The price of dynamic random-access memory (DRAM) doubled in 2025 and rose a further 40 to 90 per cent in the first quarter of 2026, according to Taipei-based analysts TrendForce. Contract prices then jumped 58 to 63 per cent for DRAM and 70 to 75 per cent for NAND flash in the second quarter, the steepest quarterly increase in a decade. Apple raised Mac and iPad prices by hundreds of dollars, and Microsoft increased Xbox prices for the third time in 13 months, both citing a memory shortage driven by artificial-intelligence infrastructure.

The world’s three dominant memory producers—Samsung, SK Hynix and Micron—have shifted more than 80 per cent of advanced production lines toward high-bandwidth memory (HBM) and server-grade chips that feed AI accelerators. Data centres now absorb nearly half of global DRAM output, up from 32 per cent five years ago. Micron’s chief executive told investors in late June that demand continues to significantly exceed supply and that tight conditions will persist beyond 2027. Analysts in China and the United States expect a further 15 to 20 per cent of consumer-electronics memory capacity to migrate to data centres in 2027.

The same AI buildout is straining electricity grids. A Capgemini survey of more than 600 senior utility executives found that 77 per cent struggle to forecast demand and 68 per cent expect power shortages as data-centre demand outpaces supply expansion. In the United States, the pressure has triggered a record $203.6 billion in electric-sector mergers and acquisitions in the first five months of 2026. Grid interconnection queues worldwide hold more than 2,500 gigawatts of projects; in Texas, nearly 90 per cent of the 438 GW waiting is linked to data centres. The US Federal Energy Regulatory Commission issued show-cause orders in June requiring grid operators to justify how they handle large-load requests above 50 MW.

The infrastructure squeeze coincides with a transformation of labour markets. The World Economic Forum projects that 39 per cent of core job skills will change by 2030, with AI automating routine tasks but increasing demand for critical thinking and cross-functional collaboration. Swedish healthcare experts note that AI’s clinical usefulness remains limited not by algorithms but by data locked in incompatible systems, underscoring a broader pattern: the value of AI depends on organisational readiness as much as on technical capability. The next factual milestones are third-quarter memory contract prices, due in the coming weeks, and the timeline for new fabrication capacity. Most large-scale fabs will not add meaningful supply before mid-2027, and the first realistic transmission plan for Texas is not expected until autumn 2027.