Declining Compute Costs and the Growing Energy Constraint
The economic landscape of artificial intelligence is shifting as the cost of compute continues to decline at a rate faster than previous transformative technologies, according to data from Epoch AI. While the cost of training individual models remains high, industry observers on platforms like Reddit suggest that the return on investment (ROI) per model remains significant, as the increased capability of newer models often translates into higher profit margins despite the substantial capital expenditure required for their development.
Infrastructure and the Energy Competition
As investment flows into AI infrastructure, semiconductors, and cloud computing, a critical bottleneck is emerging: electricity. The massive power requirements of data centers are creating direct competition with heavy industries for grid access and energy supply.
In Malaysia, the scale of this demand is becoming evident. Google has announced plans to invest $15 billion in the country through 2027 and 2028, focusing on data center campuses and clean energy initiatives. This follows reports of Chinese interest in developing an aluminum plant in Sarawak alongside a green-energy facility. Such developments highlight a growing trend where large-scale industrial projects must secure workable electricity supplies long before they can generate revenue.
The competition for power is creating a divergence in energy pricing. While the U.S. Aluminum Association notes that a globally competitive smelter requires electricity at approximately $40 per megawatt-hour under long-term contracts, technology companies have reportedly committed to prices exceeding $115 per megawatt-hour to power AI facilities.
Asset Revaluation and Industrial Shifts
The tension between computing needs and traditional manufacturing is already influencing asset valuations. In February 2026, Century Aluminum curtailed operations at its Hawesville, Kentucky, smelter following rising electricity costs. The site was subsequently sold to an affiliate of TERAWULF INC for $200 million in cash and a 6.8% equity stake, driven by the site’s high-voltage transmission lines and energized substation, which are highly valuable for computing campuses.
Similarly, ALCOA CORP disclosed in its third-quarter 2026 results that it had entered definitive agreements to sell portions of its Massena East site for data center development, with potential cash proceeds reaching up to $315 million. These transactions suggest that electrical infrastructure and grid access are becoming primary drivers of industrial real estate value, sometimes independent of the original manufacturing output.
Supply Chain and Regulatory Risks
The expansion of AI-driven demand faces logistical and political headwinds. Procurement of large power transformers can take up to four years, potentially delaying the commissioning of both data centers and industrial plants.
Furthermore, regulatory uncertainty remains a factor for developers. In the United States, discussions regarding environmental permits for hyperscale facilities and potential federal moratoriums have introduced costs associated with uncertain permit timelines. Consequently, developers are increasingly looking toward overseas locations, such as Malaysia or Indonesia, where the timeline for connecting to the power grid may be more predictable.
