The AI boom is raising costs: Where can traders look beyond Nvidia?

AI infrastructure spending is no longer only an Nvidia story. This article explores how rising demand for chips, electricity, copper and data-centre infrastructure could create opportunities and risks across technology, energy, commodities and broader stock indices.

Artificial Intelligence (AI) is usually seen as a story about productivity. Companies may use it to automate more tasks, improve their software, increase demand for cloud services, and create new revenue sources.

That may happen. But before AI becomes more profitable, it first becomes more expensive.

Building and running advanced AI models requires real infrastructure. Companies need chips, powerful servers, networking equipment, data-centre space, cooling systems, and enough electricity. These are not small details. They are the foundation of the AI economy.

This changes how I would look at the theme.

A company that sells chips, networking hardware or data-centre capacity can benefit directly from the AI buildout. But a company that has to buy that infrastructure faces a different situation. It may be investing for long-term growth, but it is also spending more money and putting pressure on cash flow.

A cloud provider, for example, may spend heavily on AI infrastructure to protect its market position. Investors may accept this spending if revenue is still growing strongly. But if AI revenue grows more slowly than expected, the same spending that once looked ambitious can start to look like a margin problem.

That is why I think the AI trade is becoming more selective.

Invest in the world’s biggest market movers

Trade stocks of the most successful global tech and industry giants.*

Explore stocks

*

Terms and conditions apply.

Why the AI boom is becoming a cost story

AI is usually presented as a productivity story. Companies could automate more processes, improve software, increase cloud usage, and create new revenue streams.

That may happen. But before AI becomes more profitable, it first has to become more expensive.

Training and running advanced models requires a physical infrastructure layer: chips, high-performance servers, networking equipment, data-centre space, cooling systems and power capacity. These inputs are not secondary details. They are the base of the entire AI economy.

This changes the way I would analyse the theme.

A company selling chips, networking hardware or data-centre capacity may benefit directly from the AI buildout. A company buying that infrastructure is in a different position. It may be investing for long-term advantage, but it is also increasing capex and possibly putting pressure on free cash flow.

A cloud provider, for example, can spend heavily on AI infrastructure to defend its market position. Investors may accept that spending while revenue growth is strong. But if monetisation becomes slower than expected, the same capex that once looked like ambition can start to look like a margin problem.

That is why I think the AI trade is becoming more selective.

Beyond Nvidia: The AI spending chain

Nvidia remains the reference point, but the AI supply chain is larger than one company. Traders looking beyond Nvidia can think in layers: semiconductors, networking, hyperscalers, data centres, power and commodities.

A practical monitoring list could include USTEC, NVDA, AMD, AVGO, TSM, MSFT, AMZN, GOOGL, META, ORCL, EQIX, XCUUSD, and XNGUSD.

Each one tells a different part of the story.

This does not mean all these instruments should move together. They should not. Nvidia, USTEC, copper, and natural gas are exposed to different forces.

That difference is useful. It helps traders see whether the market is still rewarding the obvious AI leaders or starting to price the second-order effects of the infrastructure buildout.

Semiconductors: Where expectations start to matter

Semiconductors remain the centre of the AI trade. Advanced chips are the foundation of AI training and inference, and demand for computing power continues to shape investor sentiment across the sector.

But the semiconductor story is not only about GPUs.

AI data centres also need networking chips, custom accelerators, foundry capacity and semiconductor equipment. That is why AMD, Broadcom, and TSM are useful names to monitor alongside Nvidia.

AMD gives traders a way to follow whether AI chip demand is spreading beyond the dominant leader. Broadcom connects the theme to networking, connectivity, and custom silicon. TSM is important because advanced chip manufacturing remains one of the most critical parts of the global AI supply chain.

The challenge is that investors already know this.

That makes expectations the real issue. If a semiconductor stock has already rallied sharply, strong growth may not be enough. The market may demand exceptional growth, better guidance, and evidence that AI demand is not slowing.

This is where I would be cautious. A company can be well positioned in the AI chain and still become difficult to trade if the valuation assumes too much.

Memory, networking, and the less visible AI components

The AI boom is often explained through processing power, but AI systems also need to move and store huge amounts of data efficiently.

That brings memory and networking into the conversation.

High-performance AI workloads depend on advanced memory, fast interconnections, and strong data-centre networking. These areas do not always receive the same attention as GPU leaders, but they are essential to the infrastructure layer.

For traders, this matters because market narratives often rotate. The first phase of a trade usually rewards the most obvious names. Later, investors may start looking for companies that control less visible but still critical parts of the chain.

Broadcom is relevant here because it links AI demand to networking chips and custom silicon. Cisco can also be useful to watch as a broader networking infrastructure name, although I would not treat it as a pure AI trade.

