Rethinking AI through a capital allocation lens

At a glance

• AI investment is accelerating rapidly, but investors are increasingly questioning the gap between expectations and real-world outcomes.
• Compute capacity, energy infrastructure, and organizational readiness could become critical constraints on AI adoption and monetization.
• The AI opportunity is global, but differences between US and China ecosystems may create distinct risks and return profiles.
• For investors, a disciplined capital allocation approach may prove more valuable than chasing market momentum.

AI is clearly moving through the hype cycle. Investors, however, are becoming increasingly wary of the gap between expectation and real-world outcomes.

Spending on AI is accelerating rapidly, with capex across leading technology firms expected to reach around USD 800 billion in 2026.¹ Broader market estimates point to between USD 4 trillion and USD 8 trillion of cumulative investment over the next five years.²

This acceleration in investment, alongside elevated equity valuations, rests on expectations of widespread productivity gains and scalable AI-driven revenues. Yet tangible evidence of such gains is likely to emerge only over a longer horizon, even if investments are happening today.

There are risks to the AI build-out, too. Physical constraints, including power availability, grid capacity, and emissions frameworks, are becoming potential bottlenecks to scaling.

While equity markets, for now, appear to be looking through these constraints, current valuations raise a more fundamental question: are capital markets pricing maturity into systems that are still evolving, effectively assigning value ahead of proven monetization?

These are the questions we posed to an expert panel at this year’s Global Markets Conference in Singapore.

A key theme from the discussion was that, while AI remains a transformative force, the most useful framework for investors may be to treat it as a capital allocation problem and focus on where value is demonstrably being created.

Where value is being created – and constrained

Recent market performance, with record valuations now apparent across the technology sector, assumes that productivity gains will be broad enough to justify the infrastructure build-out currently underway. That assumption warrants scrutiny.

Today, value is most apparent in relatively narrow domains such as coding, procurement, content generation, personalized marketing, and administrative tasks.

Expanding this across industries and use cases requires overcoming three key constraints:

1. Compute. A massive expansion of computing power is testing the limits of semiconductor manufacturers, hardware providers, and data center operators. Regulatory approvals may also influence the pace of infrastructure projects.

2. Energy. Data center electricity use surged 17% in 2025 and is projected to roughly double to 950 TWh by 2030.³ While this represents less than 5% of global generation, the always-on nature of data centers places a disproportionate strain on grid infrastructure. Investment in power generation, transmission, and distribution is not keeping pace.

3. Organizational readiness. A significant gap remains between what advanced AI models can theoretically achieve and what enterprise customers can use them for. One survey in 2025 found that while 88% of organizations now use AI, only a small minority report material enterprise-level earnings impact, with 6% meeting a threshold of at least a 5% contribution to earnings.⁴ Internal governance frameworks, compliance systems, and educational frameworks are simply not evolving fast enough to close the operational readiness gap.

These constraints increase the risk of monetization challenges and disappointing risk-adjusted returns down the line. More broadly, markets may not fully capture the sequence of conditions required for the AI economy to mature.

Value creation will therefore vary significantly depending on the pace of adoption across sectors, industries, and regions. This brings us to the question of capital allocation. Where should investors focus their attention?

Differences across ecosystems

AI is a global phenomenon, but the ecosystems driving it differ in important ways, as highlighted in our discussion in Singapore. Constraints around compute, energy, and readiness differ across regions, while competition between these ecosystems will continue to shape valuations and returns for investors.

In the US, leading AI labs are currently training the next generation of models on chips that are purpose-built for AI-native environments.

In China, companies are pursuing highly efficient, low-cost, open-source models, reflecting differences in access to capital, compute, and talent relative to their US peers.

Lower revenues, however, could place valuations under pressure if Chinese models fail to stay competitive over time.

Timeframes matter

The panel highlighted the characteristics shared by companies that have successfully adopted AI: strategic clarity about whether AI is being used to reduce costs, generate new revenue, or modernize operations; genuine leadership buy-in; and the right foundations in place, including data, talent, and technology.

In the near term, monetization is most likely when AI is used to develop customized models for specific industries or to build agents that solve problems within well-defined workflows.

This dynamic has already disrupted valuations in the SaaS sector and is forcing investors to rethink earnings expectations and cash flow models across multiple industries.

On a longer-term horizon, infrastructure and certain other industries may prove more resilient to disruption. Higher-value cognitive work, which requires human judgment and contextual expertise, remains beyond the limits of today’s AI tools. In these cases, the value of AI is likely to be concentrated in structured, verifiable tasks within narrow, well-defined workflows.

Anchoring allocations

AI has moved from a niche technology to a dominant theme in investor conversations globally. Its implications touch every asset class; this is no longer a story confined to growth equities and semiconductor stocks.

Investors need to price the various opportunities carefully and resist the pull of market momentum. Proven routes to monetization must be weighed against execution risk.

A capital allocation approach, as emphasized in the discussion, offers a useful anchor. By focusing on risk-adjusted returns and the specific conditions under which AI value is actually being realized, investors can construct portfolios that play the global AI story in a way that fits their mandate, recognizing that disciplined allocation will matter well before broad-based productivity gains fully materialize.

References

_{1. North American AI Data Center Expansion Drives 2026 Capex of Top Nine CSPs to US$830 Billion}

_{2. Tracking Trillions: The Assumptions Shaping the Scale of the AI Build-Out}

_{3. IEA: Data Centre Electricity Use Surged in 2025}

_{4. McKinsey: The State of AI}