Industrias: la cadena de la IA y las megatendenciasIndustries: the AI value chain and megatrends
Sector a sector, qué empresas dominan cada eslabón, con qué ETFs se puede invertir en él y qué riesgos tiene.Sector by sector: which companies lead each link, which ETFs give you exposure and what the risks are.
IA y semiconductores
The heart of it all. Running a single AI chip takes seven layers; the higher up, the stronger the moat. Almost every household-name company lives in the lower half.
1Design · EDA software
The software chips are designed with. A pure oligopoly with recurring revenue. The most underrated layer.
Design · EDA software
The software chips are designed with. A pure oligopoly with recurring revenue. The most underrated layer.
El software de diseño electrónico (EDA) es el "AutoCAD de los chips": sin él es imposible diseñar un procesador de miles de millones de transistores. Es un oligopolio casi perfecto con ingresos recurrentes por licencia, márgenes altísimos y clientes que no pueden cambiar de herramienta sin rehacer todo su flujo de trabajo.
Líderes: Synopsys · Cadence · Siemens EDA
Para invertir: Synopsys (SNPS) y Cadence (CDNS) cotizan como pure-plays. Siemens EDA (antes Mentor Graphics) no cotiza por separado: es una división de Siemens AG (ADR: SIEGY), un conglomerado industrial donde el EDA es solo una parte.
Electronic design automation (EDA) software is the "AutoCAD of chips": without it you can't design a processor with billions of transistors. It's an almost perfect oligopoly with recurring licence revenue, huge margins and customers who can't switch tools without redoing their whole workflow.
Leaders: Synopsys · Cadence · Siemens EDA
To invest: Synopsys (SNPS) and Cadence (CDNS) trade as pure-plays. Siemens EDA (formerly Mentor Graphics) isn't listed separately: it's a division of Siemens AG (ADR: SIEGY), an industrial conglomerate where EDA is just one part.
2Equipment & lithography
The machines that make the chips. ASML holds the EUV lithography monopoly. Picks and shovels.
Equipment & lithography
The machines that make the chips. ASML holds the EUV lithography monopoly. Picks and shovels.
Las máquinas que convierten el diseño en silicio real. ASML es la joya de la corona: monopoliza la litografía EUV, sin la cual no existen los chips más avanzados, y cada máquina cuesta cientos de millones. El resto domina otros pasos —deposición, grabado, control de calidad—. Es la jugada picks and shovels: ganan venda quien venda, porque todos los fabricantes les compran.
Líderes: ASML · Applied Materials · Lam Research · KLA · Tokyo Electron
Foso enorme y barreras de entrada casi infranqueables: décadas de I+D y propiedad intelectual acumulada. El punto débil: dependen de la geopolítica (restricciones de exportación a China).
The machines that turn a design into real silicon. ASML is the crown jewel: it monopolises EUV lithography, without which the most advanced chips don't exist, and each machine costs hundreds of millions. The rest dominate other steps —deposition, etching, quality control—. It's the picks and shovels play: they win no matter who sells the chips, because every manufacturer buys from them.
Leaders: ASML · Applied Materials · Lam Research · KLA · Tokyo Electron
A huge moat with almost insurmountable barriers to entry: decades of R&D and accumulated IP. The weak spot: they depend on geopolitics (export restrictions to China).
3Foundry
Who makes the chip. TSMC has ~72% of the market and controls CoWoS packaging, today the sharpest bottleneck.
Foundry
Who makes the chip. TSMC has ~72% of the market and controls CoWoS packaging, today the sharpest bottleneck.
Las fábricas que producen físicamente los chips que otros diseñan. TSMC es el rey absoluto: ~72% del mercado y, sobre todo, el control del empaquetado avanzado (CoWoS), hoy el cuello de botella más agudo de toda la industria. Samsung pelea en segundo lugar e Intel intenta volver al negocio de fundición para terceros.