The important point is not to force every infrastructure company into the AI basket. The question is whether the demand for AI is large enough to affect revenue, margins, or guidance. Without that, the connection may be more narrative than financial.

Hyperscalers: When AI spending becomes a tolerance test

The other side of the AI trade is the group of companies spending heavily to build and deploy AI infrastructure.

Microsoft, Amazon, Alphabet, Meta, and Oracle are useful examples. They may benefit from AI, but they are also paying for the infrastructure that makes AI possible.

This is where the analysis becomes more nuanced.

Microsoft can connect AI to cloud demand, enterprise software, and productivity tools. Amazon can link the theme to AWS and data-centre scale. Alphabet has exposure through cloud, search, advertising, and model development. Meta can use AI to improve advertising efficiency, recommendation systems, and future products. Oracle has become more relevant as investors focus on demand for cloud infrastructure.

But the market will not reward spending forever just because it carries the AI label.

At some point, investors need evidence that higher capex is producing returns. That could mean faster cloud growth, stronger margins, better advertising performance, higher enterprise adoption, or clearer monetisation.

Meta is a useful example because investors can reward aggressive infrastructure spending when it supports advertising efficiency and user engagement. But tolerance for that spending can fall quickly if capex increases faster than visible returns.

Alphabet faces a different version of the same question. AI can strengthen its cloud business and product ecosystem, but it can also increase competitive pressure in search and raise the cost of defending its core business.

That is why I would not put all hyperscalers in the same category. The trade depends on whether the company can turn AI spending into pricing power, retention, productivity, or new revenue.

Electricity: AI’s hidden input cost

One of the most underestimated parts of the AI boom is electricity.

Data centres consume large amounts of power, and AI workloads are more energy-intensive than many traditional computing tasks. As AI adoption expands, electricity demand becomes more than an operational detail. It becomes a constraint.

This creates two market effects.

Power-linked assets may benefit from stronger demand. At the same time, companies building AI infrastructure may face higher operating costs if electricity prices rise or grid capacity becomes limited.

For traders, this means AI is not only a technology theme. It is also an energy theme.

Power demand may become especially important in regions where data-centre growth is concentrated. If grid expansion fails to keep pace with AI infrastructure demand, power availability could become a bottleneck. That would affect cloud providers, data-centre operators, utilities, and industrial infrastructure companies.

Natural gas deserves attention in this context. It is not a pure AI instrument, but it can become relevant if rising power demand increases the need for reliable electricity generation. That makes XNGUSD a useful instrument to monitor within the broader AI energy story.

AI infrastructure spending is expected to increase electricity demand from data centres. The International Energy Agency projects that global electricity generation needed to supply data centres could rise from 460 TWh in 2024 to more than 1,000 TWh in 2030 and around 1,300 TWh in 2035, supporting the view that AI is also an energy, infrastructure and cost-pressure theme. Source: International Energy Agency.

Copper and data-centre infrastructure

AI may look like a software revolution, but it depends heavily on physical materials.

Data centres require electrical wiring, transformers, cooling systems, grid connections, and other infrastructure. Copper is important because of its role in power transmission and electrical equipment.

That makes XCUUSD a useful macro instrument for traders following the AI infrastructure theme.

However, copper will not move only because of AI. It is influenced by China, global growth, mining supply, inventories, and the US dollar as well. But if investors continue to associate AI with data-centre growth, grid expansion and electrification, copper can become part of the same broader conversation.

Equinix is relevant from a different angle. It gives traders exposure to the physical data-centre layer rather than the chip or cloud layer. That can be useful if the market starts rewarding infrastructure capacity and not only semiconductor performance.

This is one of the areas I find most interesting. The obvious AI trade is already crowded. The less obvious question is whether the market will start paying more attention to the companies and commodities that make the infrastructure possible.

USTEC and broader stock indices

For traders who prefer a broader view, USTEC remains one of the clearest ways to monitor AI-related sentiment.

The index includes many of the companies most associated with technology, cloud computing, semiconductors, and digital infrastructure. When AI optimism rises, USTEC often reflects that mood. When investors worry about valuations, interest rates, or earnings disappointment, the same index can quickly come under pressure.

The advantage of tracking USTEC is that it captures the broader market reaction rather than relying on a single company. The disadvantage is that AI is only one part of the index story.

USTEC can rise on AI optimism, but it can also fall on rate expectations, dollar strength, weak earnings guidance, or broader risk-off sentiment.

For me, USTEC is useful as a sentiment gauge. It helps answer one basic question: is the market still rewarding AI exposure broadly, or is it becoming more selective?

