platforms, from Google to Meta to Anthropic, now deploy Broadcom’s custom silicon alongside Nvidia GPUs, because at hyperscaler volumes, custom chips optimised for a specific workload consistently outperform general-purpose GPUs on cost per inference.
The $750 billion they are spending this year flows into infrastructure businesses that most Indian retail portfolios have never touched.
Power and energy
A single Nvidia Blackwell rack consumes up to 132 kilowatts. Multiply that across a $750 billion capex cycle and the electricity bill ceases to be a technology problem and becomes a grid infrastructure problem.
Goldman Sachs has revised its data-centre power demand forecast upward three times in 18 months, each time in the same direction, landing at 1,350 terawatt-hours by 2030, a 220% increase from 2023 levels. The IEA projects US data centre electricity consumption will reach 426 terawatt-hours by 2030, up 133% from today. That’s roughly a quarter of India’s entire annual power consumption, from a sector that barely registered a decade ago.
The companies capturing that spend sit outside every conventional technology index. GE Vernova, which makes the power equipment that feeds data centres, booked $18.3 billion in orders in Q1 2026, up 71% year-on-year, carrying a $163 billion backlog. Constellation Energy has signed 20-year nuclear power agreements with both Microsoft and Meta for separate plants, and Vistra locked in its own deal with Meta in January 2026, covering over 2.1 gigawatts from nuclear plants across Ohio and Pennsylvania. Two hyperscalers, two operators, three separate 20-year nuclear procurement agreements executed inside 18 months: the AI buildout has turned nuclear energy from a regulatory headache into a procurement priority.
Cooling
Vertiv builds the thermal management systems that keep GPU clusters from destroying themselves. It carries a $15 billion backlog, saw orders surge 252% in Q4 2025, and raised full-year EPS guidance to $6.35 on its Q1 2026 print, up 51% from 2025. The infrastructure that makes Nvidia’s chips usable at scale has a ticker and a backlog measured in billions.
Metals and mining: the bottleneck beneath the bottlenecks
Every data centre runs on copper. Every transformer needs it. Every power cable running from a Constellation nuclear plant to a Microsoft server farm, every liquid cooling loop inside a Vertiv rack, every high-speed networking switch in an Arista cluster carries copper at its core, which means the AI buildout created a commodity demand shock that the mining industry was structurally unprepared for.
Copper hit an all-time high of $14,527 per tonne in January 2026.COPX, the copper miners ETF, returned 93% in 2025 against copper’s own 43% gain, a 2.3x leverage ratio for investors who tracked the demand picture rather than the commodity headline. REMX, tracking rare earth miners, delivered 92% in the same year. China controls roughly 90% of global rare earth processing capacity, and when that concentration becomes a geopolitical liability every hyperscaler is now actively spending capital to reduce, it stops being a mining industry statistic and starts being embedded in every AI infrastructure dollar spent.
The access for Indian investors
None of these companies trade on Indian exchanges. No domestic mutual fund gives meaningful stock-level exposure to Vertiv, GE Vernova, Broadcom, copper miners, or rare earths.
Most Indian investors who wanted AI exposure bought Nvidia. That was the right call. The stocks that compounded alongside it, Broadcom, Vertiv, GE Vernova, copper miners, are available on the same exchange, and require the same LRS remittance via affordable platforms.
