AI’s next bottleneck is not chips. It is electricity, finance and time

The most important conversation about artificial intelligence may no longer be taking place inside a laboratory.

It is happening between lenders, power companies, grid operators, governments and the people who must build enormous physical assets quickly enough to keep the AI expansion moving.

A panel at AixEnergy in Bangkok put that reality in plain sight. Its title was “Financing the convergence: Capital allocation across AI and energy infrastructure”. The subject sounds technical. The underlying question is anything but: who pays for the electricity system behind the AI boom, and who carries the risk if expectations change?

The scale hidden behind the screen

AI feels weightless to the person typing a prompt. Its infrastructure is not. Data centres require land, high-voltage connections, cooling, backup systems and a supply of electricity that must be available every hour.

The International Energy Agency expects global data-centre electricity consumption to more than double to about 945 terawatt-hours by 2030, slightly more than Japan uses today. AI is the largest driver of that growth. Accelerated servers, mainly associated with AI, are expected to account for almost half of the net increase.

That forecast turns an apparent technology story into an energy-buildout story. Chips can improve rapidly. Transmission lines, substations and large generating assets move on planning, permitting and construction timetables measured in years.

The mismatch creates a valuable but uncomfortable market. Data-centre developers want power quickly. Utilities need confidence that the demand will remain long enough to repay new infrastructure. Lenders want contracted revenues, creditworthy customers and clarity over who bears the cost of a delayed grid connection.

Who was on the panel

The line-up reflected the different pieces that must come together before an AI infrastructure project becomes financeable:

• Suluksana Artamnuayvipas, Partner, Strategy and Transactions at EY-Parthenon, moderated the discussion.
• Sajal Kishore, Managing Director for Infrastructure and Project Finance Ratings at Fitch Ratings, brought the lender and credit-risk perspective.
• Supriy Singh, Senior Investment Specialist in Infrastructure Finance at the Asian Development Bank, represented the role development finance can play in mobilising private capital.
• Abishek Srivastava, Co-Founder and President of Teknobuilt, brought a project-delivery and construction technology perspective.
• Le Nguyen Thien Nga, Director at the Institute of Policy Administration and Development Strategy, added the policy and institutional dimension.

That combination matters. No single investor can solve the convergence alone. It requires commercial equity, infrastructure debt, utility investment, public policy and, in some markets, development-bank support or guarantees to make risks acceptable.

The hardest risk is that AI moves faster than infrastructure

A gas plant, renewable project, transmission link or data-centre campus is financed against years of expected cash flow. AI demand, by contrast, rests on technology and business models that can change much faster.

That raises questions financiers cannot avoid. Will today’s specialised AI facility remain useful if hardware becomes more efficient? Can the site be repurposed for cloud or other computing demand? Is the power contract strong enough to support project debt? Who pays if the data centre is ready before the grid?

Fitch has already highlighted location and adaptability as important protections. A facility able to serve broader cloud demand is less exposed than a remote project built around one speculative AI use. In other words, the most exciting project may not be the safest asset.

Asia already had an infrastructure gap

The competition for capital is especially sharp in developing Asia. The Asian Development Bank has estimated that the region needs $1.7 trillion of infrastructure investment each year through 2030 when climate costs are included. Power accounts for the largest share of that requirement.

AI does not replace those existing needs. It arrives on top of them. Data centres will compete with cities, factories, transport and household electrification for grid capacity and investment. A project can therefore be commercially attractive while still creating a public-policy problem if ordinary consumers are left carrying network costs.

This is where development finance can be catalytic. Technical assistance, guarantees, concessional capital and clear regulation can reduce risks that private investors cannot price comfortably. But public support should buy a public benefit: stronger grids, additional clean generation, transparent cost allocation and infrastructure that remains useful beyond one customer.

The real AI race

The popular AI race is described in models, chips and computing power. The physical race is different. It is about bankable electricity, credible counterparties, construction capacity and permission to connect.

Capital is available for convincing projects. What is scarce is the combination of power, certainty and time.

That is why this panel matters. It moves the AI debate away from demonstrations and into the machinery of delivery. The next breakthrough model may capture the headlines, but the decisive advantage could belong to the market that can finance a substation.

The Stratawire is reporting from Gastech 2026 and AixEnergy in Bangkok.

• Company websiteAIxEnergy
• LinkedInAIxEnergy on LinkedIn