US500 can add context. If USTEC outperforms US500, technology and AI sentiment may still be leading. If USTEC begins to underperform, traders may need to ask whether capital is rotating away from the AI trade or simply expanding into other parts of the market.

The margin problem: Not every AI spender is a winner

The cleanest distinction in this theme is between companies selling AI infrastructure and companies buying it.

A chipmaker, networking provider, data-centre company, or power-linked instrument can benefit directly from rising AI investment. Their products or services become more valuable because other companies need them.

The situation is different for companies spending heavily on AI infrastructure. For them, AI can increase capex and operating costs, placing pressure on free cash flow.

Investors may accept that if they believe the spending builds a stronger long-term position. But if monetisation remains unclear, the same AI investment can become a concern.

This is already one of the key debates around large technology companies. Heavy AI spending can be interpreted as a sign of future growth, but it can also raise questions about efficiency and returns.

Big Tech AI infrastructure spending has increased sharply, with combined capex across Alphabet, Amazon, Meta, Microsoft and Oracle rising from approximately USD 162.3 billion in 2022 to USD 448.3 billion in 2025. The increase shows the AI boom is becoming a major infrastructure spending cycle, not just a software or chip story. Source: Visual Capitalist and company reports.

The market may eventually separate companies with pricing power from those simply paying more to stay competitive.

That separation is where the trading opportunity may appear.

Key risks for traders

Valuations may already reflect the AI boom

The first risk is valuation. Many AI-related assets have already rallied strongly. If expectations are high, even solid earnings may not be enough to drive further upside.

In that environment, traders should ask whether the market is pricing in realistic growth or assuming near-perfect execution.

Capex may become a problem

AI infrastructure spending can support future growth, but can also reduce free cash flow in the short term. If companies keep investing without showing clear returns, investors may become less patient.

This is especially important for large cloud and technology companies. The market may reward AI spending during the early phase of the cycle, but later demand evidence that the spending is producing revenue and margin improvement.

Energy constraints could slow expansion

Data-centre growth depends on power availability. If electricity demand rises faster than grid capacity, expansion could become more expensive or slower than expected.

This would not necessarily end the AI boom, but it could change which companies benefit most from it.

The AI narrative could rotate

Markets rarely keep rewarding the same part of a theme forever. Chip leaders dominated the first phase of the AI boom. A later phase could shift attention towards memory, networking, power, copper, cooling, or data-centre infrastructure.

It could also rotate away from AI if earnings disappoint or if investors decide valuations have become too stretched.

My approach to trading the AI infrastructure theme

I would not treat AI as a single-stock story.

Nvidia may remain central, but the broader opportunity is in understanding the spending chain behind artificial intelligence. The most interesting trades may come from identifying where the market has not fully priced the second-order effects of AI infrastructure demand.

Before considering a trade, I would first ask whether the company is selling into AI demand or paying more to participate in it.

That distinction changes the analysis. Nvidia, AMD, Broadcom, and TSM sit closer to the supply side of the AI trade. Microsoft, Amazon, Alphabet, Meta and Oracle sit closer to the spending and monetisation side. Equinix, Cisco, XCUUSD, and XNGUSD help traders follow the physical infrastructure and energy layer.

In practical terms, I would focus on four questions:

• Is the instrument exposed to AI demand as a supplier, a spender, or a sentiment proxy?
• Is earnings growth keeping pace with valuation?
• Are margins improving, stable, or deteriorating?
• Is the instrument outperforming its closest benchmark or simply following the broader market?

For broader exposure, I would monitor USTEC to identify whether AI sentiment remains strong across technology. For more specific exposure, I would look at semiconductors, data-centre infrastructure, copper and natural gas.

I would also avoid chasing headlines. When every market conversation is about AI, the risk of crowded positioning increases. In that environment, pullbacks, earnings revisions, and technical structure become more important than the narrative itself.

Trade stocks with zero commission

Trade the biggest names in tech and industry with low transaction costs.*

Try free demo

*

Terms and conditions apply.

Final thoughts: AI trade is becoming more selective

The AI boom is not only about who builds the best model or who sells the most advanced GPU.

It is also about the companies supplying the physical infrastructure behind artificial intelligence and the companies absorbing the rising cost of that infrastructure.

That is why I think traders should look beyond Nvidia.

Semiconductors, networking, cloud infrastructure, electricity, copper, data centres, natural gas, and USTEC all tell part of the story. The challenge is identifying where AI spending creates pricing power and where it creates margin pressure.

The opportunity is not simply to trade the AI narrative.

It is to understand where the money is flowing, where costs are rising, and where the market may be underestimating the next stage of the AI infrastructure cycle.