The Next Economy Research Centre

“Capital builds the infrastructure; productivity determines the enduring wealth.”

The most consequential development in the AI Supercycle is no longer simply the extraordinary amount of capital being invested. It is the widening financial distinction between companies already generating exceptional profits from AI infrastructure and those still committing extraordinary capital to build and operate it.

The evidence has strengthened sharply in the past 24 hours: SK Hynix reporting record memory profits in South Korea, ASM International pulling a 2030 target three years forward in the Netherlands, Meta pricing enlarged offerings in the United States, and a specialized developer seeking $5 billion for a Polish data centre.

The suppliers of intelligence infrastructure are already monetizing the Supercycle. The companies financing its deployment must still demonstrate that the resulting computing capacity will consistently generate returns exceeding its full cost of capital.

SK Hynix reported Q3 revenue of ₩104.62 trillion (+328% YoY), operating profit of ₩83.33 trillion at roughly 80% margin, net income of ₩39.78 trillion and net cash of ₩98.7 trillion. Long-term customer agreements already cover more than 50% of planned production capacity for 2027–2028.

The company also announced approximately ₩100 trillion in 2026 shareholder returns — a figure that includes gains from its Kioxia investment, not recurring operating earnings alone — while discontinuing a possible Solidigm IPO.

The signal is financial, not merely a demand headline: the semiconductor investment cycle may increasingly finance itself through operating profits, diverging from infrastructure operators that require repeated equity or debt issuance. But an 80% margin reflects cyclical scarcity and should not be extrapolated indefinitely.

Meta priced $25 billion in Class A common equity — structured through forward-sale agreements settling by November 2027 — and $12.5 billion in mandatory convertible preferred securities carrying a 6.50% dividend. The combined $37.5 billion, plus a described but unestablished $60 billion at-the-market program, represents priced size and potential issuance, not completed cash proceeds. The 2026 capital-expenditure outlook remains approximately $135–145 billion, against substantial operating cash flow.

Separately, Argentum AI is negotiating roughly $5 billion to finance computing equipment for a Poland project tied to DL Invest Group — with some capacity serving customers lacking investment-grade credit, and a reported but unverified $12 billion-plus in contracted revenue. Contracted revenue is not collected cash, and a long-term contract does not eliminate default, utilization or refinancing risk.

ASM International reported Q3 revenue of €1.159 billion (+45% YoY), a 52.0% gross margin and a 2027 outlook above €5.7 billion — reaching its previously set 2030 objective three years early, with a new €300 million buyback authorization. Dassault Systèmes raised its full-year revenue-growth target to 4–5% after Q3 revenue of €1.55 billion (+5%).

Yet the industrial conversion problem remains unresolved. European AI adoption in strategic sectors rose to 18% in 2025 — but only 16% in manufacturing — and no economy-wide productivity inflection has been independently verified. Europe does not need to control every layer to capture returns; the distinction between deploying AI and extracting value from it remains fundamental.

US September durable-goods data confirmed strong physical capital formation: core capital-goods orders rose 0.5% and shipments 1.2% month on month (10.9% and 8.8% year on year). The AI boom is no longer only a market narrative — though strong capex does not guarantee future returns.

China's industrial economy is becoming two-speed: September industrial profits rose 14.4%, with computer, communications and electronic-equipment profits more than doubling over nine months, while furniture makers declined. CXMT reportedly intends to proceed with HBM3E despite estimated usable yields of just 20–30%, versus roughly 70% at SK Hynix — exposing the near-term cost of technological sovereignty.

The physical electricity constraint has not disappeared: no new verified European grid milestone offsets the growing pressure to deliver computing infrastructure. The Federal Reserve concludes today and the ECB decides tomorrow — both material to the cost of financing long-duration AI assets, neither incorporated here as a completed outcome.

Five key takeaways

For investors, boards and policymakers

• The AI Supercycle is generating extraordinary profits — but not uniformly: SK Hynix and ASM show strong supplier economics, while infrastructure financing stays capital intensive.
• The quality and timing of financing matter increasingly, as Meta's priced securities and Argentum's negotiations reveal different structures and credit exposures.
• The investment cycle is already affecting the real economy, with US equipment data showing substantial capital formation — though future returns remain uncertain.
• Europe's comparative strengths extend beyond computing infrastructure, into equipment, industrial software, engineering and productive AI applications.
• Sustainable capital productivity — the relationship among operating cash flow, capital commitments and incremental returns — is becoming the decisive financial measure, not investment scale alone.

Capital builds the infrastructure. Technology creates the capability. Scarcity determines the initial profits. Productivity determines the enduring wealth.