The artificial intelligence (AI) investment cycle is entering a second phase in 2026, with demand broadening beyond GPUs to inference, AI agents and commercial applications, creating new opportunities across optical connectivity, power, advanced packaging and semiconductor equipment, Kenanga Research said on Monday.

Kenanga, which recently attended Huatai Securities’ investment forum in Shanghai, said in a note that Huatai expects the next stage of the AI cycle to be driven increasingly by capacity expansion and volume growth rather than the initial GPU-led infrastructure buildout.

The investment house said investors should focus on bottlenecks across five areas — compute, storage, optical, power and equipment — while tracking operational milestones such as orders, customer qualification, equipment installation, utilization and delivery lead times before assessing the impact on revenue, margins and return on invested capital.

“AI exposure” alone is increasingly insufficient to support premium valuations, Kenanga said, citing Huatai’s view that the market is shifting from an AI narrative towards evidence of execution and earnings conversion.

Huatai characterises 2023 to 2025 as the “first inning” of AI, dominated by large-model training and benefiting companies supplying GPUs and high-bandwidth memory (HBM).

From 2026, the industry is expected to enter a “second inning”, with inference and AI agents broadening demand across the infrastructure stack as AI moves from training a relatively small number of frontier models towards serving a much larger base of users, applications and agents.

This could shift incremental capital expenditure towards memory, networking, optical connectivity, power, cooling and advanced packaging, while eventually creating greater opportunities further downstream in AI-enabled devices, software and services, Kenanga said.

Huatai places the current AI boom within the broader concept of a Kondratieff technology cycle, which describes long periods of investment and economic transformation following major technological breakthroughs.

Previous cycles have been associated with technologies such as steam power, railways, electrification, automobiles and information technology. Under the framework, AI, alongside technologies such as humanoid robotics and biotechnology, could form part of another long-duration technology cycle.

For investors, however, the relevance is less about timing the market precisely and more about recognizing that breakthrough technologies typically require years of supporting infrastructure investment before reaching mass adoption.

Huatai also uses Nvidia chief executive Jensen Huang’s “five-layer cake” framework to illustrate the AI ecosystem, comprising energy, chips, infrastructure, models and applications.

The framework highlights how growth at the application and model layers ultimately creates demand for more data-center capacity, semiconductors and electricity.

Kenanga said this suggests some of the more durable opportunities could initially remain in the lower layers of the stack, particularly where supply is constrained, including power, advanced packaging, optical connectivity, cooling and AI infrastructure.

Over time, however, value creation could migrate towards models and applications as infrastructure matures and commercialization becomes more important.

Bottlenecks become key investment focus

Huatai’s five-part framework — compute, storage, optical, power and equipment — is designed to identify where incremental AI capital expenditure is flowing and where supply constraints are emerging.

Compute remains the foundation, but competitive advantages increasingly extend beyond chip performance to software ecosystems, developers and customer lock-in.

Storage and optical connectivity are becoming increasingly important as AI clusters grow. HBM remains a critical component of high-end AI computing, although Huatai cautions investors against relying solely on memory pricing as an indicator of structural demand.

Instead, capacity expansion and market-share gains are viewed as stronger evidence of underlying demand growth than temporary increases in average selling prices.

Optical connectivity is also becoming more strategic as AI clusters require higher bandwidth and lower latency to connect thousands of accelerators. This could increase demand for optical modules and silicon photonics as cluster sizes expand.

Power could become another major constraint

Huatai expects the future location of computing capacity to increasingly depend on the availability of secure and scalable electricity, making power generation, grid infrastructure, backup systems and energy efficiency more important to the development of AI data centres.

The availability of power could determine which projects proceed, how quickly they scale and which locations emerge as major AI hubs.

Advanced packaging is another structural enabler as conventional transistor scaling becomes more challenging.

As AI processors become larger and more complex and require tighter integration between logic, memory and interconnects, advanced packaging is becoming an increasingly important way to sustain performance improvements. This creates opportunities for packaging providers as well as equipment and materials suppliers supporting capacity expansion.

Huatai’s preferred investment principle is that “capacity expansion beats price increases”.

Kenanga said earnings growth driven by higher shipment volumes, utilisation and market-share gains provides stronger evidence of structural demand than growth driven mainly by pricing.

