Smartphones are becoming harder to afford as global shipments are projected to fall 13% in 2026, with declines exceeding 20% in Africa and the Middle East. The downturn is concentrated at the cheapest end of the market, where many people are priced out of ownership. The main driver is memory supply, since smartphones rely on DRAM and DRAM production is difficult and expensive. DRAM supply is inelastic, and recent years have shifted large volumes of memory toward AI data centers. Samsung, SK Hynix, and Micron dominate production, and building advanced fabrication capacity requires $15 to $20 billion and years to reach competitive yields. Memory makers prioritize capital discipline and avoid oversupplying demand. AI increases memory intensity through HBM, which stacks DRAM dies vertically and uses far more wafer capacity per gigabyte than standard DRAM.
"In 1985, the best computer a reasonably affluent American could buy was the IBM PC AT, which cost $19,400 in today's money. Today, a Tecno Spark Go costs $30 in a Nairobi market stall and runs a processor billions of times faster. No other good in history has experienced a cost decline on that scale. That era is now ending. The International Data Corporation predicts worldwide smartphone shipments will fall 13% in 2026, the largest single-year decline ever. In Africa and the Middle East, the drop exceeds 20%."
"The crash is concentrated at the cheapest end of the market. IDC calls it " a structural reset of the entire market. " A huge share of the world's population is getting priced out of smartphone ownership. The reason is memory. Smartphones, like all computers, use DRAM. The global supply of DRAM is remarkably inelastic because memory is extraordinarily difficult and expensive to produce."
"For decades, most DRAM went to smartphones and laptops. In the last three years, AI emerged as an enormous and hugely profitable consumer of memory. The result is a massive reallocation of DRAM production away from consumer electronics and toward AI data centres. Three companies, Samsung, SK Hynix, and Micron, produce more than 90% of the world's DRAM. Building a single state-of-the-art memory fabrication facility costs $15 to $20 billion. It takes years of producing defective chips before yields become competitive."
"These memory makers learned a brutal lesson from decades of boom-and-bust cycles: always leave demand unmet. Intel, Texas Instruments, IBM, Germany's Qimonda, and Japan's Elpida all exited or collapsed. The survivors' strategy is capital discipline above all else. AI changed the calculus. Training and running AI models requires high-bandwidth memory, or HBM, which stacks multiple DRAM dies vertically and connects them with thousands of tiny channels. A single gigabyte of HBM consumes more than three times the wafer capacity of a gigabyte of standard D"
Read at TNW | Artificial-Intelligence
Unable to calculate read time
Collection
[
|
...
]