Yet for investors who are keen to invest in the boom, framing the story around just the headline-dominating names, such as Nvidia, Micron, Google, and Meta, is myopic. Behind the chatbots we interact with lies a long, capital-intensive value chain, comprised of a diverse ecosystem of companies. These stretch from software models, to semiconductor fabricators, data centres and firms responsible for energy infrastructure.
This more complex reality matters, because AI is not one sector or one investment. Rather, it is a chain of specialised companies with different economics and risks. Perhaps more importantly, this also means that the more interesting investment opportunities lie in understanding how value flows through the whole system.
To do that, perhaps, the easiest way to see all of this in action is to follow the prompt; envision yourself typing a question into your favourite AI Chat, and press “Enter”. Your question just activated one of the most complex industrial systems ever built.
Meet the model: The hyperscalers
The journey begins when the words in your prompt or question are broken into small units called tokens. A token may be a punctuation mark, or even just part of it.
See also: Compute is king in the AI era, says Neuberger fund manager
This is where your question meets the AI model at the application layer. Companies such as OpenAI, Anthropic, Google and Meta build the models and services users interact with. This interaction does not happen in your phone but in a data centre linked to a large cloud platform.
Amazon Web Services, Microsoft Azure and Google Cloud are called hyperscalers because they build computing capacity on an enormous scale. Think of them as digital factories. They combine servers, networks, storage and software, then use this capacity themselves or rent it to AI developers. As demand for AI rises, these firms are committing vast sums to chips, data centres, networks and energy.
Inside the digital factory
Within these factories, your tokens reach specialised processors.
See also: Structural shifts across Asia with China’s rise and AI revolution, says Nomura
Nvidia’s graphics processing units (GPUs) perform many calculations simultaneously, making them particularly effective at training and running AI models. The company’s advantage, however, extends beyond the chip; it also provides software, networking and complete systems that allow thousands of GPUs to work together.
Broadcom plays a different role. It helps technology companies develop customised AI accelerators, and supplies networking chips that allow processors to exchange information rapidly. If Nvidia offers a ready-made engine, Broadcom can help hyperscalers design their own ones and then proceed to connect thousands of them.
The processors also need high bandwidth memory from companies such as SK Hynix, Micron and Samsung. Without fast memory, even an advanced GPU can be left wanting: The equivalent of re-fueling a gas-guzzling racing car through a drinking straw.
From blueprints to actual silicon
While Nvidia and Broadcom design many chips, they generally do not manufacture them.
Rather, their blueprints travel to foundries — most notably Taiwan Semiconductor Manufacturing Company (TSMC) — where microscopic transistors are printed onto silicon wafers. Advanced packaging places processors and memory close together so they communicate faster and consume less power.
Behind TSMC stands ASML, which makes the lithography systems used to print the tiny patterns forming advanced chips. While ASML does not create the AI model or finished processor, many bleeding edge chips could not possibly be produced at scale without its sophisticated and expensive machines. Some of the most important AI beneficiaries therefore sit several layers behind the companies that consumers recognise.
From data centres to power stations
Those chips are then installed in servers, stacked in racks and connected inside a data centre. The building needs data or optic fibre cables, transformers, backup power and cooling. AI servers also generate far more heat than conventional systems.
For more stories about where money flows, click here for Capital Section
Again, this brings in yet another sector of interest. Companies such as Vertiv, Schneider Electric and Eaton supply cooling, power management and electrical equipment. Follow the power cable beyond the data centre and it leads to substations, transmission lines and power plants. Follow this chain even further and it reaches the resources needed to build them: From copper, steel, aluminium and concrete to even water.
In essence, a question typed into a chat window can create demand stretching from software laboratories to factories, ports, mines and power plants; it also reveals potentially untapped investment opportunities in the process.
The Red Dot in the route
Singapore marks an important junction in this journey. Its semiconductor cluster spans wafer fabrication, memory, manufacturing equipment, testing and packaging.
Micron has major memory operations here, while equipment companies use Singapore as a manufacturing and regional base. Chips and equipment move through its trade and logistics network, and many MNCs coordinate Asian operations from here.
Your simple query may therefore touch Singapore through a memory chip, manufacturing equipment, data cable or the regional team coordinating the system. In fact, Singapore’s position in the value chain has emerged as one of the main reasons for our resilience during the Iran-US conflict, especially vis-à-vis its negative impact on many of our neighbours.
Why this matters for investors
Building awareness of the complex ecosystem of companies dotted along the value chain quickly reveals a vast universe of players beyond just the big names we associate most immediately with AI.
Crucially, the more-than-US$800 billion ($1.034 trillion) of capital expenditure committed by hyperscalers for this year alone does not remain just with them. Rather, it cascades through this chain.
More AI capacity means more accelerators, memory and networking. More chips require foundry and packaging capacity. More servers require data centres, cooling, transformers and electricity. For investors keen on participating meaningfully in the AI boom, there are investment opportunities to be had that clearly cut across multiple sectors and industries.
But this does not mean every player or company will benefit from this equally. Competition is intensifying at every layer. Hyperscalers are developing custom chips to lower costs and reduce dependence on Nvidia. Chipmakers compete for manufacturing capacity. Data centre operators compete for land and power. Suppliers must expand without creating excess capacity.
For investors and advisors, the key questions, then, surround what role a company plays, how scarce that capability is, how difficult it is to replace, and how much of the spending it can turn into lasting profits. The strongest beneficiaries may be those that own a bottleneck, are well-poised to remove one, or can provide the tools everyone else needs.
As the world increasingly moves towards AI-ready societies, the more exciting entry points into the boom thus extend far beyond the names dominating the headlines today. Instead, they lie in tracing precisely how different companies along the value chain help move a prompt from your screen, through the machine, and back again as an answer.
Chez Anbu is head of wealth advisory at OCBC
