Over the past few weeks, several US banks have pulled off from lending to Oracle for expanding its AI data centres, as per a report.
Over the past few weeks, several US banks have pulled off from lending to Oracle for expanding its AI data centres, as per a report.
That’s not true, from what I’ve read:
www.trendforce.com/…/20251113-12780.html
despite higher ASPs boosting profitability across the memory industry, capital spending on DRAM and NAND Flash is only anticipated to increase modestly in 2026. This limited investment growth is unlikely to significantly affect bit output.
Memory makers seem skeptical, hence aren’t planning to spend on more capacity in 2026.

TrendForce’s latest investigations reveal that despite higher ASPs boosting profitability across the memory industry, capital spending on DRAM and NAND Flash is only anticipated to increase modestly in 2026. This limited investment growth is unlikely to significantly affect bit output. Instead, the emphasis is shifting from capacity expansion to advancements such as process technology upgrades, higher-layer stacking, hybrid bonding, and high-value products such as HBM.
That’s not true, unfortunately. It’s not economical to transplant RAM ICs once they’re packaged and soldered onto something.
And if they’re produced as, say, HBM modules, they absolutely cannot be repurposed for, say, DDR5 or LPDDR5 CPUs, or GDDR GPUs. There’s no reworking, the memory buses on processors simply do not support them electrically, and altering those processors would have a massive development cost with years of lead time.
Some of the RAM (like the LPDDR5X for the Nvidia Grace Hopper ARM CPUs) can be re-used, but it seems most is being made as HBM.