By Mike Smith, senior portfolio manager and head of the growth equity team, and Jon Baranko, chief investment officer at Allspring Global Investors
The artificial intelligence (AI) boom is supported by a vast infrastructure stack. Understanding the role of key players such as hyperscalers, semiconductors, and data centres is crucial for identifying investment opportunities beyond the headlines.
AI has evolved rapidly, as has the language around it. Our framework shows where investment is being spent and where growth opportunities are emerging.
Most people interact with AI through applications such as ChatGPT, Claude, Gemini or Copilot. Their providers, or “AI labs”, generally earn revenue through subscriptions or usage-based charges linked to token consumption. But every interaction depends on an infrastructure stack, with participants capturing value at each step.
Three layers are central: hyperscalers, semiconductors and data centres.
Hyperscalers: Supplying cloud capacity
Cloud computing is essential for training and deploying AI models as adoption accelerates.
Hyperscalers such as Microsoft, Amazon, Google, Meta and Oracle provide AI compute and infrastructure, investing heavily to become the industry’s “toll roads”.
Many model providers have strategic, sometimes “circular”, relationships with hyperscalers. OpenAI has ties to Microsoft Azure and Oracle, while Anthropic uses Amazon Web Services and Google Cloud. Hyperscalers can be both investors in and customers of model providers, helping deploy AI applications on a scale.
Neoclouds provide similar services on a smaller, specialised basis. CoreWeave, Nebius, Hut 8 and DigitalOcean span niches including model training, physical power and enterprise deployment.
Semiconductors: Processing AI workloads
Semiconductors are critical to AI. This layer is often discussed in “racks”: systems of processing and memory chips, including GPUs, plus networking equipment used to develop large-scale models.
A single rack can cost several million dollars and data centres can house tens of thousands, so semiconductors capture a disproportionate share of hyperscaler spending. Much of the wider stack exists to support this compute capacity.
GPUs, or accelerators, perform AI’s high-speed calculations. Nvidia captures a significant share of this market, with Advanced Micro Devices a major competitor.
Application-specific integrated circuits (ASICs) perform specific AI tasks more efficiently and with lower power use than general-purpose GPUs. Broadcom, Marvell and Cerebras design ASICs and rely on foundries including TSMC, Samsung and Intel for manufacturing.
Memory is also essential: DRAM and high-bandwidth memory feed active processing, while NAND flash stores data until needed. Providers include Micron, SK Hynix, and Samsung, plus SanDisk, Western Digital and Seagate in storage.
Data centres: The physical infrastructure
Data centres house AI workloads, creating opportunities for companies supplying the technologies needed to expand capacity.
Power is emerging as one of the biggest constraints. By 2030, data centres are estimated to consume around 1,000 terawatt-hours of electricity annually, roughly Japan’s current annual consumption.
Meeting this demand will require major grid investment, benefitting companies such as Quanta Services, while “behind the meter” generators from Bloom Energy and GE Vernova can help bypass lengthy grid connections.
Cooling is equally critical. HVAC systems from Comfort Systems and EMCOR, components from Johnson Controls and Modine Manufacturing, and liquid-cooling solutions from Vertiv help keep equipment operating reliably.
Interconnect technologies transmit power and signals. Copper is effective but heated intensive; fibre optics are more efficient but harder to manufacture, install and test. Credo, Astera Labs and Amphenol are expanding into optical products, while Viavi Solutions provides network testing and measurement.
Colocation providers such as Equinix, Digital Realty, QTS and CyrusOne operate data-centre facilities, while Schneider Electric and Siemens supply sensors and control systems that optimise power and operations.
Allspring Growth Team perspective: Go deeper into the stack
The scale of the AI infrastructure buildout is unlike anything the US has seen in modern times, rivalled by the railroad boom of the 1800s. In the first stage, hyperscalers and mega-cap technology companies such as Amazon, Microsoft, Alphabet, Micron and Nvidia dominated headlines and returns.
Now investors are increasingly rewarding the recipients of AI capital spending, not only the spenders. Attention is broadening to companies enabling the buildout and solving constraints in power, cooling and storage.
Next-generation data centres are measured in gigawatts. One gigawatt is comparable with the power consumed by Indianapolis, with around 800,000 households. A 1-gigawatt data centre is estimated to require about $35bn, and some forecasts suggest 100GW could be added by decade-end.
This buildout is bringing industrial companies into sharper focus. Multi-year order backlogs can create more visible earnings growth and help address concerns about speculative excess.
The opportunity is also to broaden small and mid-sized specialists. Temperature control can account for as much as 40% of data-centre energy use, creating a strong incentive for more efficient power and cooling technologies.
Ultimately, investing in AI means looking beyond the headline names. We believe investors who go deeper into the stack may benefit from the next stage of growth. Fundamental research can identify recipients of AI-related capital spending, particularly where demand is outpacing capacity and bottlenecks create opportunities for infrastructure, equipment, and service providers.