This means investors should distinguish between companies benefiting from actual capacity additions and those whose earnings are being boosted primarily by short-term pricing tightness.

Huatai also recommends identifying “toll-road” companies that control scarce or difficult-to-replace parts of the AI ecosystem and can benefit regardless of which model, hyperscaler or application ultimately emerges as the winner.

Malaysia semiconductor read-through

For Malaysia’s technology sector, Kenanga said the framework provides a way to distinguish genuine AI beneficiaries from companies benefiting mainly from thematic interest.

The screening process should move from AI exposure to customer qualification, capacity expansion, utilisation, revenue contribution and, ultimately, margin and return on invested capital.

Kenanga said Malaysian companies positioned in semiconductor equipment, advanced packaging and outsourced semiconductor assembly and test (OSAT), photonics, data-centre infrastructure and power-related supply chains could benefit from AI-driven capacity expansion, particularly where growth is volume-led through new programmes, equipment installations, production ramps and market-share gains.

The research house maintained a neutral stance on Malaysia’s technology sector and said it continued to favor companies with visible volume growth, improving utilisation and earnings conversion rather than AI exposure alone.

Kenanga said Huatai’s emphasis on optical bottlenecks reinforced its positive view on NATGATE.

Global optical transceiver demand continues to exceed supply, according to Huatai, with limited near-term risk of oversupply as customers are directly involved in suppliers’ capacity planning.

NATGATE is investing in dedicated capacity for new customer programmes and has expanded beyond conventional surface-mount technology and final assembly into chip-on-carrier and chip-on-substrate packaging, TOSA/ROSA/TROSA assembly, fibre attachment, active alignment, burn-in and optical testing.

Kenanga said discussions with Chinese optical transceiver peers also reinforced Malaysia’s role in the supply chain serving leading US customers.

For NATGATE, its targeted production start in November 2026 and subsequent production ramp are expected to be key milestones.

Kenanga said higher volumes and utilisation could generate a more meaningful contribution to earnings in financial year 2027 and support margin expansion.

Other Malaysian semiconductor-related companies could benefit from the broader infrastructure cycle.

It noted rising orders, shipments, customer qualification and utilization would provide clearer evidence that the AI-led capital expenditure cycle is translating into sustainable revenue and earnings growth.

China demand and geopolitical risks

Beyond AI infrastructure, Huatai also highlighted macroeconomic and geopolitical risks.

China continues to face an imbalance between relatively strong supply and weak domestic demand, with industrial production remaining resilient while consumption, property investment and parts of manufacturing investment remain softer.

Huatai believes a durable recovery would require stronger household income, confidence and consumption rather than liquidity measures alone.

The investment house attributed weak domestic demand to factors including a relatively low household-income share, high precautionary savings, property-market weakness and subdued confidence.

Policy measures discussed at the forum included directing more fiscal resources towards social security, healthcare and education, increasing the contribution of state-owned enterprise profits to welfare funding and accelerating the urbanisation of migrant workers.

Meanwhile, China-US technology competition could increasingly extend beyond semiconductor chips into areas such as AI intellectual property, scientific research, talent, visas and cross-border collaboration.

Kenanga said this could raise risk premiums for companies heavily dependent on cross-border research and development, US-origin technology or sensitive overseas markets.

For AI investors, Huatai also suggested that efficiency should be assessed beyond the commonly cited cost per token.

As enterprise customers ultimately seek usable outcomes, cost per completed task could become a more relevant measure, particularly where a higher-priced model can deliver better accuracy, fewer retries, greater cache efficiency or stronger workflow integration.

Overall, Kenanga said AI remains a long-duration structural theme, but the investment cycle is broadening from the initial GPU and model-training phase.

The next stage is likely to be characterised by wider infrastructure requirements and greater emphasis on execution, with capacity expansion, customer qualification, utilisation and volume growth becoming increasingly important indicators of whether AI investment is translating into sustainable earnings growth.

For Malaysian technology companies, Kenanga said this favours businesses positioned at supply-chain bottlenecks and those able to demonstrate a clear progression from new orders and customer programmes to capacity ramp-up, higher utilisation and earnings contribution.

Kenanga Research foresees Malaysia’s logistics sector to ride on e-commerce boom