The Next Economy Research Centre

“Capital can finance the promise of intelligence. Electricity makes its delivery possible.”

The AI Supercycle is entering a phase in which electricity availability, financial resilience and infrastructure execution increasingly determine whether technological ambition becomes economic reality.

The decisive development: on October 18, Reuters confirmed that Microsoft had terminated its agreement with Nscale for computing capacity at Loughton, Essex — a project designed for roughly 23,000 Nvidia GPUs and 50 MW of infrastructure, whose grid connection has reportedly slipped to the 2030s. For the first time, the link between inadequate electricity infrastructure and the loss of a major AI customer is visible in a named, commercially significant project.

The question has changed. It is no longer whether the AI Supercycle can attract capital, but whether that capital can be converted into operational infrastructure before financing costs, contractual delays and depreciation undermine returns.

Nscale's October 2025 partnership envisaged some 200,000 GB300 GPUs across multiple sites; Loughton was to deliver ~23,000 by Q1 2027. Nscale maintains its commitment and says the broader Microsoft relationship is intact; Microsoft has not confirmed that power delays caused the termination. The causal detail remains unconfirmed — but the withdrawal of an anchor customer is a materially stronger signal than a postponed build schedule.

The mechanism matters: grid delay → commissioning uncertainty → customer-contract risk → lower revenue visibility → higher financing costs. A signed agreement is not necessarily irrevocable, and future capacity is not productive until operational and utilized. A new distinction could emerge between operators with deliverable electricity capacity and those holding pipelines dependent on future connections.

Britain's Ofgem reported connection applications rising from ~41 GW to 125 GW in under a year, with data centres accounting for at least 80 GW and ~73 GW in the queue — requests, not validated demand. Electricity availability now determines not only where infrastructure is built but whether it attracts bankable financing: reliable electricity → faster commissioning → earlier revenue → stronger cash flow → lower financing risk.

Europe should separate three problems often conflated: generation; transmission and distribution; and the cost and reliability of delivered electricity at the project location. Faster grid connections could unlock private investment more effectively than further subsidy. In markets, Brent retreated ~1.8% to around $107.83 on October 19 amid possible Russian supply developments — not evidence the energy shock has ended.

TSMC posted Q3 revenue of $46.72B, net income up 68.7%, gross margin 66.4%; ASML delivered €11.66B revenue at a 57.5% margin. China's CXMT expects nine-month revenue of CNY 265–270B (vs ~32.1B a year earlier) and preliminary net profit of CNY 131–132.5B (vs a ~5.3B loss) — company estimates, not audited.

The asymmetry is the point: higher semiconductor profits can coexist with lower expected returns on downstream infrastructure. Manufacturers benefit from memory prices and equipment demand; infrastructure customers face rising hardware costs and tighter financing. The financial winners need not be those announcing the largest investments.

Today, Siemens and Nvidia present industrial AI at the Erlangen electronics factory — a forthcoming demonstration, not yet verified productivity. The European Commission's JRC reports AI adoption in strategic sectors rose from 7% (2021) to 18% (2025), ~20% across business — but much reflects general-purpose tools, not deep integration. Using AI is not operating an AI-native enterprise.

China's Q3 GDP grew 4.5% (above the 4.4% consensus), with high-tech manufacturing +14.3% and industrial robot production +29.1% year-to-date — against private investment −10.4% and property development −20.6%. This is the Great Split within an economy: advanced production outpacing domestic demand, forcing reliance on external markets. China has also proposed export restrictions on 12 solar-manufacturing technologies (consultation to November 15) — competitive equipment for Europe, but deepening supply-chain dependence.

The European Council's October 16 conclusions reaffirmed the 28th company-law regime, the Industrial Accelerator Act, the grids package and the Savings and Investment Union — but reached no final agreement on the 2028–2034 budget. The competitiveness problem is now one of coordination: financing, grids, permitting and industrial strategy operate on different timescales. Europe must improve the conversion rate between announced investment, financeable infrastructure, deliverable electricity, operational compute and measurable productivity.

Five Key Takeaways

For investors, boards and policymakers

• Electricity availability is becoming a financial variable — build verified grid access and commissioning schedules into valuations.
• Contracted AI demand is not bankable revenue — evaluate termination rights, counterparty concentration and commissioning probability.
• Semiconductor profitability does not guarantee infrastructure profitability — strong chip earnings can coexist with deteriorating operator economics.
• Europe must accelerate infrastructure-to-productivity conversion — coordinate AI investment with energy, financing and measurable outcomes.
• The next rotation may reward endurance and execution — operational infrastructure and sustainable cash flow over distant capacity promises.

Capital can finance the promise of intelligence. Electricity makes its delivery possible. Only productivity converts it into wealth.