It's easy to think about Artificial Intelligence (AI) as a breakthrough that only impacts software – smarter chatbots, better search engines or automated assistants.
But Nvidia CEO Jensen Huang has recently pushed back on that idea.
In a blog published in March 2026, Huang described AI as a “five-layer cake” – not just a digital innovation, but a physical industrial system stretching from electricity generation all the way through to our daily applications.
The key takeaway for investors is this, AI doesn’t just sit neatly inside one sector. It spans everything from power grids and semiconductor factories to data centres and enterprise software. Understanding how these layers fit together can help identify where the AI ecosystem is creating value – not just at the top, but all the way down to its foundations.
This article isn’t personal advice. If you’re not sure an investment is right for you, seek advice. Investments and any income from them will rise and fall in value, so you could get back less than you invest.
Investing in an individual company isn’t right for everyone because if that company fails, you could lose your whole investment. If you cannot afford this, investing in a single company might not be right for you. You should make sure you understand the companies you’re investing in and their specific risks. You should also make sure any shares you own are part of a diversified portfolio.
Breaking Down the Five Layers of AI
1. Energy
At the very bottom of the AI stack sits energy, supporting everything built on top.
Training and running AI systems requires enormous amounts of electricity, with every response generated by an AI model ultimately powered by the grid (or an independent energy source). Without large-scale energy production, alongside cooling systems, AI simply cannot operate.
As AI adoption grows, access to reliable power becomes a key constraint on the amount of computing capacity that can be deployed. In other words, the amount of intelligence we can generate is directly linked to the amount of energy available.
This problem requires a mix of traditional and new energy sources. Nuclear offers one path forward, and we’ve seen some of the major AI builders sign contracts to set up their own reactors with companies like Constellation Energy, the largest clean-power producer in the US.
We expect this trend to continue, and, aside from the power companies themselves, we see uranium miners like Cameco benefitting. The risk here is that much of the good news is already priced in, and, as with all large buildouts, execution risk is high.
2. Chips
Once energy is available, it needs to be converted into something usable – AI.
That’s where specialised chips come in.
Unlike traditional processors found in personal computers or smartphones, AI workloads require huge numbers of calculations to be carried out simultaneously. Modern AI chips are designed specifically for this, enabling machines to process massive datasets far more efficiently than older computing architectures.
Progress at this layer plays a major role in determining how fast AI technology can scale – and how affordable it becomes for businesses looking to deploy it.
Nvidia is our preferred way to gain exposure to the AI chips layer. It’s supported by a strong pipeline of next-generation products that continue to push the performance frontier and make it increasingly difficult for rivals to compete at scale.
We also see Broadcom as a potential diversifier. Its growing presence in custom AI chips offers exposure to a more niche but rapidly expanding segment of the market, as some of the larger AI names look to tailor infrastructure for specific workloads.
The main risk to both companies, and a key debate at the moment, is whether major customers can keep up their investment plans.
3. Infrastructure
AI chips don’t operate in isolation.
Infrastructure includes everything needed to house and connect thousands of processors into a single functioning system – from data centres and networking equipment to cooling technology and physical construction.
This is perhaps the broadest segment so far. Nvidia works with over 200 partners to bring its chip designs to life in a modern data centre or “AI factory”.
Companies like Dell have emerged as key partners in the buildout of AI infrastructure, helping businesses deploy the servers and supporting systems needed to scale AI. Momentum in its AI server business continues to build, and it's fast becoming a leader in the space – though the competition is tough.
Caterpillar and Sunbelt Rentals (formerly Ashtead) sit on the construction side, supplying the heavy machinery and equipment needed to physically build out data centres and supporting infrastructure. As investment in AI capacity grows, we expect demand for construction services to follow suit.
4. Models
Once the physical foundations are in place, the next step is developing AI models capable of performing useful tasks.
Models are trained on large datasets to recognise patterns, generate content, make predictions or carry out specific instructions. Some models are broad in scope and can be applied across multiple use cases, while others are built with particular industries in mind – like healthcare, finance or scientific research.
These systems form the “brains” behind AI-powered tools and services across a wide range of sectors.
The model layer spans both public and private companies. OpenAI (ChatGPT) and Anthropic (Claude) have seen valuations rise sharply in private markets. In public markets, Alphabet (Gemini), Microsoft through its OpenAI stake, and Meta (Llama) are among the biggest names.
We think 2026 is the year when models move beyond simply answering questions to taking on more complex tasks, possibly even using external tools to automate real-world workflows. Still, we see this as one of the more competitive layers, where it's hard to pick a winner just yet.
5. Applications
At the top of the stack sit the AI-driven applications that businesses and consumers interact with directly.
This includes everything from productivity assistants and recommendation engines to fraud detection platforms and autonomous machines. These tools often receive a lot of attention, as they represent the visible face of AI adoption.
However, every application relies on the layers beneath it – from trained models and physical infrastructure to the chips running in data centres and the electricity powering them.
The application layer is likely to see the most disruption. Over recent months, software companies have come under pressure as markets have begun to price in an uncertain future, with new AI-driven applications seen as potential disruptors.
We believe this uncertainty has created attractive opportunities across the software landscape. High-quality companies are more likely to benefit from AI than be hindered by it and are now trading at compelling valuations.
We favour businesses with one or more of these characteristics.
1. Deep, hard-to-replicate data sets
2. Regulated operating environments
3. High switching costs and deeply embedded software.
Names like RELX and Microsoft fit this profile, though each carries its own risks.
What’s next for investors
We see opportunities emerging across the AI stack at each layer of development. Some are more specialised, offering higher potential rewards but also greater risk, while others provide more generalised exposure as critical enablers of AI adoption.
The foundational layers of energy, chips and infrastructure are already underway. This is an area where we think the market is underestimating the scale of this buildout. Demand is showing no signs of slowing, and we think investment will continue longer than many think, offering opportunities at each layer.
Models continue to improve rapidly, but we still struggle to pick the winners. The mix of private and public means it's not always possible to invest in the leading names – though we note that could change if rumoured IPOs happen.
The application layer is only just getting started, offering the biggest risk-reward opportunity. While software companies have come under pressure due to fears of AI disruption, we believe many will emerge as AI winners by evolving and adapting their existing applications. Our preference is for companies with strong data moats, exposure to regulated industries, or high switching costs.
Progress is unlikely to be smooth, and there will be bumps along the way. But while the path may not be linear, we think the longer-term direction of travel across the AI ecosystem is becoming increasingly clear.
The author holds shares in Meta, Microsoft, Nvidia, and RELX.
This article is original Hargreaves Lansdown content, published by Hargreaves Lansdown. It was correct as at the date of publication, and our views may have changed since then. Investments rise and fall in value so investors could make a loss.
This article is not advice or a recommendation to buy, sell or hold any investment. No view is given on the present or future value or price of any investment, and investors should form their own view on any proposed investment.