Equity Insights

Path to monetisation in Al cycle

Current cycle may have similarities to a bubble, but this phase of Al development is being driven by physical bottlenecks rather than user adoption curves."

Global equity markets remain near record highs, yet performance is increasingly concentrated in a narrow segment of US mega-caps, which now account for roughly a quarter of global equity market capitalisation. Semiconductor manufacturers and cloud infrastructure providers have delivered repeated earnings upgrades, reinforcing the perception that AI is a durable multi-year investment cycle, rather than a short-lived thematic rally. In 2025, Al-related sectors alone accounted for approximately 60 per cent of S&P 500 earnings growth. Forecast revisions for leading chip companies continue to rise, with several large semiconductor firms now guiding to mid-teens to high-20s EPS growth for 2025–27. At the same time, segments further downstream in the value chain exhibit far weaker visibility and softer expectations.

Naturally, this divergence invites the familiar question of whether markets are witnessing a sustainable structural shift or the formation of a speculative bubble reminiscent of the early 2000s.

Figure 1: Technology and communication services sector weight share in global equities (%)

Click the image to enlarge

Source: HSBC AM, Refinitiv, Datastream. Data as of November 2025.

Is this cycle a bubble?

There are legitimate reasons for drawing parallels with past excesses. Valuations in parts of the AI ecosystem are elevated, capital intensity is rising rapidly, and markets are making long-dated assumptions about eventual monetisation that remain unproven. In particular, private markets show clear signs of stretched valuations, with venture-backed AI application companies commanding revenue multiples that far exceed public-market norms, often based on limited operating history and highly concentrated use cases. This mirrors earlier cycles, where in some cases, enthusiasm for transformative technology pulled forward years of expected value creation.

At the same time, today's cycle differs in that unprecedented levels of capital are being committed upfront to physical infrastructure. Hyperscalers, semiconductor manufacturers and network providers are deploying capital at scale, driven by rising compute usage, model complexity and time-on-task metrics. This does not eliminate speculative behaviour, but it does anchor the cycle in tangible investment rather than purely conceptual demand.

In other words, the foundations are physical, but valuations remain forward-looking. The cycle is neither purely speculative nor immune to speculative risk — it contains elements of both, and its ultimate outcome depends on whether downstream monetisation required to justify that deployment ultimately materialises.

Physical build-out and infrastructure-led momentum

It is clear that the current phase is characterised by intense demand for compute infrastructure rather than confirmed end-user monetisation. The infrastructure required to train and run foundation models is intensive in power, land, specialised cooling, high-performance silicon and grid connectivity. Hyperscalers have not waited for monetisation visibility before spending. They are expanding aggressively, securing transformer allocations, substation access and land zoning rights alongside long-term power agreements, reflecting confidence in future usage rather than realised application revenues.

Concurrently, semiconductor fabrication investment is rising at a pace rarely seen outside mobile and broadband infrastructure cycles, with equipment supply chains operating at capacity. The binding constraint today is not willingness to deploy capital, but the availability of power infrastructure, high-bandwidth memory, transformers, advanced packaging capacity, and the space to house all these.

Figure 2: Annual medium-term EPS growth expectations (%)

Click the image to enlarge

Source: Refinitiv, IBES, Datastream, HSBC Asset Management, November 2025.

Importantly, recent earnings upgrades for semiconductors reflect these physical constraints. Several major chipmakers have reported order backlogs extending well into 2026, with high-bandwidth memory demand exceeding prior-cycle peaks and contributing meaningfully to forward EPS revisions. This does not imply certainty of end-user monetisation, but it does indicate utilisation-led scarcity at the infrastructure layer.

Compute cost inflation further illustrates this dynamic. A single rack of compute that cost roughly USD20,000 a decade ago now exceeds USD250,000, driven both by demand intensity and component scarcity.¹ Nvidia’s margins reflect this duality. Its pricing power stems from constrained fabrication and packaging capacity rather than pure sentiment.

Figure 3: AI datacenter economics¹

Click the image to enlarge

Source: HSBC AM, November 2025.

Similarly, AI build-outs across the US, Europe and Asia are reshaping power markets, accelerating transmission upgrades and aligning multi-year capital plans for utilities and industrial producers. While infrastructure being built today supports near-term earnings visibility upstream, it does not eliminate risk should downstream adoption fall short.

The first beneficiaries of AI have therefore been hardware-centric names, consistent with prior compute cycles in which silicon and infrastructure monetise before software. However, infrastructure revenues remain robust only so long as deployment plans hold. Should downstream monetisation disappoint materially, hyperscalers may defer, resize or redesign build-outs, introducing cancellation risk even at the hardware level.

1 - Dylan Patel and GeraldWong, ‘AI Server Cost Analysis – Memory is the biggest loser’,May 2023.

The invisible variable in equity pricing

Today, equity valuations embed significant confidence that downstream profit pools will ultimately materialise. Yet enterprise AI adoption remains uneven. Budget cycles are slow, many organisations remain in proof-of-concept stages, and large scale workflow integration is still developing. Productivity capture, licensing structures and usage based pricing models are evolving, with timelines varying widely by sector. Given application-layer revenues remain largely conceptual, markets are extrapolating long term outcomes rather than pricing established earnings streams.

Figure 4: Price to cash earnings per share (x)

Click the image to enlarge

Source: HSBC AM, Refinitiv, Datastream. Data as of November.

This makes valuation risk worth examining carefully. Price to cash earnings per share for US mega-caps are approaching levels last seen during the late-1990s cycle. While earnings quality remains strong, reported figures benefit from capitalisation choices, depreciation assumptions and non-recurring add-backs. These practices are common in fast-growing sectors, but they increase sensitivity to any delay in monetisation. Private markets appear even more stretched, with AI-focused venture investments commanding aggressive revenue multiples despite limited operating histories, amplifying the risk of valuation correction if adoption disappoints.