Líderes: TSMC · Samsung · Intel
Construir una fábrica nueva ("fab") cuesta más de 20.000 millones y varios años: por eso el poder está tan concentrado. Riesgo geopolítico claro por la concentración en Taiwán.
The factories that physically produce the chips others design. TSMC is the undisputed king: ~72% of the market and, above all, control of advanced packaging (CoWoS), today the sharpest bottleneck in the whole industry. Samsung fights for second place and Intel is trying to re-enter the third-party foundry business.
Leaders: TSMC · Samsung · Intel
Building a new fab costs over $20 billion and several years: that's why power is so concentrated. Clear geopolitical risk from the concentration in Taiwan.
4HBM / DRAM memory
The memory that feeds the GPUs. Only three makers; sold out through 2026. One of the hottest themes.
HBM / DRAM memory
The memory that feeds the GPUs. Only three makers; sold out through 2026. One of the hottest themes.
La memoria de alta velocidad (HBM) que alimenta a las GPUs. Solo tres empresas en el mundo la fabrican, y la demanda supera con mucho a la oferta: SK Hynix y Micron tienen vendida su producción de 2026 por contrato. Fue el epicentro de la caída de junio de 2026 y, a la vez, la razón por la que Micron resistió: sus ingresos del año ya estaban comprometidos.
Líderes: SK Hynix · Micron · Samsung
Históricamente la memoria es muy cíclica (sube y baja con fuerza). La IA ha cambiado las reglas, pero sigue siendo el eslabón más volátil de toda la cadena.
The high-speed memory (HBM) that feeds the GPUs. Only three companies in the world make it, and demand far outstrips supply: SK Hynix and Micron have their 2026 output sold under contract. It was the epicentre of the June 2026 selloff and, at the same time, the reason Micron held up: its yearly revenue was already committed.
Leaders: SK Hynix · Micron · Samsung
Historically memory is very cyclical (it swings hard). AI has changed the rules, but it's still the most volatile link in the whole chain.
5Chip design · GPU & ASIC
GPUs (Nvidia, AMD) and custom chips for hyperscalers (Broadcom, Marvell). Top growth, more competition.
Chip design · GPU & ASIC
GPUs (Nvidia, AMD) and custom chips for hyperscalers (Broadcom, Marvell). Top growth, more competition.
La capa más famosa y la que más titulares genera. Nvidia diseña las GPUs que mueven la IA y es la empresa más valiosa del mundo; AMD persigue. Pero el dinero también va a los chips a medida (ASIC) que Broadcom y Marvell diseñan para Google, Meta o Amazon, y a Arm, que cobra un royalty por casi cada chip móvil del planeta.
Líderes: Nvidia · AMD · Broadcom · Marvell · Arm
Máximo crecimiento, pero también máxima competencia y valoraciones exigentes: es la capa donde más se nota una corrección, como se vio en junio de 2026.
The most famous layer and the one that grabs the headlines. Nvidia designs the GPUs that power AI and is the most valuable company in the world; AMD chases. But money also flows to the custom chips (ASIC) that Broadcom and Marvell design for Google, Meta or Amazon, and to Arm, which charges a royalty on nearly every mobile chip on the planet.
Leaders: Nvidia · AMD · Broadcom · Marvell · Arm
Top growth, but also top competition and demanding valuations: it's the layer where a correction bites hardest, as seen in June 2026.
6Physical infrastructure
Connecting, powering and cooling the GPUs. The
Physical infrastructure
Connecting, powering and cooling the GPUs. The
La capa "aburrida" que casi nadie incluye y que crece de verdad. Cuando un centro de datos de IA cuesta miles de millones, esta "fontanería" se lleva una parte enorme del gasto: conectar miles de GPUs (Arista), alimentar y refrigerar (Vertiv), ensamblar los servidores (Super Micro) y mover los datos por óptica (Coherent).
