AI, Power and the Emerging Constrainton Compute

Big Tech may be able to buy compute, but it cannot buy grid capacity on the same timeline

Large technology companies are set to spend more than $700 billion on artificial intelligence (AI)-related capital expenditures in 2026, fueling a rapid expansion of data center infrastructure. That level of spending can secure chips, servers and data-center capacity. It cannot as easily compress the timelines required to generate, transmit and deliver electricity. The recent announcement of the first futures market for AI computing power underscores this, with contracts tied to graphics processing unit rental prices. While the buildout of AI infrastructure is often framed as a compute problem, power, rather than chips, is increasingly becoming the gating

factor in how quickly AI adoption can scale.

The growing focus on power availability comes as electricity demand is accelerating across the US economy. Recent industry estimates suggest US data center IT load capacity could roughly double from approximately

80 gigawatts (GW) in 2025 to 150 GW in 2028, highlighting the pace at which AI infrastructure is scaling. Training and inference workloads are materially more energy-intensive than traditional computing tasks, while higher-performance chips are increasing power consumption per workload. As a result, the AI buildout is making digital infrastructure materially more power intensive.

The ability to deliver incremental power is not keeping pace

The traditional grid model is constrained by both physical infrastructure and regulatory processes. Transmission buildout can take years. Interconnection queues, which formalize access to the electric grid, are increasingly backlogged, and utilities must balance new sources of demand against existing capacity and reliability requirements. These processes typically introduce multi-year delays before new power can be delivered.

The core challenge is whether new generation capacity can come online fast enough to support rapid data center growth. Today, the evidence suggests it cannot. Projects completed in 2024 took roughly four to five years from initial interconnection request to commercial operation, highlighting the mismatch between AI-driven demand and grid-scale supply. For hyperscalers, power may represent a smaller cost than compute hardware, but it can be the more binding constraint as data centers can only go where reliable power can be secured.

The core challenge is whether new generation capacity can come online fast enough to support rapid data center growth. Today, the evidence suggests it cannot.

In the near term, natural gas is likely to play a central role in meeting new demand

These constraints are changing how power is sourced. Developers are increasingly prioritizing generation that can be deployed relatively quickly and provide reliable, continuous output. This helps explain the recent increase in natural gas capacity in interconnection queues, while solar, storage and wind declined year over year.

Natural gas is unlikely to be the final state of the power system, but it is well suited to meeting immediate needs. It is dispatchable, scalable and supported by existing infrastructure, making it one of the more practical near-term solutions for large loads that require high reliability. In some cases, this may include grid-connected generation. In others, operators may explore onsite or co-located generation to reduce reliance on constrained transmission infrastructure and accelerate deployment timelines.

This does not imply that localized or behind-the-meter generation will replace the grid. Large-scale operators value the reliability, cost efficiency and simplicity of centralized infrastructure. Rather, these solutions are likely to complement grid-based power in regions where electricity access has become a limiting factor.

Implications across the power value chain

Diversified power sourcing is creating a multi-layered investment opportunity across the energy and industrial ecosystem. Companies that enable additional power supply, such as GE Vernova, are seeing increased demand for generation equipment as well as improved pricing dynamics. Besides supplying gas turbines to meet the demand for more generation, service revenue tied to installed equipment can extend the profit cycle over many years.

Greater reliance on natural gas power generation also supports demand for pipeline and processing infrastructure. For example, Williams operates critical transmission systems that connect major supply basins to end markets, particularly along the Eastern Seaboard where data center development is concentrated. Importantly, much of the incremental electricity demand has already been secured through long-term capacity agreements with power producers. In many cases, expansion can be achieved through compression, pipeline extensions or other investments in existing systems, which may be faster and less complex than entirely new infrastructure.

At the same time, the solution set is broader than gas. Fuel cells and other modular systems may also serve as complementary solutions in constrained regions, particularly where grid access or permitting timelines limit the pace of deployment. While these technologies are unlikely to replace traditional generation at scale, they can help address localized power shortages.

As data centers become more power-intensive, the electrical systems inside the facility become more complex. We believe power management providers such as Eaton are positioned to benefit from rising demand for transformers, switchgear, backup systems and controls that help distribute electricity efficiently and reliably across higher-density computing environments. While this demand is being accelerated by near-term power bottlenecks, these systems should remain important long-term as data centers become larger, denser and more electricity intensive.

Over the longer term, additional sources of generation, including nuclear and other low-carbon technologies, may play a role in meeting sustained demand growth. However, these solutions are typically characterized by longer development timelines and greater regulatory complexity, meaning their impact is more likely to be gradual than immediate.

The AI buildout may be driven by digital innovation, but scaling it is increasingly a physical infrastructure challenge. For investors, that shifts the opportunity set beyond chips and cloud platforms toward generation, power delivery and localized energy solutions.

To download the PDF, click here.

---

DISCLAIMER This information has been prepared by Copia Investment Partners Limited (AFSL 229316, ABN 22 092 872 056) the issuer, distributor and responsible entity of the Artisan Global Discovery Fund. This document provides information to help investors and their advisers assess the merits of investing in financial products. We strongly advise investors and their advisers to read information memoranda and product disclosure statements carefully and seek advice from qualified professionals where necessary. The information on this document does not constitute personal advice and does not take into account your personal objectives, financial situation or needs. It is therefore important that if you are considering investing in any financial products and services referred to on this document, you determine whether the relevant investment is suitable for your objectives, financial situation or needs. You should also consider seeking independent advice, particularly on taxation, retirement planning and investment risk tolerance from a suitably qualified professional before making an investment decision. Neither Copia Investment Partners Limited, nor any of our associates, guarantee or underwrite the success of any investments, the achievement of investment objectives, the payment of particular rates of return on investments or the repayment of capital. Copia Investment Partners Limited publishes information on the document that is, to the best of its knowledge, current at the time and Copia is not liable for any direct or indirect losses attributable to omissions from the document, information being out of date, inaccurate, incomplete or deficient in any other way. Investors and their advisers should make their own enquiries before making investment decisions. © 2026 Copia Investment Partners Ltd.

The Artisan Global Discovery Fund invests all or substantially all of its assets in Artisan Global Discovery Fund (Fund), a sub-fund of Artisan Partners Global Funds plc. Artisan Partners Limited Partnership serves as investment manager to the Fund. Artisan Partnership Limited Partnership, its affiliates and Artisan Partners Global Funds plc (together, Artisan Partners) are not affiliated with Copia Investment Partners. Artisan Partners does not take any responsibility for the accuracy or completeness of the contents of these materials, any representations made herein, or the performance of the Artisan Global Discovery Fund offered by Copia Investment Partners. Artisan Partners disclaims any liability for any direct, indirect, consequential or other losses or damages, including loss of profits, incurred by you or by any third party that may arise from any reliance on these materials. Artisan Partners is not responsible for, nor involved in, the marketing, distribution or sales of shares or interests in the Artisan Global Discovery Fund and is not responsible for compliance with any marketing or promotion laws, rules or regulations; and no third party, other than Copia Investment Partners, is authorised to make any statement about any of Artisan Partners’ products or services in connection with any such marketing, distribution or sales. Past performance by any other funds or accounts advised by Artisan Partners, including the Fund into which the Artisan Global Discovery Fund invests, is not indicative of any future performance by the Artisan Global Discovery Fund. © 2026 Artisan Partners. All rights reserved.