AI infrastructure: Strong demand does not remove execution risks

Video: Top of Mind in APAC | What Fed hikes mean for Asian assets (6 mins)
Video: Mark Haefele on why the equity rally can continue (3 mins)
Video: Market Playbook | Why we still like Japanese equities (5 mins)
Podcast: Signal over Noise: Takeaways from Silicon Valley, on Apple and Spotify (6 mins)

Already contending with the surge in bond yields, higher energy costs, and the US-Iran conflict, AI investors this week faced fresh concerns over data center financing. Oracle's shares fell around 3.5% on Thursday after reports that its Project Jupiter data center campus could face a one-year delay. The Financial Times and Bloomberg report that Oracle is seeking to invoke a contractual clause that would allow it to defer certain payments, in the event power-related delays prevent the campus from opening on schedule.

The 1,400-acre campus in New Mexico is being developed with partners to support Oracle’s computing commitments to OpenAI. The project has secured USD 18 billion in loans, according to media reports cited by Reuters. Neither Oracle nor its partners have indicated an intention to terminate the development.

Without taking a view on any individual companies, we recognize that financing and execution risks could extend beyond a single development. That said, we would caution against extrapolating project-level difficulties to the wider industry at this stage:

AI demand and delivery are different questions. We forecast AI industry capital spending of USD 1.2 trillion in 2027, up roughly 33% from our USD 900 billion estimate for 2026. This forecast already assumes that some announced projects will be delayed, resized, or canceled. The UBS Evidence Lab identifies nine North American projects affected by delays or cancellations, representing roughly 7 gigawatts (GW) of initial capacity, or equivalent to around 5% of the region’s 130GW announced pipeline. We view risks to our forecast as manageable for now, but it is important to note that resilient demand does not guarantee timely delivery.

Power constraints are both an opportunity and an execution risk. Access to electricity increasingly determines when computing capacity can begin generating revenue, while limited availability of suitable land, local opposition, and environmental concerns can further complicate development. Long grid-connection queues are increasing interest in power generated at or near data centers, although these solutions cannot replace wider investment in electricity networks. With the IEA estimating global grid investment could exceed USD 540bn in 2026, we see opportunities across generation, transmission, storage, electrical equipment, cooling, and industrial automation. We must also caution that potentially slower deployment would risk pushing out revenue recognition, increase financing costs, and affect supplier order visibility.

Private-market investors should prioritize brownfield projects with visible cash flows. Across real estate and infrastructure, we favor core and core-plus exposure, which can offer greater income visibility and limit development risk, without eliminating financing or tenant risk. Within data centers, we maintain our longstanding preference for existing assets over greenfield developments (new facilities that must complete construction, secure power, and attract tenants before generating income), particularly in primary markets with low vacancy rates and high-quality tenants under long-term leases. Investors should also assess power availability, financing structures, and the capacity of each project to absorb delays. Geographic diversification remains important because development timelines, policy risks, and access to land and power vary across regions.

So, stepping back, this reinforces the need for investors to fully understand the quality of the AI assets in which they invest. Within technology, we favor quality semiconductor and hardware companies alongside large technology platforms and defensive technology businesses that may prove more resilient in a risk scenario where AI spending moderates. In private markets, we favor established infrastructure with visible income over projects whose returns depend heavily on construction and permitting. We also continue to recommend diversified exposure to our longer-term Transformational Innovation Opportunities of AI and Power and resources, while considering complementary Longevity exposure to diversify portfolio growth drivers rather than rely solely on the AI investment cycle.

When investing in alternative investments including private markets, investors must always be aware of the risks inherent to the asset class, such as a lack of liquidity, lack of control, limited disclosure, blind pool risk, uncertain cash flows, and the use of leverage.