Líderes: Arista · Vertiv · Super Micro · Coherent
Enlaza directamente con la industria de Energía: el siguiente cuello de botella no es el chip, es la electricidad para hacerlo funcionar.
The "boring" layer almost nobody includes and that really grows. When an AI data center costs billions, this "plumbing" takes a huge share of the spend: connecting thousands of GPUs (Arista), powering and cooling them (Vertiv), assembling the servers (Super Micro) and moving data over optics (Coherent).
Leaders: Arista · Vertiv · Super Micro · Coherent
It links directly to the Energy industry: the next bottleneck isn't the chip, it's the electricity to run it.
7Cloud & hyperscalers
Those who pay for the party and build their own chips (TPU, Trainium, Maia).
Cloud & hyperscalers
Those who pay for the party and build their own chips (TPU, Trainium, Maia).
Los que pagan la fiesta: gastarán entre 720.000 y 745.000 millones de dólares en 2026 en infraestructura de IA, según sus previsiones de julio. Además diseñan sus propios chips (TPU de Google, Trainium de Amazon, Maia de Microsoft) para depender menos de Nvidia. Para un inversor son la exposición "diversificada": negocios gigantes y rentables donde la IA es una palanca de crecimiento, no toda la apuesta.
Líderes: Microsoft · Google · Amazon · Meta
El riesgo: si el retorno de la IA decepciona, este gasto récord es lo primero que el mercado teme que se recorte. Es el debate de fondo de la "burbuja".
The ones who pay for the party: they'll spend between $720 and $745 billion in 2026 on AI infrastructure, per their July guidance. They also design their own chips (Google's TPU, Amazon's Trainium, Microsoft's Maia) to depend less on Nvidia. For an investor they're the "diversified" exposure: giant, profitable businesses where AI is a growth lever, not the whole bet.
Leaders: Microsoft · Google · Amazon · Meta
The risk: if AI's payoff disappoints, this record spending is the first thing the market fears will be cut. It's the underlying "bubble" debate.
The bottlenecks
Advanced packaging (CoWoS)
TSMC · Amkor · IntelIt joins the logic chip with the HBM memory. TSMC's capacity is sold out well into 2026 and oversubscribed. Building a new plant needs equipment with 12-18 month lead times. The most pressing one.
HBM memory
SK Hynix · Micron · SamsungSK Hynix has already sold all its 2026 HBM and Micron's is reserved. Goldman calls it the most severe memory shortage in 15 years, with strain until ~2028.
EUV lithography
ASML (en solitario)The underlying structural bottleneck: without ASML's EUV machines there are no advanced nodes. Less acute short-term, but the deepest moat in the industry —a real monopoly—.
Electric power
Red · Vertiv · utilitiesThe one that's coming, and maybe the definitive one. GPUs are no longer so scarce; the power grid is. Expanding it runs on industrial timelines, not chip-fab ones. It links to the Energy industry.
The four bottleneck companies
| Company | Bottleneck it controls | Position | Track it |
|---|---|---|---|
| ASML | EUV lithographyMonopoly (~100% of EUV) | World's only EUV maker | ASML |
| TSMC | Foundry + CoWoSPackaging sold out to 2026 | ~72% of the foundry market | TSM |
| SK Hynix | HBM memoryAll of 2026 sold | ~58% of global HBM · leader | SK Hynix |
| Micron | HBM/DRAM memoryOnly pure-play in the US | ~21% of HBM · cap > $1T | MU |
UCITS semiconductor ETFs
Energy & electrification
Most tied to AIThe power grid is AI's bottleneck No. 4 and, at the same time, the engine of electric cars and reindustrialization. Data-center consumption could double to ~945 TWh by 2030. Nuclear, including small modular reactors (SMR), is the only clean 24/7 source; Microsoft, Amazon and Google are already signing direct deals. Within the theme: nuclear/uranium, renewables and the grid itself (transformers, cabling).
