AI is booming yet the supply chain is not keeping up. There lies the opportunity

Money is pouring into AI at an unprecedented pace. The only problem? AI infrastructure bottlenecks. Chips, memory, wafers, networking equipment, optical infrastructure: the global manufacturing base cannot scale with the pace of AI capital spending.

For investors paying attention to the right layer of the market, that constraint is becoming one of the most compelling plays in global equities right now.

That is the central argument from ClearBridge Investments, whose head of global growth, Eliza Mazen, has been positioning the firm’s Global Growth strategy around companies sitting at AI’s most acute pressure points.

For Mazen, the shift is already well underway.

“Generative artificial intelligence is no longer just about the most visible technology platforms. It is becoming a global supply chain bottleneck story.”

She adds that the scale of AI capital spending is overwhelming the existing manufacturing base for advanced computing, creating constraints across multiple product categories including graphics processing units (GPUs), central processing units (CPUs), memory, wafers, advanced packaging, storage, networking and optical infrastructure.

For companies operating at those AI infrastructure bottlenecks, the result is something they rarely had before: pricing power.

From cyclical suppliers to structural winners

Historically, semiconductor and hardware suppliers operated in deeply cyclical businesses. Prices rose and fell with inventory cycles. Margins compressed when supply caught up with demand.

AI has changed that dynamic, at least for the suppliers directly in its path.

High-bandwidth memory (HBM), which powers AI workloads, is one of the clearest examples. South Korea’s Samsung Electronics is seeing higher margins for both commodity memory and HBM. Because HBM requires more silicon wafers, demand has flowed through to wafer suppliers.

Japan’s Shin-Etsu Chemical has moved rapidly from dealing with a supply glut to facing genuine shortages.

The same dynamic is playing out in analogue chips. Germany’s Infineon, which makes power management circuits that regulate voltage for specific server workloads, pushed through two price increases in the second quarter alone.

“These constraints are giving pricing power to companies that historically had little of it, while transforming these once-cyclical industries into structural growth beneficiaries,” Mazen says.

GPUs, CPUs and the rise of new entrants

Nvidia dominates the GPU market, and supply remains constrained. That constraint is opening doors for new entrants.

Taiwan’s MediaTek, better known for smartphone and broadband chips, has developed AI accelerator chips called tensor processing units (TPUs) for customers including Alphabet. It is a meaningful expansion of its addressable market.

CPUs, long considered the workhorse of traditional computing, are also seeing renewed demand. Agentic AI, which involves AI systems that plan and execute multi-step tasks, requires CPUs to complement GPUs and coordinate complex workloads.

US chipmaker Advanced Micro Devices has gained new customers and raised prices as AI infrastructure bottlenecks tighten CPU supply.

At the foundry level, Taiwan Semiconductor (TSMC) holds a near-monopoly on leading-edge manufacturing. That position has strengthened its pricing power considerably.

Samsung is investing in advanced manufacturing technologies to win new customers, with Tesla and Apple among those it has attracted.

Networking and optical: the overlooked layer

Mazen highlights an area of the AI infrastructure buildout that receives far less attention than it deserves.

“The networking infrastructure that amplifies AI computation processing power by connecting GPUs within data centres and AI clusters across multiple facilities is often overlooked,” she says.

She adds that within data centres, US-based Arista Networks provides high-speed switching and software packages that enable low-latency communication across GPUs, with clients including Microsoft and Meta Platforms.

Canada’s Celestica manufactures more flexible “white box” switches that allow hyperscalers to customise their infrastructure at lower hardware cost.

Further out, optical networking companies convert electrical signals into light for transmission through fibre optic cables, enabling high-speed connectivity over longer distances.

Finland’s Nokia has transformed itself from a telecom provider into a leading optical and laser supplier. With key rival Ciena at capacity, Nokia is expanding its own production capabilities fivefold to meet accelerating demand.

UK-based Halma’s optical photonics business has doubled in revenue over two years. The firm is projected to grow approximately 30 per cent annually. Italian manufacturer Prysmian is also benefiting from supply constraints in fibre optic cable, giving it room to raise prices as demand outpaces supply.

Managing the momentum risk

Mazen is clear-eyed about the risks that come with a theme this dominant. AI has at times concentrated returns in a narrow, momentum-driven cohort, and sentiment can shift fast.

The strategy’s answer to that risk is discipline. Rather than chasing the hype or following momentum, the firm targets companies whose earnings growth rests on genuine pricing power and structural demand, not market enthusiasm.

“By targeting growth companies that provide products and services addressing critical AI infrastructure bottlenecks, and whose competitive positioning is backed by pricing power, we have confidence in our technology exposure,” she says.

For advisers reviewing technology allocations, the most durable exposure to the AI buildout is not in the names generating the most headlines. It is in the companies solving the problems those headline names cannot function without.