End user demand therefore remains a critical uncertainty. Enterprises may not scale AI usage at the pace implied by current investment plans, particularly if IT budgets tighten, or productivity gains prove slower to realise. If adoption curves flatten or elongate, revenue realisation may lag capex, and margin expectations embedded in equity pricing, especially for software and consumer-facing platforms, could reset lower.

Hence, the current phase can be characterised as infrastructure-led, with monetisation expected to follow rather than already visible. The uncertainty lies less in whether applications emerge, and more in the timing and breadth of adoption required to justify today’s capital intensity. This asymmetry explains why earnings risk is skewed toward downstream software and platform businesses, while upstream suppliers retain greater near-term visibility — albeit still exposed if adoption assumptions are ultimately revised.

Global multipliers

Even if broader monetisation timelines remain difficult to forecast, rather than dismissing the cycle, it is important to recognise that the first phase of returns has accrued where supply cannot respond quickly.

First, opportunity lies in power infrastructure. Data-centre demand is now consuming grid allocations faster than residential construction in several US states. Transmission and transformer shortages extend build-out timelines by years, creating suppliers’ pricing power. This has positioned Korean and European high-voltage manufacturers, alongside select US grid equipment providers, to capture structurally constrained demand.

Another overlooked area is advanced packaging and server design. Taiwan dominates advanced packaging, while three firms alone, concentrated in Taiwan and China, control more than 90 per cent of global server manufacturing. These companies are not beneficiaries of AI hype — they are indispensable enablers of the AI value chain.

Commodities underpinning AI infrastructure expand the investment universe further. The compute and power intensity of AI increases demand for copper, silicon, rare earths, and platinum, much of which is extracted or refined in Chile, South Africa, Indonesia and China. Their importance rises as compute and energy intensity scales.

Broader emerging markets exemplify why looking beyond US mega-caps for the next wave can be beneficial. China, for example, is frequently underestimated in its AI potential, largely due to the perception that its capabilities are constrained by a lag in cutting-edge chip development and capex spending. However, China’s competitive playbook has never relied on technological first-mover advantage.

Figure 5: Regional capital expenditure in technology and communications sector (USD bn)

Click the image to enlarge

Source: HSBC AM, Refinitiv, Datastream. Data as of November 2025.

Instead, it competes through scale, cost disruption, state-backed industrial learning, and deep integration of AI into real-economy applications. This is critical in AI value capture which does not depend exclusively on building the most advanced model but relies on deploying models at scale across transaction-heavy ecosystems where monetisation is repeatable.

China operates closed loop data ecosystems tied directly to transactions in which payments, logistics, identity, retail, entertainment, and enterprise services sit within unified platforms. Alibaba, Meituan, Tencent and ByteDance collectively control an end-to-end chain of consumer interaction across transactions, data, and service delivery. In contrast, US platforms manage data fragmented across advertising, productivity tools, and software services. This allows China to monetise per user at a level that exceeds Western platforms even with smaller models.

Cost structure reinforces this advantage. China’s large language models are already priced at a fraction of their US and European peers, while offering increasingly competitive performance. For example, Baidu’s autonomous operations undercut Western competitors by as much as 80 per cent in unit economics, accelerating commercialisation timelines.

This mirrors past disruptions in solar, batteries, and EVs, where China closed capability gaps and reshaped global pricing. If AI follows a similar path, with adoption and cost outpacing innovation, China could define the economics of deployment even while trailing the US in frontier chips.

Chip restrictions are a constraint, but not a deadlock. China’s policy mandating local chips for AI infrastructure supports domestic semiconductor growth, backed by its expanding cloud market, which could reach USD1 trillion. This scale can foster a self-sufficient ecosystem to bridge cost and capability gaps.

Similarly, other markets are also leapfrog adopters rather than laggards. India and Gulf economies, for instance, are adopting AI without legacy system constraints and are deploying generative models directly into public administration, logistics, credit underwriting and education. This underscores that these markets are capable of absorbing AI faster once infrastructure is accessible at appropriate price points. The investment opportunity is therefore not a single geography, but a distributed ecosystem spanning US architecture, Asian manufacturing depth and EM resource supply.

Where to position now versus later

Given today’s phase, monetisation is happening, but unevenly. For now, near-term earnings visibility remains reliable upstream, supported by scarcity and committed investment. However, the realisation of the full value of current investments ultimately remains contingent on downstream monetisation. Should it fail to materialise at scale, order deferrals, redesigns and cancellations remain possible, particularly if compute requirements evolve or efficiency gains reduce hardware intensity.

Downstream opportunities are larger but more uncertain. Software monetisation is not absent from the cycle, it is delayed. Growth curves may steepen once infrastructure constraints ease, but the dispersion of outcomes remains wide, and winners are not yet clear. In fact, not all participants are likely to survive.

Hence, portfolio construction may benefit from barbell positioning by owning what prints cash now while keeping a stake in what could monetise later. This could be achieved by balancing infrastructure certainty with software possibility, and America’s leadership with Asia’s scale and emerging markets’ resource leverage.

Source HSBC AM, as of January 2026 For informational purposes only and should not be construed as a recommendation to invest in the specific country, product, strategy, sector or security The views expressed above were held at the time of preparation and are subject to change without notice Any forecast, projection or target where provided is indicative only and not guaranteed in any way HSBC Asset Management accepts no liability for any failure to meet such forecast, projection or target