Key companies
Cameco · Constellation Energy · Vistra · BWX Technologies · GE Vernova · Siemens Energy · Schneider Electric · NextEra
UCITS ETFs
Copper & critical raw materials
The base of everythingEvery megatrend (AI, defense, electrification) pulls on the same copper, lithium and rare earths. Copper faces a supply deficit of ~30% by 2035, lithium demand could quintuple by 2040, and China refines 19 of the 20 strategic minerals. It's the
Key companies
Freeport-McMoRan · Southern Copper · BHP · Glencore · Antofagasta · Albemarle · MP Materials
UCITS ETF
Defense & security
Multipolar worldDeglobalization is structural and governments are raising military spending: NATO is aiming for 5% of GDP by 2035, with a share for cyberdefense. It includes weapons, military tech and reshoring. Risk: it depends on political budgets and carries ethical considerations for some investors.
Key companies
Lockheed Martin · RTX · Northrop Grumman · Palantir · Rheinmetall · BAE Systems · Leonardo · Thales · Indra
UCITS ETFs
Robotics & automation
Physical AIAI stepping out of the screen into the real world. Citi estimates up to $7 trillion in humanoid robots by 2050; component costs are falling fast and ageing plus labor shortages push adoption. Lots of potential, but also lots of short-term hype.
Key companies
Nvidia · ABB · Keyence · Fanuc · Intuitive Surgical · Rockwell Automation · Tesla
UCITS ETF
Health, longevity & GLP-1
DemographicsAgeing is one of the few nearly guaranteed trends. The visible engine is GLP-1 drugs (obesity/diabetes), with sales up +500% between 2018 and 2023 and oral versions arriving in 2026, plus AI applied to drug discovery. For pure GLP-1 there's no clean UCITS ETF: exposure goes through broad pharma or the leading stocks.
Key companies
Novo Nordisk · Eli Lilly · Stryker · Medtronic · Intuitive Surgical
UCITS ETF
Cybersecurity
Toll of digitalizationThe mandatory toll of digitalization: the more AI, the larger the attack surface, and threats from non-human agents are accelerating. Structural, low-discretion demand —no company can cut here—. Risk: a competitive sector with constant consolidation.
Key companies
CrowdStrike · Palo Alto Networks · Zscaler · Fortinet · Cloudflare
UCITS ETFs
Water
The quiet trendThe least-discussed trend: the World Economic Forum estimates a water funding gap of $7.5 trillion by 2040. Growing demand from industry, AI data-center cooling and climate. Stable, low-cyclical businesses with long-duration concessions.
Key companies
Veolia · Xylem · American Water Works · Ecolab · Geberit
UCITS ETF
Gold & real assets
A hedge, not a sectorMore than a growth sector, a portfolio hedge against currency
Key companies (miners)
Newmont · Agnico Eagle · Barrick · Franco-Nevada
UCITS vehicles
How to fit them in: satellite themes, not the whole portfolio
La idea habitual: un núcleo global diversificado (por ejemplo, un MSCI World) como base, y estos temas como satélites pequeños —algo así como un 5-10% cada uno como tope— para no concentrar en exceso. Cuantos más temas añadas, más se solapan (el cobre aparece en energía y en IA; Nvidia, en IA y en robótica), así que evita duplicar la misma apuesta. Puedes probar repartos concretos en la pestaña Simulador.The usual idea: a diversified global core (for example, an MSCI World) as the base, and these themes as small satellites —something like 5-10% each as a cap— so you don't over-concentrate. The more themes you add, the more they overlap (copper shows up in energy and AI; Nvidia in AI and robotics), so avoid doubling the same bet. You can try specific allocations in the Simulator tab.
Links open Yahoo Finance in a new tab. For ETFs they go to the ISIN search (Yahoo indexes by ticker; click the result that matches your market). ISINs and data verified as of today: holdings and fees change, so confirm them on justETF or the issuer's site before investing.